fix(main): re-evaluate chain per call to respect caller model override

buildOrch now returns a closure instead of *Orchestrator. Each invocation
calls models.ChainFor(skill, req.Model) so a non-empty caller override
collapses to a single-entry chain (no escalation). The attempts slice is
also allocated fresh per call, preventing unbounded growth across requests.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

.gitea/workflows/ci.yml

@@ -53,6 +53,6 @@ jobs:
           chmod 600 ~/.ssh/id_rsa_gh_mirror
           ssh-keyscan github.com >> ~/.ssh/known_hosts 2>/dev/null
           GIT_SSH_COMMAND="ssh -i ~/.ssh/id_rsa_gh_mirror -o IdentitiesOnly=yes" \
-            git push git@github.com:mathiasb/hyperguild.git HEAD:main --tags
+            git push git@github.com:mathiasb/hyperguild.git HEAD:main --follow-tags
           rm ~/.ssh/id_rsa_gh_mirror
           echo "✓ Mirrored to GitHub"

cmd/supervisor/main.go

@@ -13,6 +13,10 @@ import (
 	"github.com/mathiasbq/supervisor/internal/skills/brain"
 	"github.com/mathiasbq/supervisor/internal/skills/org"
 	"github.com/mathiasbq/supervisor/internal/skills/retrospective"
+	skilldebug "github.com/mathiasbq/supervisor/internal/skills/debug"
+	"github.com/mathiasbq/supervisor/internal/skills/review"
+	"github.com/mathiasbq/supervisor/internal/skills/spec"
+	"github.com/mathiasbq/supervisor/internal/skills/trainer"
 	"github.com/mathiasbq/supervisor/internal/skills/sessionlog"
 	"github.com/mathiasbq/supervisor/internal/skills/tdd"
 	"github.com/mathiasbq/supervisor/internal/tier"
@@ -51,11 +55,55 @@ func main() {
 		os.Exit(1)
 	}
 
-	executor := iexec.New(iexec.Config{
+	reviewPrompt, err := os.ReadFile(cfg.ConfigDir + "/review.md")
+	if err != nil {
+		logger.Error("read review.md", "path", cfg.ConfigDir+"/review.md", "err", err)
+		os.Exit(1)
+	}
+
+	debugPrompt, err := os.ReadFile(cfg.ConfigDir + "/debug.md")
+	if err != nil {
+		logger.Error("read debug.md", "path", cfg.ConfigDir+"/debug.md", "err", err)
+		os.Exit(1)
+	}
+
+	specPrompt, err := os.ReadFile(cfg.ConfigDir + "/spec.md")
+	if err != nil {
+		logger.Error("read spec.md", "path", cfg.ConfigDir+"/spec.md", "err", err)
+		os.Exit(1)
+	}
+
+	trainerReaderPrompt, err := os.ReadFile(cfg.ConfigDir + "/trainer-reader.md")
+	if err != nil {
+		logger.Error("read trainer-reader.md", "path", cfg.ConfigDir+"/trainer-reader.md", "err", err)
+		os.Exit(1)
+	}
+	trainerWriterPrompt, err := os.ReadFile(cfg.ConfigDir + "/trainer-writer.md")
+	if err != nil {
+		logger.Error("read trainer-writer.md", "path", cfg.ConfigDir+"/trainer-writer.md", "err", err)
+		os.Exit(1)
+	}
+
+	claudeExec := iexec.New(iexec.Config{
 		SystemPrompt:   string(systemPrompt),
 		LiteLLMBaseURL: cfg.LiteLLMBaseURL,
 		LiteLLMAPIKey:  cfg.LiteLLMAPIKey,
 	})
+	litellmExec := iexec.NewLiteLLM(cfg.LiteLLMBaseURL, cfg.LiteLLMAPIKey, 0)
+	verifier := iexec.NewVerifier("", models.Verifier(), 0)
+
+	buildOrch := func(skill string) func(ctx context.Context, req iexec.Request) (iexec.Result, error) {
+		return func(ctx context.Context, req iexec.Request) (iexec.Result, error) {
+			rawChain := models.ChainFor(skill, req.Model)
+			chain := make([]iexec.ChainEntry, len(rawChain))
+			for i, m := range rawChain {
+				chain[i] = iexec.EntryFor(m)
+			}
+			attempts := make([]iexec.AttemptRecord, 0, len(chain))
+			orch := iexec.NewOrchestrator(chain, litellmExec.Run, claudeExec.Run, verifier, models.LlamaSwapURL(), &attempts)
+			return orch.Run(ctx, req)
+		}
+	}
 
 	tierFn := func(ctx context.Context) tier.Info {
 		return tier.Detect(ctx, "https://api.anthropic.com", cfg.LiteLLMBaseURL)
@@ -65,8 +113,9 @@ func main() {
 	reg.Register(tdd.New(tdd.Config{
 		SystemPrompt: string(systemPrompt),
 		SkillPrompt:  string(tddPrompt),
-		DefaultModel: models.Resolve("tdd", ""),
-		ExecutorFn:   executor.Run,
+		DefaultModel: models.ChainFor("tdd", "")[0],
+		ExecutorFn:   buildOrch("tdd"),
+		SessionsDir:  cfg.SessionsDir,
 	}))
 	reg.Register(brain.New(brain.Config{
 		IngestBaseURL: cfg.IngestBaseURL,
@@ -79,9 +128,35 @@ func main() {
 	}))
 	reg.Register(retrospective.New(retrospective.Config{
 		SkillPrompt:  string(retroPrompt),
-		DefaultModel: models.Resolve("retrospective", ""),
+		DefaultModel: models.ChainFor("retrospective", "")[0],
 		SessionsDir:  cfg.SessionsDir,
-		ExecutorFn:   executor.Run,
+		ExecutorFn:   buildOrch("retrospective"),
+	}))
+	reg.Register(review.New(review.Config{
+		SkillPrompt:  string(reviewPrompt),
+		DefaultModel: models.ChainFor("review", "")[0],
+		ExecutorFn:   buildOrch("review"),
+		SessionsDir:  cfg.SessionsDir,
+	}))
+	reg.Register(skilldebug.New(skilldebug.Config{
+		SkillPrompt:  string(debugPrompt),
+		DefaultModel: models.ChainFor("debug", "")[0],
+		ExecutorFn:   buildOrch("debug"),
+		SessionsDir:  cfg.SessionsDir,
+	}))
+	reg.Register(spec.New(spec.Config{
+		SkillPrompt:  string(specPrompt),
+		DefaultModel: models.ChainFor("spec", "")[0],
+		ExecutorFn:   buildOrch("spec"),
+		SessionsDir:  cfg.SessionsDir,
+	}))
+	reg.Register(trainer.New(trainer.Config{
+		ReaderPrompt: string(trainerReaderPrompt),
+		WriterPrompt: string(trainerWriterPrompt),
+		DefaultModel: models.ChainFor("trainer", "")[0],
+		ExecutorFn:   buildOrch("trainer"),
+		SessionsDir:  cfg.SessionsDir,
+		BrainDir:     cfg.BrainDir,
 	}))
 
 	srv := mcp.NewServer(reg)

config/models.yaml

@@ -1,11 +1,41 @@
-# Model routing table — three-layer priority:
-# 1. model param in MCP tool call (caller override)
-# 2. per-skill entry here
-# 3. default (fallback)
-default: ollama/qwen3-coder-30b-tuned
+# Model routing chains — three-layer priority:
+# 1. model param in MCP tool call (caller override — collapses to single entry, no escalation)
+# 2. per-skill chain here
+# 3. default_chain fallback
+
+verifier: claude-sonnet-4-6   # fixed verifier for all local tiers
+
+llama_swap_url: http://koala:8080   # for warm-state probing
+
+default_chain:
+  - ollama/qwen3-coder-30b-tuned
+  - claude-sonnet-4-6
 
 skills:
-  tdd:           ollama/qwen3-coder-30b-tuned
-  review:        ollama/devstral-tuned
-  debug:         ollama/deepseek-r1-tuned
-  retrospective: ollama/qwen3-coder-30b-tuned
+  tdd:
+    chain:
+      - ollama/qwen3-coder-30b-tuned
+      - claude-sonnet-4-6
+  review:
+    chain:
+      - ollama/devstral-tuned
+      - ollama/gemma4
+      - claude-sonnet-4-6
+  debug:
+    chain:
+      - ollama/deepseek-r1-tuned
+      - claude-sonnet-4-6
+  spec:
+    chain:
+      - ollama/phi4
+      - ollama/gemma4
+      - claude-sonnet-4-6
+      - claude-opus-4-6
+  retrospective:
+    chain:
+      - ollama/qwen3-coder-30b-tuned
+      - claude-sonnet-4-6
+  trainer:
+    chain:
+      - ollama/qwen3-coder-30b-tuned
+      - claude-sonnet-4-6

config/supervisor/debug.md

@@ -0,0 +1,31 @@
+# Debug Discipline
+
+You are a systematic debugger. Form hypotheses before suggesting fixes.
+
+## Iron laws
+1. Never suggest "try X and see what happens" — every hypothesis must have a specific expected outcome if correct
+2. Generate exactly 3-5 hypotheses, ordered by likelihood (most likely first)
+3. Never fix the bug — diagnose only; the caller decides what to do with the hypotheses
+
+## Output contract
+Return JSON result with:
+- `status`: "pass" (hypotheses generated) or "error" (error too ambiguous to analyse)
+- `phase`: "debug"
+- `skill`: "debug"
+- `file_path`: the most relevant file to the error (read it)
+- `runner_output`: your hypotheses, formatted as:
+  ```
+  HYPOTHESIS 1 (likelihood: high): <mechanism>
+  VERIFY: <exact command or file to check> → expected if correct: <specific output>
+
+  HYPOTHESIS 2 (likelihood: medium): <mechanism>
+  VERIFY: <exact command or file to check> → expected if correct: <specific output>
+  ```
+- `verified`: false — verification is the caller's job
+- `message`: "N hypotheses for: <one-line error summary>"
+
+## Rules
+1. Read the error and any context files provided before forming hypotheses
+2. Identify the failure mode first — what actually went wrong, not just what the error says
+3. For each hypothesis: name the mechanism, explain why it would produce this exact error, give a concrete verification command with expected output
+4. If the error is clearly a typo or trivial mistake, still form 3 hypotheses — surface the most likely cause as #1

config/supervisor/review.md

@@ -0,0 +1,30 @@
+# Code Review Discipline
+
+You are a disciplined code reviewer. Read files carefully before commenting.
+
+## Iron laws
+1. Never approve security vulnerabilities: command injection, SQL injection, credential exposure, path traversal, unchecked input at system boundaries
+2. Never approve silently swallowed errors — `err != nil` without wrapping or handling is always wrong
+3. Never approve missing validation at system boundaries (user input, external APIs, file reads)
+
+## Output contract
+Return JSON result with:
+- `status`: "pass" if no blocking issues; "fail" if any iron law is violated
+- `phase`: "review"
+- `skill`: "review"
+- `file_path`: first file reviewed
+- `runner_output`: full review formatted as:
+  ```
+  CRITICAL: <issue> at <file>:<line>
+  WARNING: <issue> at <file>:<line>
+  SUGGESTION: <issue> at <file>:<line>
+  ```
+- `verified`: true if you read all specified files; false if any were missing or unreadable
+- `message`: "N critical, M warnings, K suggestions" or "clean: <which iron law checks passed and why>"
+
+## Rules
+1. Read every file listed before writing feedback
+2. Check iron laws first — any violation is CRITICAL and sets status to "fail"
+3. Then check: correctness, test coverage for new code, Go style conventions
+4. Never rubber-stamp — if nothing is wrong, explain specifically which iron law checks you ran and why they passed
+5. Line references are required for every finding — "roughly around the middle" is not acceptable

config/supervisor/spec.md

@@ -0,0 +1,46 @@
+# Spec Writing Discipline
+
+You write structured implementation specs. Nothing is left ambiguous.
+
+## Iron laws
+1. Success criteria must be measurable — "the system is fast" is banned; "p99 < 200ms under 100 RPS" is valid
+2. Always include an explicit "Out of scope" section — if you don't draw the boundary, the developer will guess wrong
+3. Every technical decision in the approach must have a rationale
+
+## Output contract
+Return JSON result with:
+- `status`: "pass" (spec written) or "error" (requirements too ambiguous to spec without more input)
+- `phase`: "spec"
+- `skill`: "spec"
+- `file_path`: the output_path where the spec was written (absolute path)
+- `runner_output`: ""
+- `verified`: true if the file was written successfully
+- `message`: "spec written: <one-line summary of what was specced>"
+
+## Spec structure
+Write the spec as markdown to the output_path:
+
+```markdown
+# [Feature] Spec
+
+## Problem statement
+[What problem does this solve? For whom? Why now?]
+
+## Success criteria
+- [ ] [Criterion 1 — measurable and verifiable]
+- [ ] [Criterion 2 — measurable and verifiable]
+
+## Constraints
+[Non-negotiable requirements the solution must satisfy]
+
+## Out of scope
+[What we are explicitly NOT doing in this iteration]
+
+## Technical approach
+[Architecture decisions, key components, rationale for each choice]
+
+## Risks
+[What could go wrong, and how we'd mitigate it]
+```
+
+If the requirements are too vague to produce measurable success criteria, return status "error" with a message listing the specific questions that need answers.

config/supervisor/trainer-reader.md

@@ -0,0 +1,31 @@
+# Trainer Reader Discipline
+
+You scan session logs and identify candidate learning moments worth converting to training data.
+
+## What to look for
+- **SFT candidates**: the worker did exactly the right thing — a clean pattern worth reinforcing
+- **DPO candidates**: the worker first produced a wrong or suboptimal response, then corrected — you have both rejected and chosen
+
+## Scoring (1–5)
+- 5: novel pattern, clearly correct, generalises across projects
+- 4: good pattern, correct, somewhat project-specific but still useful
+- 3: correct but obvious — include only if especially clean
+- 2 or below: skip — too ambiguous or too context-specific
+
+## Output contract
+Return JSON result with:
+- `status`: "pass" or "error"
+- `phase`: "trainer"
+- `skill`: "trainer"
+- `file_path`: ""
+- `runner_output`: JSON array of candidates (valid JSON, not markdown):
+  [{"type":"sft","moment":"<what happened>","prompt":"<what was asked>","completion":"<what was done right>","score":4},
+   {"type":"dpo","moment":"<what happened>","prompt":"<what was asked>","chosen":"<correct>","rejected":"<incorrect>","score":3}]
+- `verified`: true
+- `message`: "N sft candidates, M dpo candidates found"
+
+## Rules
+1. Read all session entries in the task prompt
+2. Score each entry — only include entries scoring >= 3
+3. Prompt/completion fields must be phrased to generalise: no project-specific paths or names
+4. If no candidates score >= 3, return an empty array `[]` — never force low-quality candidates

config/supervisor/trainer-writer.md

@@ -0,0 +1,35 @@
+# Trainer Writer Discipline
+
+You receive candidate learning moments from the reader and write clean SFT/DPO training pairs.
+
+## Quality gate (apply before writing)
+- SFT: prompt must be phrased so it could come from any project, not just this one
+- DPO: chosen and rejected must be clearly distinguishable — skip if a reader can't tell which is better
+- Never include project-specific paths, variable names, or identifiers in any pair
+
+## Output contract
+Return JSON result with:
+- `status`: "pass" (pairs written or skipped due to quality) or "error" (candidates JSON was malformed)
+- `phase`: "trainer"
+- `skill`: "trainer"
+- `file_path`: path of the last file written (empty if nothing passed quality gate)
+- `runner_output`: "N SFT pairs written to brain/training-data/sft/, M DPO pairs to brain/training-data/dpo/" or "0 pairs passed quality gate"
+- `verified`: true if files were written; false if nothing passed
+- `message`: "N sft + M dpo pairs for session <id>" or "no pairs passed quality gate"
+
+## File format
+JSONL — one JSON object per line.
+
+SFT: `{"prompt": "...", "completion": "..."}`
+DPO: `{"prompt": "...", "chosen": "...", "rejected": "..."}`
+
+Write SFT to: `<brain_dir>/training-data/sft/<session_id>.jsonl`
+Write DPO to: `<brain_dir>/training-data/dpo/<session_id>.jsonl`
+
+Append to existing files if they exist (don't overwrite).
+
+## Rules
+1. Parse the `reader_candidates` JSON from the task prompt
+2. For each candidate: apply quality gate
+3. Write passing SFT candidates to sft JSONL, DPO candidates to dpo JSONL
+4. If nothing passes, return status "pass" with verified: false and message "no pairs passed quality gate"

docs/superpowers/specs/2026-04-20-model-orchestration-design.md

@@ -0,0 +1,322 @@
+# Model Orchestration Design
+
+**Date:** 2026-04-20  
+**Status:** Approved for implementation
+
+## Problem statement
+
+The hyperguild supervisor currently spawns a `claude --print` subprocess for every skill call. The model routing config (`models.yaml`) exists but is dead weight — the model name is injected as text into the task prompt and ignored. Every skill call costs Claude tokens regardless of task complexity or data sensitivity.
+
+## Goal
+
+Route skill work to the most appropriate model — weighing cost, latency, and quality — with Claude acting as the real supervisor: verifying outputs and deciding when to escalate. Local models on owned hardware handle the common case; Claude escalates through a chain to frontier models only when local quality is insufficient.
+
+## Success criteria
+
+- [ ] Each skill dispatches generation to its configured local model via LiteLLM by default
+- [ ] Claude verifies every local output and either accepts or escalates
+- [ ] Escalation walks a per-skill chain (local small → local large → Sonnet → Opus) with one attempt per tier
+- [ ] Every attempt (model, tier, duration, warm state, verdict) is logged in the session JSONL
+- [ ] Cloud tiers (Sonnet/Opus) self-certify — no separate verifier call
+- [ ] Zero changes to skill handlers — they call `ExecutorFn` exactly as today
+- [ ] `LiteLTMBaseURL` already in config; no new env vars required beyond `LLAMA_SWAP_URL`
+
+## Constraints
+
+- One attempt per tier before escalating (no retry within a tier)
+- Anthropic T&C: Claude is called normally via Anthropic API; local models are called directly via LiteLLM HTTP — no API redirection
+- `models.yaml` remains the single routing config file
+
+## Out of scope
+
+- Auto-rerouting based on real-time warm state (logged, not acted on — Phase 4)
+- Multi-tenant / public service exposure
+- RAG/CAG model boosting
+- Managed Agent cloud delegation (chain stub only in Phase 3)
+
+---
+
+## Architecture
+
+```
+MCP tool call (Claude Code)
+    ↓
+Skill handler — calls ExecutorFn (unchanged)
+    ↓
+Orchestrator.Run (implements ExecutorFn)
+    ├─ Resolve chain from models.yaml
+    ├─ For each model in chain:
+    │   ├─ [ollama/*] → LiteLLM executor → generate
+    │   │       ↓
+    │   │   Claude verifier (task + output + discipline)
+    │   │       ├─ accept  → return Result (log attempt)
+    │   │       └─ escalate → next tier (log attempt)
+    │   │
+    │   └─ [claude-*] → Claude executor (current) → generate + self-certify
+    │           └─ return Result (log attempt)
+    │
+    └─ All tiers exhausted → return best attempt with escalation note
+```
+
+Claude is always the verifier for local tiers. At cloud tiers, Claude generates and self-certifies — the verifier call is skipped.
+
+---
+
+## Components
+
+### 1. `internal/exec/litellm.go` — LiteLLM executor
+
+Calls `POST /v1/chat/completions` on the configured LiteLLM server. Implements the same `ExecutorFn` signature as the existing claude executor.
+
+```go
+type LiteLLMExecutor struct {
+    BaseURL    string
+    APIKey     string
+    HTTPClient *http.Client
+    Timeout    time.Duration
+}
+
+func NewLiteLLM(baseURL, apiKey string, timeout time.Duration) *LiteLLMExecutor
+
+func (e *LiteLLMExecutor) Run(ctx context.Context, req Request) (Result, error)
+```
+
+Request mapping:
+- `req.SkillPrompt` → system message
+- `req.TaskPrompt` → user message
+- `req.Model` → `model` field in the chat completions request
+
+Response handling: local models are prompted (via the discipline file output contract) to return a JSON object matching the `Result` schema. The executor attempts `json.Unmarshal` into `Result` directly — no envelope unwrapping needed (unlike the `--output-format json` claude envelope). If unmarshalling fails, the executor returns an error that the orchestrator treats as an automatic escalation trigger.
+
+### 2. `internal/exec/verifier.go` — Claude verifier
+
+A focused Claude call that judges local model output. Uses the existing `Executor` (claude subprocess) internally.
+
+```go
+type Verdict struct {
+    Accept   bool   `json:"accept"`
+    Feedback string `json:"feedback"` // reason if not accepting; empty if accept
+}
+
+type Verifier struct {
+    executor *Executor // the existing claude executor
+}
+
+func NewVerifier(executor *Executor) *Verifier
+
+func (v *Verifier) Verify(ctx context.Context, skillPrompt, taskPrompt string, output Result) (Verdict, error)
+```
+
+The verifier prompt gives Claude:
+1. The skill discipline file (so it knows the iron laws and output contract)
+2. The original task prompt (informed verification — Claude sees what was asked)
+3. The generated output
+4. A short instruction: "Does this output satisfy the discipline's iron laws and output contract? Reply with JSON: `{\"accept\": true|false, \"feedback\": \"...\"}`"
+
+The verifier uses a lightweight JSON schema for its own output (a `Verdict` schema), keeping the call fast.
+
+### 3. `internal/exec/orchestrator.go` — chain walker
+
+Implements `ExecutorFn`. Walks the escalation chain, delegating generation and verification per tier.
+
+```go
+type Chain []ChainEntry
+
+type ChainEntry struct {
+    Model    string // e.g. "ollama/phi4", "claude-sonnet-4-5"
+    Tier     string // "local" | "subagent" | "managed"
+    IsCloud  bool   // true for claude-* models; skips verifier
+}
+
+type Orchestrator struct {
+    chain    Chain
+    litellm  *LiteLLMExecutor
+    claude   *Executor
+    verifier *Verifier
+    llamaSwapURL string // for warm-state probe
+}
+
+func NewOrchestrator(chain Chain, litellm *LiteLLMExecutor, claude *Executor, verifier *Verifier, llamaSwapURL string) *Orchestrator
+
+func (o *Orchestrator) Run(ctx context.Context, req Request) (Result, error)
+```
+
+Algorithm:
+```
+for each entry in chain:
+    warm = probe llama-swap (if local tier)
+    start = now()
+    if entry.IsCloud:
+        result, err = claude.Run(ctx, req with entry.Model)
+        log attempt(model, tier, duration, warm, verified=true)
+        if err == nil: return result
+    else:
+        result, err = litellm.Run(ctx, req with entry.Model)
+        duration = now() - start
+        if err != nil:
+            log attempt(model, tier, duration, warm, verified=false)
+            continue  // automatic escalation on parse/network error
+        verdict = verifier.Verify(ctx, req.SkillPrompt, req.TaskPrompt, result)
+        log attempt(model, tier, duration, warm, verified=verdict.Accept)
+        if verdict.Accept: return result
+        // inject verifier feedback into next tier's task prompt
+        req.TaskPrompt = req.TaskPrompt + "\n\nPrior attempt feedback: " + verdict.Feedback
+
+return error("all tiers exhausted")
+```
+
+### 4. `internal/config/models.go` — chain parser
+
+Replaces the current single-model resolution with chain parsing.
+
+Updated `models.yaml` format:
+
+```yaml
+verifier: claude-sonnet-4-6   # fixed verifier for all local tiers
+
+llama_swap_url: http://koala:8080   # for warm-state probing
+
+default_chain:
+  - ollama/qwen3-coder-30b-tuned
+  - claude-sonnet-4-5
+
+skills:
+  tdd:
+    chain:
+      - ollama/qwen3-coder-30b-tuned
+      - claude-sonnet-4-5
+  review:
+    chain:
+      - ollama/devstral-tuned
+      - ollama/gemma4
+      - claude-sonnet-4-5
+  debug:
+    chain:
+      - ollama/deepseek-r1-tuned
+      - claude-sonnet-4-5
+  spec:
+    chain:
+      - ollama/phi4
+      - ollama/gemma4
+      - claude-sonnet-4-5
+      - claude-opus-4-6
+  retrospective:
+    chain:
+      - ollama/qwen3-coder-30b-tuned
+      - claude-sonnet-4-5
+  trainer:
+    chain:
+      - ollama/qwen3-coder-30b-tuned
+      - claude-sonnet-4-5
+```
+
+The parser exposes:
+```go
+func (m *Models) ChainFor(skill string) Chain
+func (m *Models) Verifier() string
+func (m *Models) LlamaSwapURL() string
+```
+
+Caller override (`model` param in MCP tool call) pins the chain to a single entry — one model, no escalation. This preserves the existing override behaviour for power users.
+
+### 5. `internal/session/session.go` — updated `Attempt` struct
+
+```go
+type Attempt struct {
+    Attempt       int    `json:"attempt"`
+    Model         string `json:"model"`
+    Tier          string `json:"tier"`          // local | subagent | managed
+    DurationMs    int64  `json:"duration_ms"`
+    WarmStart     bool   `json:"warm_start"`    // model was already loaded in llama-swap
+    Verified      bool   `json:"verified"`
+    Verdict       string `json:"verdict,omitempty"` // accept | escalate | error
+    Feedback      string `json:"feedback,omitempty"` // verifier feedback on escalation
+    OutputSummary string `json:"output_summary,omitempty"`
+    RunnerOutput  string `json:"runner_output,omitempty"`
+}
+```
+
+### 6. `cmd/supervisor/main.go` — one wiring change
+
+```go
+// Before:
+reg.Register(review.New(review.Config{ExecutorFn: executor.Run, ...}))
+
+// After:
+chain := models.ChainFor("review")
+orch := exec.NewOrchestrator(chain, litellmExec, claudeExec, verifier, models.LlamaSwapURL())
+reg.Register(review.New(review.Config{ExecutorFn: orch.Run, ...}))
+```
+
+One orchestrator per skill, sharing the same `litellmExec`, `claudeExec`, and `verifier` instances.
+
+---
+
+## Data flow example: `review` skill call
+
+1. Claude Code calls `review` tool with `files: ["internal/foo.go"]`
+2. Skill handler builds task prompt, calls `orch.Run`
+3. Orchestrator resolves chain: `[devstral, gemma4, sonnet]`
+4. Probes llama-swap: devstral is warm
+5. LiteLLM calls devstral → returns JSON result
+6. Verifier asks Claude: "does this review satisfy the iron laws?"
+7. Claude: `{"accept": false, "feedback": "missing line references for all findings"}`
+8. Orchestrator logs attempt #1 (devstral, local, 4200ms, warm, escalate)
+9. Injects feedback into task prompt, calls gemma4
+10. Verifier: `{"accept": true}`
+11. Orchestrator logs attempt #2 (gemma4, local, 6100ms, cold, accept)
+12. Returns result to skill handler → MCP response
+
+Session JSONL records both attempts. You can see: devstral was warm but produced weak output; gemma4 was cold but passed.
+
+---
+
+## Observability
+
+Session JSONL is the primary store. Each `Entry.Attempts` slice records the full escalation trail. To analyse across sessions:
+
+```bash
+# Which models are escalating most?
+jq -r '.attempts[] | select(.verdict == "escalate") | .model' brain/sessions/*.jsonl | sort | uniq -c
+
+# Average latency per model
+jq -r '.attempts[] | [.model, .duration_ms] | @tsv' brain/sessions/*.jsonl | awk '{sum[$1]+=$2; n[$1]++} END {for (m in sum) print m, sum[m]/n[m]}'
+
+# Cold start frequency
+jq -r '.attempts[] | select(.warm_start == false) | .model' brain/sessions/*.jsonl | sort | uniq -c
+```
+
+No new metrics infrastructure needed for Phase 3. Phase 4 can build a dashboard on top of this data.
+
+---
+
+## Error handling
+
+| Scenario | Behaviour |
+|----------|-----------|
+| LiteLLM unreachable | Log attempt as error, escalate immediately |
+| Local model returns unparseable JSON | Log attempt as error, escalate |
+| Verifier call fails | Log, treat as escalate (safe default) |
+| All tiers exhausted | Return error to skill handler; skill returns MCP error to caller |
+| Caller passes `model` override | Single-entry chain, no escalation, no verifier call |
+
+---
+
+## Testing approach
+
+- `TestLiteLLMExecutor`: mock HTTP server returning valid/invalid JSON; verify parse logic and error escalation
+- `TestVerifier`: fake claude executor returning accept/escalate verdicts; verify prompt construction
+- `TestOrchestrator`: table-driven — chains of 1/2/3 tiers, various accept/escalate/error combinations; verify attempt log contents and final result
+- `TestModelsChainFor`: YAML parsing for all skill overrides and default_chain fallback
+- Integration smoke test: start real LiteLLM (or mock), call `review` tool via MCP, verify attempt log written
+
+---
+
+## Risks
+
+| Risk | Mitigation |
+|------|------------|
+| Local models ignore output contract → bad JSON | Discipline files already specify JSON output contract; parse failure auto-escalates |
+| Verifier Claude call adds latency to every local attempt | Verifier prompt is small and fast; acceptable tradeoff for quality gate |
+| llama-swap warm probe adds overhead | Probe is a single lightweight HTTP GET; timeout at 200ms, treat failure as `warm_start: false` |
+| Chain exhaustion leaves caller with no result | Return structured error via MCP; caller can retry with explicit `model` override |

internal/config/models.go

@@ -7,9 +7,15 @@ import (
 	"gopkg.in/yaml.v3"
 )
 
+type skillChain struct {
+	Chain []string `yaml:"chain"`
+}
+
 type modelsFile struct {
-	Default string            `yaml:"default"`
-	Skills  map[string]string `yaml:"skills"`
+	Verifier     string                `yaml:"verifier"`
+	LlamaSwapURL string                `yaml:"llama_swap_url"`
+	DefaultChain []string              `yaml:"default_chain"`
+	Skills       map[string]skillChain `yaml:"skills"`
 }
 
 type Models struct {
@@ -28,16 +34,23 @@ func LoadModels(path string) (Models, error) {
 	return Models{data: f}, nil
 }
 
-// Resolve returns the model for a skill, respecting three-layer priority:
-// 1. override (from MCP call) — highest
-// 2. per-skill default from models.yaml
-// 3. global default
-func (m Models) Resolve(skill, override string) string {
+// Verifier returns the model name to use for all local-tier output verification.
+func (m Models) Verifier() string { return m.data.Verifier }
+
+// LlamaSwapURL returns the llama-swap base URL for warm-state probing.
+func (m Models) LlamaSwapURL() string { return m.data.LlamaSwapURL }
+
+// ChainFor returns the ordered list of model names for a skill.
+// If override is non-empty, returns a single-entry chain (no escalation).
+// Falls back to default_chain when the skill has no explicit entry.
+func (m Models) ChainFor(skill, override string) []string {
 	if override != "" {
-		return override
+		return []string{override}
 	}
-	if model, ok := m.data.Skills[skill]; ok {
-		return model
+	if sc, ok := m.data.Skills[skill]; ok && len(sc.Chain) > 0 {
+		return sc.Chain
 	}
-	return m.data.Default
+	out := make([]string, len(m.data.DefaultChain))
+	copy(out, m.data.DefaultChain)
+	return out
 }

internal/config/models_test.go

@@ -10,35 +10,71 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
-func TestModelsResolve(t *testing.T) {
-	yaml := `
-default: ollama/default-model
+const testYAML = `
+verifier: claude-sonnet-4-6
+llama_swap_url: http://koala:8080
+
+default_chain:
+  - ollama/qwen3-coder-30b-tuned
+  - claude-sonnet-4-6
+
 skills:
-  tdd: ollama/qwen3-coder-30b-tuned
-  review: ollama/devstral-tuned
+  review:
+    chain:
+      - ollama/devstral-tuned
+      - ollama/gemma4
+      - claude-sonnet-4-6
+  spec:
+    chain:
+      - ollama/phi4
+      - claude-opus-4-6
 `
+
+func writeModels(t *testing.T, content string) string {
+	t.Helper()
 	f := filepath.Join(t.TempDir(), "models.yaml")
-	require.NoError(t, os.WriteFile(f, []byte(yaml), 0644))
-
-	m, err := config.LoadModels(f)
-	require.NoError(t, err)
-
-	assert.Equal(t, "ollama/qwen3-coder-30b-tuned", m.Resolve("tdd", ""))
-	assert.Equal(t, "ollama/devstral-tuned", m.Resolve("review", ""))
-	assert.Equal(t, "ollama/default-model", m.Resolve("unknown", ""))
+	require.NoError(t, os.WriteFile(f, []byte(content), 0644))
+	return f
 }
 
-func TestModelsOverride(t *testing.T) {
-	yaml := `
-default: ollama/default-model
-skills:
-  tdd: ollama/qwen3-coder-30b-tuned
-`
-	f := filepath.Join(t.TempDir(), "models.yaml")
-	require.NoError(t, os.WriteFile(f, []byte(yaml), 0644))
+func TestModelsVerifier(t *testing.T) {
+	m, err := config.LoadModels(writeModels(t, testYAML))
+	require.NoError(t, err)
+	assert.Equal(t, "claude-sonnet-4-6", m.Verifier())
+}
 
-	m, err := config.LoadModels(f)
+func TestModelsLlamaSwapURL(t *testing.T) {
+	m, err := config.LoadModels(writeModels(t, testYAML))
+	require.NoError(t, err)
+	assert.Equal(t, "http://koala:8080", m.LlamaSwapURL())
+}
+
+func TestModelsChainForSkillOverride(t *testing.T) {
+	m, err := config.LoadModels(writeModels(t, testYAML))
 	require.NoError(t, err)
 
-	assert.Equal(t, "anthropic/claude-sonnet-4-6", m.Resolve("tdd", "anthropic/claude-sonnet-4-6"))
+	chain := m.ChainFor("review", "")
+	require.Len(t, chain, 3)
+	assert.Equal(t, "ollama/devstral-tuned", chain[0])
+	assert.Equal(t, "ollama/gemma4", chain[1])
+	assert.Equal(t, "claude-sonnet-4-6", chain[2])
+}
+
+func TestModelsChainForDefaultFallback(t *testing.T) {
+	m, err := config.LoadModels(writeModels(t, testYAML))
+	require.NoError(t, err)
+
+	chain := m.ChainFor("trainer", "") // not in skills map
+	require.Len(t, chain, 2)
+	assert.Equal(t, "ollama/qwen3-coder-30b-tuned", chain[0])
+	assert.Equal(t, "claude-sonnet-4-6", chain[1])
+}
+
+func TestModelsChainForCallerOverride(t *testing.T) {
+	m, err := config.LoadModels(writeModels(t, testYAML))
+	require.NoError(t, err)
+
+	chain := m.ChainFor("review", "claude-opus-4-6")
+	require.Len(t, chain, 1)
+	assert.Equal(t, "claude-opus-4-6", chain[0])
 }

internal/exec/executor.go

@@ -72,8 +72,11 @@ func (e *Executor) Run(ctx context.Context, req Request) (Result, error) {
 		"--tools", tools,
 		"--json-schema", Schema,
 		"--output-format", "json",
-		prompt,
 	}
+	if strings.HasPrefix(req.Model, "claude-") {
+		args = append(args, "--model", req.Model)
+	}
+	args = append(args, prompt)
 
 	cmd := exec.CommandContext(ctx, e.cfg.ClaudeBinary, args...)
 	cmd.Env = append(os.Environ(), "LITELLM_API_KEY="+e.cfg.LiteLLMAPIKey)

internal/exec/executor_test.go

@@ -75,3 +75,58 @@ func TestExecutorTimesOut(t *testing.T) {
 	_, err := ex.Run(context.Background(), iexec.Request{TaskPrompt: "slow"})
 	assert.ErrorContains(t, err, "timeout")
 }
+
+func TestExecutorPassesModelFlagForCloudModel(t *testing.T) {
+	// The script captures its args to a temp file so we can assert --model was passed.
+	argsFile := filepath.Join(t.TempDir(), "args.txt")
+	envelope := `{"type":"result","subtype":"success","is_error":false,"structured_output":{"status":"pass","phase":"review","skill":"review","file_path":"","runner_output":"","verified":true,"model_used":"claude-sonnet-4-6","message":"ok"}}`
+
+	dir := t.TempDir()
+	script := filepath.Join(dir, "claude")
+	content := "#!/bin/sh\necho \"$@\" > " + argsFile + "\necho '" + envelope + "'\n"
+	require.NoError(t, os.WriteFile(script, []byte(content), 0755))
+
+	ex := iexec.New(iexec.Config{
+		ClaudeBinary: script,
+		SystemPrompt: "sys",
+		Timeout:      5 * time.Second,
+	})
+
+	_, err := ex.Run(context.Background(), iexec.Request{
+		SkillPrompt: "review rules",
+		TaskPrompt:  "do review",
+		Model:       "claude-sonnet-4-6",
+	})
+	require.NoError(t, err)
+
+	argsData, err := os.ReadFile(argsFile)
+	require.NoError(t, err)
+	assert.Contains(t, string(argsData), "--model claude-sonnet-4-6")
+}
+
+func TestExecutorSkipsModelFlagForLocalModel(t *testing.T) {
+	argsFile := filepath.Join(t.TempDir(), "args.txt")
+	envelope := `{"type":"result","subtype":"success","is_error":false,"structured_output":{"status":"pass","phase":"review","skill":"review","file_path":"","runner_output":"","verified":true,"model_used":"ollama/devstral","message":"ok"}}`
+
+	dir := t.TempDir()
+	script := filepath.Join(dir, "claude")
+	content := "#!/bin/sh\necho \"$@\" > " + argsFile + "\necho '" + envelope + "'\n"
+	require.NoError(t, os.WriteFile(script, []byte(content), 0755))
+
+	ex := iexec.New(iexec.Config{
+		ClaudeBinary: script,
+		SystemPrompt: "sys",
+		Timeout:      5 * time.Second,
+	})
+
+	_, err := ex.Run(context.Background(), iexec.Request{
+		SkillPrompt: "review rules",
+		TaskPrompt:  "do review",
+		Model:       "ollama/devstral",
+	})
+	require.NoError(t, err)
+
+	argsData, err := os.ReadFile(argsFile)
+	require.NoError(t, err)
+	assert.NotContains(t, string(argsData), "--model")
+}

internal/exec/litellm.go

@@ -0,0 +1,103 @@
+package exec
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"time"
+)
+
+// LiteLLMExecutor calls a LiteLLM-compatible /v1/chat/completions endpoint.
+// Local models are expected to return a JSON object matching the Result schema
+// as their response content — no envelope.
+type LiteLLMExecutor struct {
+	baseURL    string
+	apiKey     string
+	httpClient *http.Client
+}
+
+// NewLiteLLM creates a LiteLLMExecutor.
+// timeout applies to the full HTTP round-trip per call.
+func NewLiteLLM(baseURL, apiKey string, timeout time.Duration) *LiteLLMExecutor {
+	return &LiteLLMExecutor{
+		baseURL: baseURL,
+		apiKey:  apiKey,
+		httpClient: &http.Client{Timeout: timeout},
+	}
+}
+
+type litellmMessage struct {
+	Role    string `json:"role"`
+	Content string `json:"content"`
+}
+
+type litellmRequest struct {
+	Model    string           `json:"model"`
+	Messages []litellmMessage `json:"messages"`
+}
+
+type litellmChoice struct {
+	Message litellmMessage `json:"message"`
+}
+
+type litellmResponse struct {
+	Choices []litellmChoice `json:"choices"`
+}
+
+// Run dispatches req to the LiteLLM server and parses the Result from the
+// assistant message content. Returns an error on network failure, non-200
+// status, or unparseable/invalid JSON — all of which the Orchestrator treats
+// as automatic escalation triggers.
+func (e *LiteLLMExecutor) Run(ctx context.Context, req Request) (Result, error) {
+	body := litellmRequest{
+		Model: req.Model,
+		Messages: []litellmMessage{
+			{Role: "system", Content: req.SkillPrompt},
+			{Role: "user", Content: req.TaskPrompt},
+		},
+	}
+
+	bodyBytes, err := json.Marshal(body)
+	if err != nil {
+		return Result{}, fmt.Errorf("litellm: marshal request: %w", err)
+	}
+
+	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, e.baseURL+"/v1/chat/completions", bytes.NewReader(bodyBytes))
+	if err != nil {
+		return Result{}, fmt.Errorf("litellm: create request: %w", err)
+	}
+	httpReq.Header.Set("Content-Type", "application/json")
+	if e.apiKey != "" {
+		httpReq.Header.Set("Authorization", "Bearer "+e.apiKey)
+	}
+
+	resp, err := e.httpClient.Do(httpReq)
+	if err != nil {
+		return Result{}, fmt.Errorf("litellm: request failed: %w", err)
+	}
+	defer resp.Body.Close() //nolint:errcheck
+
+	if resp.StatusCode != http.StatusOK {
+		return Result{}, fmt.Errorf("litellm: server returned status %d", resp.StatusCode)
+	}
+
+	var chatResp litellmResponse
+	if err := json.NewDecoder(resp.Body).Decode(&chatResp); err != nil {
+		return Result{}, fmt.Errorf("litellm: decode response: %w", err)
+	}
+	if len(chatResp.Choices) == 0 {
+		return Result{}, fmt.Errorf("litellm: no choices in response")
+	}
+
+	content := chatResp.Choices[0].Message.Content
+	var result Result
+	if err := json.Unmarshal([]byte(content), &result); err != nil {
+		return Result{}, fmt.Errorf("litellm: parse result JSON: %w — content: %s", err, content)
+	}
+	if err := result.Validate(); err != nil {
+		return Result{}, fmt.Errorf("litellm: invalid result: %w", err)
+	}
+	return result, nil
+}

internal/exec/litellm_test.go

@@ -0,0 +1,112 @@
+package exec_test
+
+import (
+	"context"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+	"time"
+
+	iexec "github.com/mathiasbq/supervisor/internal/exec"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func validLiteLLMResult() iexec.Result {
+	return iexec.Result{
+		Status:    "pass",
+		Phase:     "review",
+		Skill:     "review",
+		ModelUsed: "ollama/devstral",
+		Message:   "looks good",
+	}
+}
+
+func chatResponseFor(t *testing.T, result iexec.Result) []byte {
+	t.Helper()
+	content, err := json.Marshal(result)
+	require.NoError(t, err)
+	resp := map[string]any{
+		"choices": []map[string]any{
+			{"message": map[string]any{"role": "assistant", "content": string(content)}},
+		},
+	}
+	data, err := json.Marshal(resp)
+	require.NoError(t, err)
+	return data
+}
+
+func TestLiteLLMParsesValidResult(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		assert.Equal(t, "/v1/chat/completions", r.URL.Path)
+		assert.Equal(t, "application/json", r.Header.Get("Content-Type"))
+		w.Header().Set("Content-Type", "application/json")
+		w.WriteHeader(http.StatusOK)
+		_, _ = w.Write(chatResponseFor(t, validLiteLLMResult()))
+	}))
+	defer srv.Close()
+
+	ex := iexec.NewLiteLLM(srv.URL, "", 5*time.Second)
+	result, err := ex.Run(context.Background(), iexec.Request{
+		SkillPrompt: "review rules",
+		TaskPrompt:  "review the code",
+		Model:       "ollama/devstral",
+	})
+	require.NoError(t, err)
+	assert.Equal(t, "pass", result.Status)
+	assert.Equal(t, "review", result.Skill)
+}
+
+func TestLiteLLMSendsAuthHeader(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		assert.Equal(t, "Bearer secret", r.Header.Get("Authorization"))
+		w.Header().Set("Content-Type", "application/json")
+		w.WriteHeader(http.StatusOK)
+		_, _ = w.Write(chatResponseFor(t, validLiteLLMResult()))
+	}))
+	defer srv.Close()
+
+	ex := iexec.NewLiteLLM(srv.URL, "secret", 5*time.Second)
+	_, err := ex.Run(context.Background(), iexec.Request{Model: "x", TaskPrompt: "t", SkillPrompt: "s"})
+	require.NoError(t, err)
+}
+
+func TestLiteLLMErrorOnNonOKStatus(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusServiceUnavailable)
+	}))
+	defer srv.Close()
+
+	ex := iexec.NewLiteLLM(srv.URL, "", 5*time.Second)
+	_, err := ex.Run(context.Background(), iexec.Request{Model: "x", TaskPrompt: "t"})
+	assert.ErrorContains(t, err, "503")
+}
+
+func TestLiteLLMErrorOnUnparsableJSON(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		w.WriteHeader(http.StatusOK)
+		resp := map[string]any{
+			"choices": []map[string]any{
+				{"message": map[string]any{"role": "assistant", "content": "not json at all"}},
+			},
+		}
+		data, _ := json.Marshal(resp)
+		_, _ = w.Write(data)
+	}))
+	defer srv.Close()
+
+	ex := iexec.NewLiteLLM(srv.URL, "", 5*time.Second)
+	_, err := ex.Run(context.Background(), iexec.Request{Model: "x", TaskPrompt: "t"})
+	assert.Error(t, err)
+}
+
+func TestLiteLLMRespectsContextCancellation(t *testing.T) {
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel() // Cancel immediately
+
+	ex := iexec.NewLiteLLM("http://invalid.example.com", "", 1*time.Second)
+	_, err := ex.Run(ctx, iexec.Request{Model: "x", TaskPrompt: "t"})
+	assert.Error(t, err)
+}

internal/exec/orchestrator.go

@@ -0,0 +1,197 @@
+package exec
+
+import (
+	"context"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+	"time"
+)
+
+// ChainEntry is one tier in an escalation chain.
+type ChainEntry struct {
+	Model   string // e.g. "ollama/phi4", "claude-sonnet-4-6"
+	Tier    string // "local" | "subagent" | "managed"
+	IsCloud bool   // true for claude-* models; skips verifier call
+}
+
+// EntryFor builds a ChainEntry from a model name string.
+func EntryFor(model string) ChainEntry {
+	cloud := strings.HasPrefix(model, "claude-")
+	tier := "local"
+	if cloud {
+		tier = "subagent"
+	}
+	return ChainEntry{Model: model, Tier: tier, IsCloud: cloud}
+}
+
+// AttemptRecord captures the outcome of one tier attempt for session logging.
+type AttemptRecord struct {
+	Model      string
+	Tier       string
+	DurationMs int64
+	WarmStart  bool
+	Verdict    string // "accept" | "escalate" | "error"
+	Feedback   string
+}
+
+// VerifierFn is the interface the orchestrator uses to verify local output.
+type VerifierFn interface {
+	Verify(ctx context.Context, skillPrompt, taskPrompt string, output Result) (Verdict, error)
+}
+
+// ExecutorRunFn is the signature of Executor.Run and LiteLLMExecutor.Run.
+type ExecutorRunFn func(ctx context.Context, req Request) (Result, error)
+
+// Orchestrator walks an escalation chain, delegating generation and verification.
+// It implements the ExecutorFn shape expected by skill handlers.
+type Orchestrator struct {
+	chain        []ChainEntry
+	localRun     ExecutorRunFn // for local (non-cloud) tiers; may be nil
+	cloudRun     ExecutorRunFn // for cloud tiers; may be nil
+	verifier     VerifierFn
+	llamaSwapURL string
+	attempts     *[]AttemptRecord
+}
+
+// NewOrchestrator creates an Orchestrator.
+// attempts is a pointer to a slice that will be appended to on each tier attempt.
+// Pass nil for localRun or cloudRun if no tiers of that type exist in the chain.
+func NewOrchestrator(
+	chain []ChainEntry,
+	localRun ExecutorRunFn,
+	cloudRun ExecutorRunFn,
+	verifier VerifierFn,
+	llamaSwapURL string,
+	attempts *[]AttemptRecord,
+) *Orchestrator {
+	return &Orchestrator{
+		chain:        chain,
+		localRun:     localRun,
+		cloudRun:     cloudRun,
+		verifier:     verifier,
+		llamaSwapURL: llamaSwapURL,
+		attempts:     attempts,
+	}
+}
+
+// Run walks the escalation chain and returns the first accepted result.
+// Satisfies the ExecutorFn signature: func(context.Context, Request) (Result, error).
+func (o *Orchestrator) Run(ctx context.Context, req Request) (Result, error) {
+	taskPrompt := req.TaskPrompt
+
+	for _, entry := range o.chain {
+		warm := o.probeWarm(entry.Model)
+		start := time.Now()
+
+		tierReq := req
+		tierReq.Model = entry.Model
+		tierReq.TaskPrompt = taskPrompt
+
+		if entry.IsCloud {
+			result, genErr := o.cloudRun(ctx, tierReq)
+			dur := time.Since(start).Milliseconds()
+			verdict := "accept"
+			if genErr != nil {
+				verdict = "error"
+			}
+			o.appendAttempt(AttemptRecord{
+				Model:      entry.Model,
+				Tier:       entry.Tier,
+				DurationMs: dur,
+				WarmStart:  warm,
+				Verdict:    verdict,
+			})
+			if genErr == nil {
+				return result, nil
+			}
+			continue
+		}
+
+		// Local tier.
+		result, genErr := o.localRun(ctx, tierReq)
+		dur := time.Since(start).Milliseconds()
+
+		if genErr != nil {
+			o.appendAttempt(AttemptRecord{
+				Model:      entry.Model,
+				Tier:       entry.Tier,
+				DurationMs: dur,
+				WarmStart:  warm,
+				Verdict:    "error",
+				Feedback:   genErr.Error(),
+			})
+			continue
+		}
+
+		verdict, verErr := o.verifier.Verify(ctx, req.SkillPrompt, taskPrompt, result)
+		if verErr != nil {
+			// Treat verifier failure as escalate (safe default).
+			o.appendAttempt(AttemptRecord{
+				Model:      entry.Model,
+				Tier:       entry.Tier,
+				DurationMs: dur,
+				WarmStart:  warm,
+				Verdict:    "escalate",
+				Feedback:   "verifier error: " + verErr.Error(),
+			})
+			continue
+		}
+
+		if verdict.Accept {
+			o.appendAttempt(AttemptRecord{
+				Model:      entry.Model,
+				Tier:       entry.Tier,
+				DurationMs: dur,
+				WarmStart:  warm,
+				Verdict:    "accept",
+			})
+			return result, nil
+		}
+
+		o.appendAttempt(AttemptRecord{
+			Model:      entry.Model,
+			Tier:       entry.Tier,
+			DurationMs: dur,
+			WarmStart:  warm,
+			Verdict:    "escalate",
+			Feedback:   verdict.Feedback,
+		})
+		// Inject verifier feedback into the next tier's task prompt.
+		taskPrompt = taskPrompt + "\n\nPrior attempt feedback: " + verdict.Feedback
+	}
+
+	return Result{}, fmt.Errorf("all tiers exhausted after %d attempt(s)", len(o.chain))
+}
+
+func (o *Orchestrator) appendAttempt(rec AttemptRecord) {
+	if o.attempts != nil {
+		*o.attempts = append(*o.attempts, rec)
+	}
+}
+
+// probeWarm checks whether the model is currently loaded in llama-swap.
+// Returns false on any error or if llamaSwapURL is empty.
+func (o *Orchestrator) probeWarm(model string) bool {
+	if o.llamaSwapURL == "" {
+		return false
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), 200*time.Millisecond)
+	defer cancel()
+
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, o.llamaSwapURL+"/v1/models", nil)
+	if err != nil {
+		return false
+	}
+	resp, err := http.DefaultClient.Do(req)
+	if err != nil {
+		return false
+	}
+	defer resp.Body.Close() //nolint:errcheck
+	body, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return false
+	}
+	return strings.Contains(string(body), model)
+}

internal/exec/orchestrator_test.go

@@ -0,0 +1,151 @@
+package exec_test
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	iexec "github.com/mathiasbq/supervisor/internal/exec"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// stubRunFn returns preset results sequentially.
+type stubRunFn struct {
+	calls   []stubCall
+	callIdx int
+}
+
+type stubCall struct {
+	result iexec.Result
+	err    error
+}
+
+func (s *stubRunFn) Run(_ context.Context, _ iexec.Request) (iexec.Result, error) {
+	if s.callIdx >= len(s.calls) {
+		return iexec.Result{}, errors.New("unexpected call")
+	}
+	c := s.calls[s.callIdx]
+	s.callIdx++
+	return c.result, c.err
+}
+
+// stubVerifier returns preset verdicts sequentially.
+type stubVerifier struct {
+	verdicts []iexec.Verdict
+	idx      int
+}
+
+func (s *stubVerifier) Verify(_ context.Context, _, _ string, _ iexec.Result) (iexec.Verdict, error) {
+	if s.idx >= len(s.verdicts) {
+		return iexec.Verdict{}, errors.New("unexpected verify call")
+	}
+	v := s.verdicts[s.idx]
+	s.idx++
+	return v, nil
+}
+
+func okResult(skill string) iexec.Result {
+	return iexec.Result{Status: "pass", Phase: "review", Skill: skill, Message: "ok", ModelUsed: "m"}
+}
+
+func TestOrchestratorSingleLocalAccept(t *testing.T) {
+	local := &stubRunFn{calls: []stubCall{{result: okResult("review")}}}
+	verifier := &stubVerifier{verdicts: []iexec.Verdict{{Accept: true}}}
+
+	var attempts []iexec.AttemptRecord
+	orch := iexec.NewOrchestrator(
+		[]iexec.ChainEntry{{Model: "ollama/devstral", Tier: "local", IsCloud: false}},
+		local.Run, nil, verifier, "", &attempts,
+	)
+
+	result, err := orch.Run(context.Background(), iexec.Request{TaskPrompt: "review"})
+	require.NoError(t, err)
+	assert.Equal(t, "pass", result.Status)
+	require.Len(t, attempts, 1)
+	assert.Equal(t, "local", attempts[0].Tier)
+	assert.Equal(t, "accept", attempts[0].Verdict)
+}
+
+func TestOrchestratorEscalatesOnVerifierReject(t *testing.T) {
+	local := &stubRunFn{calls: []stubCall{
+		{result: iexec.Result{Status: "fail", Phase: "review", Skill: "review", Message: "weak"}},
+		{result: okResult("review")},
+	}}
+	verifier := &stubVerifier{verdicts: []iexec.Verdict{
+		{Accept: false, Feedback: "missing line refs"},
+		{Accept: true},
+	}}
+
+	var attempts []iexec.AttemptRecord
+	orch := iexec.NewOrchestrator(
+		[]iexec.ChainEntry{
+			{Model: "ollama/devstral", Tier: "local", IsCloud: false},
+			{Model: "ollama/gemma4", Tier: "local", IsCloud: false},
+		},
+		local.Run, nil, verifier, "", &attempts,
+	)
+
+	result, err := orch.Run(context.Background(), iexec.Request{TaskPrompt: "review"})
+	require.NoError(t, err)
+	assert.Equal(t, "pass", result.Status)
+	require.Len(t, attempts, 2)
+	assert.Equal(t, "escalate", attempts[0].Verdict)
+	assert.Equal(t, "missing line refs", attempts[0].Feedback)
+	assert.Equal(t, "accept", attempts[1].Verdict)
+}
+
+func TestOrchestratorEscalatesOnLocalError(t *testing.T) {
+	local := &stubRunFn{calls: []stubCall{
+		{err: errors.New("network failure")},
+		{result: okResult("review")},
+	}}
+	verifier := &stubVerifier{verdicts: []iexec.Verdict{{Accept: true}}}
+
+	var attempts []iexec.AttemptRecord
+	orch := iexec.NewOrchestrator(
+		[]iexec.ChainEntry{
+			{Model: "ollama/devstral", Tier: "local", IsCloud: false},
+			{Model: "ollama/gemma4", Tier: "local", IsCloud: false},
+		},
+		local.Run, nil, verifier, "", &attempts,
+	)
+
+	_, err := orch.Run(context.Background(), iexec.Request{TaskPrompt: "review"})
+	require.NoError(t, err)
+	require.Len(t, attempts, 2)
+	assert.Equal(t, "error", attempts[0].Verdict)
+	assert.Equal(t, "accept", attempts[1].Verdict)
+}
+
+func TestOrchestratorCloudTierSelfCertifies(t *testing.T) {
+	cloud := &stubRunFn{calls: []stubCall{{result: okResult("review")}}}
+	verifier := &stubVerifier{} // no verdicts — must not be called
+
+	var attempts []iexec.AttemptRecord
+	orch := iexec.NewOrchestrator(
+		[]iexec.ChainEntry{{Model: "claude-sonnet-4-6", Tier: "subagent", IsCloud: true}},
+		nil, cloud.Run, verifier, "", &attempts,
+	)
+
+	result, err := orch.Run(context.Background(), iexec.Request{TaskPrompt: "review"})
+	require.NoError(t, err)
+	assert.Equal(t, "pass", result.Status)
+	require.Len(t, attempts, 1)
+	assert.Equal(t, "subagent", attempts[0].Tier)
+	assert.Equal(t, "accept", attempts[0].Verdict)
+	assert.Equal(t, 0, verifier.idx) // verifier never called
+}
+
+func TestOrchestratorAllTiersExhausted(t *testing.T) {
+	local := &stubRunFn{calls: []stubCall{{err: errors.New("unavailable")}}}
+
+	var attempts []iexec.AttemptRecord
+	orch := iexec.NewOrchestrator(
+		[]iexec.ChainEntry{{Model: "ollama/devstral", Tier: "local", IsCloud: false}},
+		local.Run, nil, &stubVerifier{}, "", &attempts,
+	)
+
+	_, err := orch.Run(context.Background(), iexec.Request{TaskPrompt: "review"})
+	assert.ErrorContains(t, err, "all tiers exhausted")
+}

internal/exec/result.go

@@ -10,7 +10,7 @@ import (
 // validates its own output before returning.
 type Result struct {
 	Status       string `json:"status"`        // pass | fail | error
-	Phase        string `json:"phase"`         // red | green | refactor
+	Phase        string `json:"phase"`         // red | green | refactor | retrospective | review | debug | spec | trainer
 	Skill        string `json:"skill"`         // tdd | review | ...
 	FilePath     string `json:"file_path"`     // absolute path to generated file
 	RunnerOutput string `json:"runner_output"` // raw stdout+stderr from test runner
@@ -25,6 +25,10 @@ var validPhases = map[string]bool{
 	"green":         true,
 	"refactor":      true,
 	"retrospective": true,
+	"review":        true,
+	"debug":         true,
+	"spec":          true,
+	"trainer":       true,
 }
 
 func (r Result) Validate() error {
@@ -33,7 +37,7 @@ func (r Result) Validate() error {
 		errs = append(errs, "status must be pass|fail|error, got: "+r.Status)
 	}
 	if !validPhases[r.Phase] {
-		errs = append(errs, "phase must be red|green|refactor, got: "+r.Phase)
+		errs = append(errs, "phase must be one of red|green|refactor|retrospective|review|debug|spec|trainer, got: "+r.Phase)
 	}
 	if r.Skill == "" {
 		errs = append(errs, "skill is required")
@@ -50,7 +54,7 @@ const Schema = `{
   "required": ["status","phase","skill","file_path","runner_output","verified","model_used","message"],
   "properties": {
     "status":        {"type": "string", "enum": ["pass","fail","error"]},
-    "phase":         {"type": "string", "enum": ["red","green","refactor"]},
+    "phase":         {"type": "string"},
     "skill":         {"type": "string"},
     "file_path":     {"type": "string"},
     "runner_output": {"type": "string"},

internal/exec/result_test.go

@@ -69,3 +69,11 @@ func TestResultValidation(t *testing.T) {
 		})
 	}
 }
+
+func TestValidateAcceptsAllPhases(t *testing.T) {
+	phases := []string{"red", "green", "refactor", "retrospective", "review", "debug", "spec", "trainer"}
+	for _, phase := range phases {
+		r := exec.Result{Status: "pass", Phase: phase, Skill: "test", ModelUsed: "self", Message: "ok"}
+		assert.NoError(t, r.Validate(), "phase %q should be valid", phase)
+	}
+}

internal/exec/verifier.go

@@ -0,0 +1,99 @@
+package exec
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"os"
+	"os/exec"
+	"time"
+)
+
+// Verdict is the output of a Claude verification call.
+type Verdict struct {
+	Accept   bool   `json:"accept"`
+	Feedback string `json:"feedback"` // empty when Accept is true
+}
+
+// Verifier runs a focused Claude call to judge local model output.
+type Verifier struct {
+	claudeBinary string
+	model        string
+	timeout      time.Duration
+}
+
+// NewVerifier creates a Verifier that calls claude with the given binary path and model.
+// Empty claudeBinary defaults to "claude". Zero timeout defaults to 30s.
+func NewVerifier(claudeBinary, model string, timeout time.Duration) *Verifier {
+	if claudeBinary == "" {
+		claudeBinary = "claude"
+	}
+	if timeout == 0 {
+		timeout = 30 * time.Second
+	}
+	return &Verifier{
+		claudeBinary: claudeBinary,
+		model:        model,
+		timeout:      timeout,
+	}
+}
+
+// Verify asks Claude whether output satisfies the skill discipline's iron laws.
+// Returns Verdict{Accept: true} to accept or Verdict{Accept: false, Feedback: "..."}
+// to escalate. Returns an error on subprocess failure or unparseable response.
+func (v *Verifier) Verify(ctx context.Context, skillPrompt, taskPrompt string, output Result) (Verdict, error) {
+	ctx, cancel := context.WithTimeout(ctx, v.timeout)
+	defer cancel()
+
+	outputJSON, err := json.Marshal(output)
+	if err != nil {
+		return Verdict{}, fmt.Errorf("verifier: marshal output: %w", err)
+	}
+
+	prompt := fmt.Sprintf(`You are a quality verifier for an AI supervisor system.
+
+Given the skill discipline, the original task, and the generated output, decide whether the output satisfies the discipline's iron laws and output contract.
+
+Reply with JSON only — no other text:
+{"accept": true, "feedback": ""}
+or
+{"accept": false, "feedback": "<one sentence reason>"}
+
+## Skill discipline
+%s
+
+## Original task
+%s
+
+## Generated output
+%s`, skillPrompt, taskPrompt, string(outputJSON))
+
+	args := []string{
+		"--print",
+		"--permission-mode", "bypassPermissions",
+	}
+	if v.model != "" {
+		args = append(args, "--model", v.model)
+	}
+	args = append(args, prompt)
+
+	cmd := exec.CommandContext(ctx, v.claudeBinary, args...)
+	cmd.Env = os.Environ()
+	var stdout, stderr bytes.Buffer
+	cmd.Stdout = &stdout
+	cmd.Stderr = &stderr
+
+	if err := cmd.Run(); err != nil {
+		if ctx.Err() != nil {
+			return Verdict{}, fmt.Errorf("verifier: timeout after %s", v.timeout)
+		}
+		return Verdict{}, fmt.Errorf("verifier: claude exited with error: %w — stderr: %s", err, stderr.String())
+	}
+
+	var verdict Verdict
+	if err := json.Unmarshal(bytes.TrimSpace(stdout.Bytes()), &verdict); err != nil {
+		return Verdict{}, fmt.Errorf("verifier: parse verdict JSON: %w — raw: %s", err, stdout.String())
+	}
+	return verdict, nil
+}

internal/exec/verifier_test.go

@@ -0,0 +1,74 @@
+package exec_test
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+
+	iexec "github.com/mathiasbq/supervisor/internal/exec"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func fakeVerifierClaude(t *testing.T, verdict iexec.Verdict) string {
+	t.Helper()
+	data, err := json.Marshal(verdict)
+	require.NoError(t, err)
+	dir := t.TempDir()
+	script := filepath.Join(dir, "claude")
+	content := fmt.Sprintf("#!/bin/sh\necho '%s'\n", string(data))
+	require.NoError(t, os.WriteFile(script, []byte(content), 0755))
+	return script
+}
+
+func TestVerifierAccepts(t *testing.T) {
+	claude := fakeVerifierClaude(t, iexec.Verdict{Accept: true, Feedback: ""})
+	v := iexec.NewVerifier(claude, "claude-sonnet-4-6", 5*time.Second)
+
+	verdict, err := v.Verify(context.Background(), "skill rules", "do the task", iexec.Result{
+		Status: "pass", Phase: "review", Skill: "review", Message: "ok",
+	})
+	require.NoError(t, err)
+	assert.True(t, verdict.Accept)
+	assert.Empty(t, verdict.Feedback)
+}
+
+func TestVerifierEscalates(t *testing.T) {
+	claude := fakeVerifierClaude(t, iexec.Verdict{Accept: false, Feedback: "missing line references"})
+	v := iexec.NewVerifier(claude, "claude-sonnet-4-6", 5*time.Second)
+
+	verdict, err := v.Verify(context.Background(), "skill rules", "do the task", iexec.Result{
+		Status: "pass", Phase: "review", Skill: "review", Message: "incomplete",
+	})
+	require.NoError(t, err)
+	assert.False(t, verdict.Accept)
+	assert.Equal(t, "missing line references", verdict.Feedback)
+}
+
+func TestVerifierErrorOnUnparsableOutput(t *testing.T) {
+	dir := t.TempDir()
+	script := filepath.Join(dir, "claude")
+	require.NoError(t, os.WriteFile(script, []byte("#!/bin/sh\necho 'not json'\n"), 0755))
+
+	v := iexec.NewVerifier(script, "claude-sonnet-4-6", 5*time.Second)
+	_, err := v.Verify(context.Background(), "rules", "task", iexec.Result{
+		Status: "pass", Phase: "review", Skill: "review", Message: "ok",
+	})
+	assert.Error(t, err)
+}
+
+func TestVerifierErrorOnNonZeroExit(t *testing.T) {
+	dir := t.TempDir()
+	script := filepath.Join(dir, "claude")
+	require.NoError(t, os.WriteFile(script, []byte("#!/bin/sh\nexit 1\n"), 0755))
+
+	v := iexec.NewVerifier(script, "claude-sonnet-4-6", 5*time.Second)
+	_, err := v.Verify(context.Background(), "rules", "task", iexec.Result{
+		Status: "pass", Phase: "review", Skill: "review", Message: "ok",
+	})
+	assert.Error(t, err)
+}

internal/session/history.go

@@ -0,0 +1,38 @@
+// internal/session/history.go
+package session
+
+import (
+	"fmt"
+	"strings"
+)
+
+// FormatHistory formats prior session entries as a structured block for
+// injection into a worker task prompt. Entries matching excludePhase are
+// omitted (pass the current phase to avoid circular injection).
+func FormatHistory(entries []Entry, excludePhase string) string {
+	var filtered []Entry
+	for _, e := range entries {
+		if e.Phase != excludePhase {
+			filtered = append(filtered, e)
+		}
+	}
+	if len(filtered) == 0 {
+		return ""
+	}
+
+	var b strings.Builder
+	b.WriteString("## Session history\n\n")
+	for _, e := range filtered {
+		fmt.Fprintf(&b, "### Phase: %s\n", e.Phase)      //nolint:errcheck // strings.Builder never errors
+		fmt.Fprintf(&b, "- Skill: %s\n", e.Skill)        //nolint:errcheck
+		fmt.Fprintf(&b, "- Status: %s\n", e.FinalStatus) //nolint:errcheck
+		if e.FilePath != "" {
+			fmt.Fprintf(&b, "- File: %s\n", e.FilePath) //nolint:errcheck
+		}
+		if e.Message != "" {
+			fmt.Fprintf(&b, "- Summary: %s\n", e.Message) //nolint:errcheck
+		}
+		b.WriteString("\n")
+	}
+	return b.String()
+}

internal/session/history_test.go

@@ -0,0 +1,41 @@
+// internal/session/history_test.go
+package session_test
+
+import (
+	"testing"
+	"time"
+
+	"github.com/mathiasbq/supervisor/internal/session"
+	"github.com/stretchr/testify/assert"
+)
+
+func TestFormatHistoryEmpty(t *testing.T) {
+	result := session.FormatHistory(nil, "")
+	assert.Equal(t, "", result)
+}
+
+func TestFormatHistoryFormatsEntries(t *testing.T) {
+	entries := []session.Entry{
+		{
+			Skill: "tdd", Phase: "red", FinalStatus: "pass",
+			FilePath: "internal/foo/foo_test.go",
+			Message:  "wrote failing test for Foo",
+			Timestamp: time.Now(),
+		},
+	}
+	result := session.FormatHistory(entries, "")
+	assert.Contains(t, result, "## Session history")
+	assert.Contains(t, result, "Phase: red")
+	assert.Contains(t, result, "wrote failing test for Foo")
+	assert.Contains(t, result, "internal/foo/foo_test.go")
+}
+
+func TestFormatHistoryExcludesCurrentPhase(t *testing.T) {
+	entries := []session.Entry{
+		{Skill: "tdd", Phase: "red", Message: "red done", FinalStatus: "pass"},
+		{Skill: "tdd", Phase: "green", Message: "green done", FinalStatus: "pass"},
+	}
+	result := session.FormatHistory(entries, "green")
+	assert.Contains(t, result, "red done")
+	assert.NotContains(t, result, "green done")
+}

internal/session/session.go

@@ -32,9 +32,14 @@ type Entry struct {
 type Attempt struct {
 	Attempt       int    `json:"attempt"`
 	Model         string `json:"model"`
+	Tier          string `json:"tier"`                    // local | subagent | managed
+	DurationMs    int64  `json:"duration_ms"`
+	WarmStart     bool   `json:"warm_start"`              // model already loaded in llama-swap
+	Verified      bool   `json:"verified"`
+	Verdict       string `json:"verdict,omitempty"`       // accept | escalate | error
+	Feedback      string `json:"feedback,omitempty"`      // verifier feedback on escalation
 	OutputSummary string `json:"output_summary,omitempty"`
 	RunnerOutput  string `json:"runner_output,omitempty"`
-	Verified      bool   `json:"verified"`
 }
 
 // Append writes entry as a single JSON line to sessionsDir/{sessionID}.jsonl.

internal/session/session_test.go

@@ -61,3 +61,22 @@ func TestRead_EmptyWhenNoFile(t *testing.T) {
 	require.NoError(t, err)
 	assert.Empty(t, entries)
 }
+
+func TestAttemptRoundTrip(t *testing.T) {
+	a := session.Attempt{
+		Attempt:    1,
+		Model:      "ollama/devstral",
+		Tier:       "local",
+		DurationMs: 4200,
+		WarmStart:  true,
+		Verified:   false,
+		Verdict:    "escalate",
+		Feedback:   "missing line references",
+	}
+	data, err := json.Marshal(a)
+	require.NoError(t, err)
+
+	var got session.Attempt
+	require.NoError(t, json.Unmarshal(data, &got))
+	assert.Equal(t, a, got)
+}

internal/skills/debug/handlers.go

@@ -0,0 +1,80 @@
+// internal/skills/debug/handlers.go
+package debug
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+
+	iexec "github.com/mathiasbq/supervisor/internal/exec"
+	"github.com/mathiasbq/supervisor/internal/session"
+)
+
+type debugArgs struct {
+	ProjectRoot string `json:"project_root"`
+	Error       string `json:"error"`
+	Context     string `json:"context"`
+	Model       string `json:"model"`
+	SessionID   string `json:"session_id"`
+}
+
+// Handle dispatches the MCP tool call to the appropriate handler.
+func (s *Skill) Handle(ctx context.Context, tool string, args json.RawMessage) (json.RawMessage, error) {
+	if tool != "debug" {
+		return nil, fmt.Errorf("unknown tool: %s", tool)
+	}
+	var a debugArgs
+	if err := json.Unmarshal(args, &a); err != nil {
+		return nil, fmt.Errorf("parse args: %w", err)
+	}
+	if a.ProjectRoot == "" {
+		return nil, fmt.Errorf("project_root is required")
+	}
+	if a.Error == "" {
+		return nil, fmt.Errorf("error is required")
+	}
+
+	model := a.Model
+	if model == "" {
+		model = s.cfg.DefaultModel
+	}
+
+	task := fmt.Sprintf(
+		"phase: debug\nproject_root: %s\nerror: %s\ncontext: %s\nmodel: %s",
+		a.ProjectRoot, a.Error, a.Context, model,
+	)
+	task = s.prependHistory(a.SessionID, "debug", task)
+
+	if s.cfg.ExecutorFn == nil {
+		return nil, fmt.Errorf("no executor configured")
+	}
+	result, err := s.cfg.ExecutorFn(ctx, iexec.Request{
+		SkillPrompt: s.cfg.SkillPrompt,
+		TaskPrompt:  task,
+		Model:       model,
+		Tools:       "Read,Bash",
+	})
+	if err != nil {
+		return nil, err
+	}
+	b, err := json.Marshal(result)
+	if err != nil {
+		return nil, fmt.Errorf("marshal result: %w", err)
+	}
+	return b, nil
+}
+
+func (s *Skill) prependHistory(sessionID, currentPhase, task string) string {
+	if sessionID == "" || s.cfg.SessionsDir == "" {
+		return task
+	}
+	entries, err := session.Read(s.cfg.SessionsDir, sessionID)
+	if err != nil || len(entries) == 0 {
+		return task
+	}
+	history := session.FormatHistory(entries, currentPhase)
+	if history == "" {
+		return task
+	}
+	return history + "\n---\n\n" + task
+}

internal/skills/debug/handlers_test.go

@@ -0,0 +1,61 @@
+// internal/skills/debug/handlers_test.go
+package debug_test
+
+import (
+	"context"
+	"encoding/json"
+	"testing"
+
+	iexec "github.com/mathiasbq/supervisor/internal/exec"
+	"github.com/mathiasbq/supervisor/internal/skills/debug"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestDebugToolRegistered(t *testing.T) {
+	sk := debug.New(debug.Config{SkillPrompt: "debug rules"})
+	names := make([]string, 0)
+	for _, tool := range sk.Tools() {
+		names = append(names, tool.Name)
+	}
+	assert.Contains(t, names, "debug")
+}
+
+func TestDebugRequiresProjectRoot(t *testing.T) {
+	sk := debug.New(debug.Config{SkillPrompt: "d"})
+	_, err := sk.Handle(context.Background(), "debug", json.RawMessage(`{"error":"panic: nil pointer"}`))
+	assert.ErrorContains(t, err, "project_root")
+}
+
+func TestDebugRequiresError(t *testing.T) {
+	sk := debug.New(debug.Config{SkillPrompt: "d"})
+	_, err := sk.Handle(context.Background(), "debug", json.RawMessage(`{"project_root":"/tmp"}`))
+	assert.ErrorContains(t, err, "error")
+}
+
+func TestDebugCallsExecutor(t *testing.T) {
+	called := false
+	var capturedTask string
+	fakeFn := func(_ context.Context, req iexec.Request) (iexec.Result, error) {
+		called = true
+		capturedTask = req.TaskPrompt
+		return iexec.Result{
+			Status: "pass", Phase: "debug", Skill: "debug",
+			RunnerOutput: "HYPOTHESIS 1 (likelihood: high): nil map access\nVERIFY: go test ./... → expected: panic line reference",
+			Verified:     false, ModelUsed: "self", Message: "3 hypotheses for: panic nil pointer at foo.go:42",
+		}, nil
+	}
+
+	sk := debug.New(debug.Config{SkillPrompt: "debug rules", ExecutorFn: fakeFn, SessionsDir: t.TempDir()})
+	out, err := sk.Handle(context.Background(), "debug", json.RawMessage(
+		`{"project_root":"/tmp/proj","error":"panic: nil pointer dereference at foo.go:42","context":"occurs on startup"}`,
+	))
+	require.NoError(t, err)
+	assert.True(t, called)
+	assert.Contains(t, capturedTask, "panic: nil pointer dereference")
+	assert.Contains(t, capturedTask, "occurs on startup")
+
+	var result iexec.Result
+	require.NoError(t, json.Unmarshal(out, &result))
+	assert.Equal(t, "debug", result.Phase)
+}

internal/skills/debug/skill.go

@@ -0,0 +1,55 @@
+// internal/skills/debug/skill.go
+package debug
+
+import (
+	"context"
+	"encoding/json"
+
+	iexec "github.com/mathiasbq/supervisor/internal/exec"
+	"github.com/mathiasbq/supervisor/internal/registry"
+)
+
+// ExecutorFn is the function signature for running a worker subprocess.
+type ExecutorFn func(ctx context.Context, req iexec.Request) (iexec.Result, error)
+
+// Config holds dependencies for the debug skill.
+type Config struct {
+	SkillPrompt  string
+	DefaultModel string
+	ExecutorFn   ExecutorFn
+	SessionsDir  string
+}
+
+// Skill implements the debug MCP tool.
+type Skill struct{ cfg Config }
+
+// New creates a new debug Skill.
+func New(cfg Config) *Skill { return &Skill{cfg: cfg} }
+
+// Name returns the skill identifier.
+func (s *Skill) Name() string { return "debug" }
+
+// Tools returns the MCP tool definitions for this skill.
+func (s *Skill) Tools() []registry.ToolDef {
+	schema := func(required []string, props map[string]any) json.RawMessage {
+		b, _ := json.Marshal(map[string]any{"type": "object", "required": required, "properties": props})
+		return b
+	}
+	str := map[string]any{"type": "string"}
+	return []registry.ToolDef{
+		{
+			Name:        "debug",
+			Description: "Analyse an error and return 3-5 hypotheses ordered by likelihood, each with a concrete verification step.",
+			InputSchema: schema(
+				[]string{"project_root", "error"},
+				map[string]any{
+					"project_root": str,
+					"error":        str,
+					"context":      str,
+					"model":        str,
+					"session_id":   str,
+				},
+			),
+		},
+	}
+}

internal/skills/review/handlers.go

@@ -0,0 +1,81 @@
+// internal/skills/review/handlers.go
+package review
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"strings"
+
+	iexec "github.com/mathiasbq/supervisor/internal/exec"
+	"github.com/mathiasbq/supervisor/internal/session"
+)
+
+type reviewArgs struct {
+	ProjectRoot string   `json:"project_root"`
+	Files       []string `json:"files"`
+	Context     string   `json:"context"`
+	Model       string   `json:"model"`
+	SessionID   string   `json:"session_id"`
+}
+
+// Handle dispatches the MCP tool call to the appropriate handler.
+func (s *Skill) Handle(ctx context.Context, tool string, args json.RawMessage) (json.RawMessage, error) {
+	if tool != "review" {
+		return nil, fmt.Errorf("unknown tool: %s", tool)
+	}
+	var a reviewArgs
+	if err := json.Unmarshal(args, &a); err != nil {
+		return nil, fmt.Errorf("parse args: %w", err)
+	}
+	if a.ProjectRoot == "" {
+		return nil, fmt.Errorf("project_root is required")
+	}
+	if len(a.Files) == 0 {
+		return nil, fmt.Errorf("files is required")
+	}
+
+	model := a.Model
+	if model == "" {
+		model = s.cfg.DefaultModel
+	}
+
+	task := fmt.Sprintf(
+		"phase: review\nproject_root: %s\nfiles: %s\ncontext: %s\nmodel: %s",
+		a.ProjectRoot, strings.Join(a.Files, ", "), a.Context, model,
+	)
+	task = s.prependHistory(a.SessionID, "review", task)
+
+	if s.cfg.ExecutorFn == nil {
+		return nil, fmt.Errorf("no executor configured")
+	}
+	result, err := s.cfg.ExecutorFn(ctx, iexec.Request{
+		SkillPrompt: s.cfg.SkillPrompt,
+		TaskPrompt:  task,
+		Model:       model,
+		Tools:       "Read,Bash",
+	})
+	if err != nil {
+		return nil, err
+	}
+	b, err := json.Marshal(result)
+	if err != nil {
+		return nil, fmt.Errorf("marshal result: %w", err)
+	}
+	return b, nil
+}
+
+func (s *Skill) prependHistory(sessionID, currentPhase, task string) string {
+	if sessionID == "" || s.cfg.SessionsDir == "" {
+		return task
+	}
+	entries, err := session.Read(s.cfg.SessionsDir, sessionID)
+	if err != nil || len(entries) == 0 {
+		return task
+	}
+	history := session.FormatHistory(entries, currentPhase)
+	if history == "" {
+		return task
+	}
+	return history + "\n---\n\n" + task
+}

internal/skills/review/handlers_test.go

@@ -0,0 +1,61 @@
+// internal/skills/review/handlers_test.go
+package review_test
+
+import (
+	"context"
+	"encoding/json"
+	"testing"
+
+	iexec "github.com/mathiasbq/supervisor/internal/exec"
+	"github.com/mathiasbq/supervisor/internal/skills/review"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestReviewToolRegistered(t *testing.T) {
+	sk := review.New(review.Config{SkillPrompt: "review rules"})
+	names := make([]string, 0)
+	for _, tool := range sk.Tools() {
+		names = append(names, tool.Name)
+	}
+	assert.Contains(t, names, "review")
+}
+
+func TestReviewRequiresProjectRoot(t *testing.T) {
+	sk := review.New(review.Config{SkillPrompt: "r"})
+	_, err := sk.Handle(context.Background(), "review", json.RawMessage(`{"files":["main.go"]}`))
+	assert.ErrorContains(t, err, "project_root")
+}
+
+func TestReviewRequiresFiles(t *testing.T) {
+	sk := review.New(review.Config{SkillPrompt: "r"})
+	_, err := sk.Handle(context.Background(), "review", json.RawMessage(`{"project_root":"/tmp"}`))
+	assert.ErrorContains(t, err, "files")
+}
+
+func TestReviewCallsExecutor(t *testing.T) {
+	called := false
+	var capturedTask string
+	fakeFn := func(_ context.Context, req iexec.Request) (iexec.Result, error) {
+		called = true
+		capturedTask = req.TaskPrompt
+		return iexec.Result{
+			Status: "pass", Phase: "review", Skill: "review",
+			Verified: true, ModelUsed: "self", Message: "2 warnings found",
+		}, nil
+	}
+
+	sk := review.New(review.Config{SkillPrompt: "review rules", ExecutorFn: fakeFn, SessionsDir: t.TempDir()})
+	out, err := sk.Handle(context.Background(), "review", json.RawMessage(
+		`{"project_root":"/tmp/proj","files":["internal/foo/foo.go"],"context":"PR: add Foo helper"}`,
+	))
+	require.NoError(t, err)
+	assert.True(t, called)
+	assert.Contains(t, capturedTask, "internal/foo/foo.go")
+	assert.Contains(t, capturedTask, "PR: add Foo helper")
+
+	var result iexec.Result
+	require.NoError(t, json.Unmarshal(out, &result))
+	assert.Equal(t, "pass", result.Status)
+	assert.Equal(t, "review", result.Phase)
+}

internal/skills/review/skill.go

@@ -0,0 +1,55 @@
+// internal/skills/review/skill.go
+package review
+
+import (
+	"context"
+	"encoding/json"
+
+	iexec "github.com/mathiasbq/supervisor/internal/exec"
+	"github.com/mathiasbq/supervisor/internal/registry"
+)
+
+// ExecutorFn is the function signature for running a worker subprocess.
+type ExecutorFn func(ctx context.Context, req iexec.Request) (iexec.Result, error)
+
+// Config holds dependencies for the review skill.
+type Config struct {
+	SkillPrompt  string
+	DefaultModel string
+	ExecutorFn   ExecutorFn
+	SessionsDir  string
+}
+
+// Skill implements the review MCP tool.
+type Skill struct{ cfg Config }
+
+// New creates a new review Skill.
+func New(cfg Config) *Skill { return &Skill{cfg: cfg} }
+
+// Name returns the skill identifier.
+func (s *Skill) Name() string { return "review" }
+
+// Tools returns the MCP tool definitions for this skill.
+func (s *Skill) Tools() []registry.ToolDef {
+	schema := func(required []string, props map[string]any) json.RawMessage {
+		b, _ := json.Marshal(map[string]any{"type": "object", "required": required, "properties": props})
+		return b
+	}
+	str := map[string]any{"type": "string"}
+	return []registry.ToolDef{
+		{
+			Name:        "review",
+			Description: "Perform a structured code review of the specified files. Returns findings with severity levels.",
+			InputSchema: schema(
+				[]string{"project_root", "files"},
+				map[string]any{
+					"project_root": str,
+					"files":        map[string]any{"type": "array", "items": map[string]any{"type": "string"}},
+					"context":      str,
+					"model":        str,
+					"session_id":   str,
+				},
+			),
+		},
+	}
+}

internal/skills/spec/handlers.go

@@ -0,0 +1,85 @@
+// internal/skills/spec/handlers.go
+package spec
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+
+	iexec "github.com/mathiasbq/supervisor/internal/exec"
+	"github.com/mathiasbq/supervisor/internal/session"
+)
+
+type specArgs struct {
+	ProjectRoot  string `json:"project_root"`
+	Requirements string `json:"requirements"`
+	OutputPath   string `json:"output_path"`
+	Context      string `json:"context"`
+	Model        string `json:"model"`
+	SessionID    string `json:"session_id"`
+}
+
+// Handle dispatches the MCP tool call to the appropriate handler.
+func (s *Skill) Handle(ctx context.Context, tool string, args json.RawMessage) (json.RawMessage, error) {
+	if tool != "spec" {
+		return nil, fmt.Errorf("unknown tool: %s", tool)
+	}
+	var a specArgs
+	if err := json.Unmarshal(args, &a); err != nil {
+		return nil, fmt.Errorf("parse args: %w", err)
+	}
+	if a.ProjectRoot == "" {
+		return nil, fmt.Errorf("project_root is required")
+	}
+	if a.Requirements == "" {
+		return nil, fmt.Errorf("requirements is required")
+	}
+	outputPath := a.OutputPath
+	if outputPath == "" {
+		outputPath = "docs/spec.md"
+	}
+
+	model := a.Model
+	if model == "" {
+		model = s.cfg.DefaultModel
+	}
+
+	task := fmt.Sprintf(
+		"phase: spec\nproject_root: %s\nrequirements: %s\noutput_path: %s\ncontext: %s\nmodel: %s",
+		a.ProjectRoot, a.Requirements, outputPath, a.Context, model,
+	)
+	task = s.prependHistory(a.SessionID, "spec", task)
+
+	if s.cfg.ExecutorFn == nil {
+		return nil, fmt.Errorf("no executor configured")
+	}
+	result, err := s.cfg.ExecutorFn(ctx, iexec.Request{
+		SkillPrompt: s.cfg.SkillPrompt,
+		TaskPrompt:  task,
+		Model:       model,
+		Tools:       "Read,Write",
+	})
+	if err != nil {
+		return nil, err
+	}
+	b, err := json.Marshal(result)
+	if err != nil {
+		return nil, fmt.Errorf("marshal result: %w", err)
+	}
+	return b, nil
+}
+
+func (s *Skill) prependHistory(sessionID, currentPhase, task string) string {
+	if sessionID == "" || s.cfg.SessionsDir == "" {
+		return task
+	}
+	entries, err := session.Read(s.cfg.SessionsDir, sessionID)
+	if err != nil || len(entries) == 0 {
+		return task
+	}
+	history := session.FormatHistory(entries, currentPhase)
+	if history == "" {
+		return task
+	}
+	return history + "\n---\n\n" + task
+}

internal/skills/spec/handlers_test.go

@@ -0,0 +1,61 @@
+// internal/skills/spec/handlers_test.go
+package spec_test
+
+import (
+	"context"
+	"encoding/json"
+	"testing"
+
+	iexec "github.com/mathiasbq/supervisor/internal/exec"
+	"github.com/mathiasbq/supervisor/internal/skills/spec"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestSpecToolRegistered(t *testing.T) {
+	sk := spec.New(spec.Config{SkillPrompt: "spec rules"})
+	names := make([]string, 0)
+	for _, tool := range sk.Tools() {
+		names = append(names, tool.Name)
+	}
+	assert.Contains(t, names, "spec")
+}
+
+func TestSpecRequiresProjectRoot(t *testing.T) {
+	sk := spec.New(spec.Config{SkillPrompt: "s"})
+	_, err := sk.Handle(context.Background(), "spec", json.RawMessage(`{"requirements":"add login"}`))
+	assert.ErrorContains(t, err, "project_root")
+}
+
+func TestSpecRequiresRequirements(t *testing.T) {
+	sk := spec.New(spec.Config{SkillPrompt: "s"})
+	_, err := sk.Handle(context.Background(), "spec", json.RawMessage(`{"project_root":"/tmp"}`))
+	assert.ErrorContains(t, err, "requirements")
+}
+
+func TestSpecCallsExecutor(t *testing.T) {
+	called := false
+	var capturedTask string
+	fakeFn := func(_ context.Context, req iexec.Request) (iexec.Result, error) {
+		called = true
+		capturedTask = req.TaskPrompt
+		return iexec.Result{
+			Status: "pass", Phase: "spec", Skill: "spec",
+			FilePath: "/tmp/proj/docs/login-spec.md",
+			Verified: true, ModelUsed: "self", Message: "spec written: login feature",
+		}, nil
+	}
+
+	sk := spec.New(spec.Config{SkillPrompt: "spec rules", ExecutorFn: fakeFn, SessionsDir: t.TempDir()})
+	out, err := sk.Handle(context.Background(), "spec", json.RawMessage(
+		`{"project_root":"/tmp/proj","requirements":"add OAuth2 login","output_path":"docs/login-spec.md"}`,
+	))
+	require.NoError(t, err)
+	assert.True(t, called)
+	assert.Contains(t, capturedTask, "OAuth2 login")
+	assert.Contains(t, capturedTask, "docs/login-spec.md")
+
+	var result iexec.Result
+	require.NoError(t, json.Unmarshal(out, &result))
+	assert.Equal(t, "spec", result.Phase)
+}

internal/skills/spec/skill.go

@@ -0,0 +1,56 @@
+// internal/skills/spec/skill.go
+package spec
+
+import (
+	"context"
+	"encoding/json"
+
+	iexec "github.com/mathiasbq/supervisor/internal/exec"
+	"github.com/mathiasbq/supervisor/internal/registry"
+)
+
+// ExecutorFn is the function signature for running a worker subprocess.
+type ExecutorFn func(ctx context.Context, req iexec.Request) (iexec.Result, error)
+
+// Config holds dependencies for the spec skill.
+type Config struct {
+	SkillPrompt  string
+	DefaultModel string
+	ExecutorFn   ExecutorFn
+	SessionsDir  string
+}
+
+// Skill implements the spec MCP tool.
+type Skill struct{ cfg Config }
+
+// New creates a new spec Skill.
+func New(cfg Config) *Skill { return &Skill{cfg: cfg} }
+
+// Name returns the skill identifier.
+func (s *Skill) Name() string { return "spec" }
+
+// Tools returns the MCP tool definitions for this skill.
+func (s *Skill) Tools() []registry.ToolDef {
+	schema := func(required []string, props map[string]any) json.RawMessage {
+		b, _ := json.Marshal(map[string]any{"type": "object", "required": required, "properties": props})
+		return b
+	}
+	str := map[string]any{"type": "string"}
+	return []registry.ToolDef{
+		{
+			Name:        "spec",
+			Description: "Generate a structured implementation spec from requirements. Writes the spec to output_path in the project.",
+			InputSchema: schema(
+				[]string{"project_root", "requirements"},
+				map[string]any{
+					"project_root": str,
+					"requirements": str,
+					"output_path":  str,
+					"context":      str,
+					"model":        str,
+					"session_id":   str,
+				},
+			),
+		},
+	}
+}

internal/skills/tdd/handlers.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 
 	iexec "github.com/mathiasbq/supervisor/internal/exec"
+	"github.com/mathiasbq/supervisor/internal/session"
 )
 
 func (s *Skill) Handle(ctx context.Context, tool string, args json.RawMessage) (json.RawMessage, error) {
@@ -51,6 +52,7 @@ type greenArgs struct {
 	TestPath    string `json:"test_path"`
 	Model       string `json:"model"`
 	TestCmd     string `json:"test_cmd"`
+	SessionID   string `json:"session_id"`
 }
 
 func (s *Skill) handleGreen(ctx context.Context, raw json.RawMessage) (json.RawMessage, error) {
@@ -68,6 +70,7 @@ func (s *Skill) handleGreen(ctx context.Context, raw json.RawMessage) (json.RawM
 		"phase: green\nproject_root: %s\ntest_path: %s\nmodel: %s\ntest_cmd: %s",
 		args.ProjectRoot, args.TestPath, s.resolveModel(args.Model), args.TestCmd,
 	)
+	task = s.prependHistory(args.SessionID, "green", task)
 	return s.execute(ctx, task)
 }
 
@@ -77,6 +80,7 @@ type refactorArgs struct {
 	ImplPath    string `json:"impl_path"`
 	Model       string `json:"model"`
 	TestCmd     string `json:"test_cmd"`
+	SessionID   string `json:"session_id"`
 }
 
 func (s *Skill) handleRefactor(ctx context.Context, raw json.RawMessage) (json.RawMessage, error) {
@@ -97,9 +101,25 @@ func (s *Skill) handleRefactor(ctx context.Context, raw json.RawMessage) (json.R
 		"phase: refactor\nproject_root: %s\ntest_path: %s\nimpl_path: %s\nmodel: %s\ntest_cmd: %s",
 		args.ProjectRoot, args.TestPath, args.ImplPath, s.resolveModel(args.Model), args.TestCmd,
 	)
+	task = s.prependHistory(args.SessionID, "refactor", task)
 	return s.execute(ctx, task)
 }
 
+func (s *Skill) prependHistory(sessionID, currentPhase, task string) string {
+	if sessionID == "" || s.cfg.SessionsDir == "" {
+		return task
+	}
+	entries, err := session.Read(s.cfg.SessionsDir, sessionID)
+	if err != nil || len(entries) == 0 {
+		return task
+	}
+	history := session.FormatHistory(entries, currentPhase)
+	if history == "" {
+		return task
+	}
+	return history + "\n---\n\n" + task
+}
+
 func (s *Skill) resolveModel(override string) string {
 	if override != "" {
 		return override

internal/skills/tdd/handlers_test.go

@@ -5,6 +5,8 @@ import (
 	"encoding/json"
 	"testing"
 
+	iexec "github.com/mathiasbq/supervisor/internal/exec"
+	"github.com/mathiasbq/supervisor/internal/session"
 	"github.com/mathiasbq/supervisor/internal/skills/tdd"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
@@ -41,5 +43,43 @@ func TestTDDRedRequiresSpec(t *testing.T) {
 	assert.ErrorContains(t, err, "spec")
 }
 
+func TestTDDGreenInjectsSessionHistory(t *testing.T) {
+	sessDir := t.TempDir()
+	require.NoError(t, session.Append(sessDir, "sess-1", session.Entry{
+		SessionID: "sess-1", Skill: "tdd", Phase: "red", FinalStatus: "pass",
+		FilePath: "internal/foo/foo_test.go",
+		Message:  "wrote failing test for Foo",
+	}))
+
+	var capturedPrompt string
+	fakeFn := func(_ context.Context, req iexec.Request) (iexec.Result, error) {
+		capturedPrompt = req.TaskPrompt
+		return iexec.Result{Status: "pass", Phase: "green", Skill: "tdd", Verified: true, ModelUsed: "self", Message: "ok"}, nil
+	}
+
+	sk := tdd.New(tdd.Config{SkillPrompt: "tdd", ExecutorFn: fakeFn, SessionsDir: sessDir})
+	_, err := sk.Handle(context.Background(), "tdd_green", json.RawMessage(
+		`{"project_root":"/tmp","test_path":"internal/foo/foo_test.go","test_cmd":"go test ./...","session_id":"sess-1"}`,
+	))
+	require.NoError(t, err)
+	assert.Contains(t, capturedPrompt, "## Session history")
+	assert.Contains(t, capturedPrompt, "wrote failing test for Foo")
+}
+
+func TestTDDGreenNoHistoryWhenSessionIDEmpty(t *testing.T) {
+	var capturedPrompt string
+	fakeFn := func(_ context.Context, req iexec.Request) (iexec.Result, error) {
+		capturedPrompt = req.TaskPrompt
+		return iexec.Result{Status: "pass", Phase: "green", Skill: "tdd", Verified: true, ModelUsed: "self", Message: "ok"}, nil
+	}
+
+	sk := tdd.New(tdd.Config{SkillPrompt: "tdd", ExecutorFn: fakeFn, SessionsDir: t.TempDir()})
+	_, err := sk.Handle(context.Background(), "tdd_green", json.RawMessage(
+		`{"project_root":"/tmp","test_path":"internal/foo/foo_test.go"}`,
+	))
+	require.NoError(t, err)
+	assert.NotContains(t, capturedPrompt, "## Session history")
+}
+
 // Ensure require is used (avoids import error).
 var _ = require.New

internal/skills/tdd/skill.go

@@ -16,6 +16,7 @@ type Config struct {
 	SkillPrompt  string
 	ExecutorFn   ExecutorFn // nil = no executor (tests that don't reach execute())
 	DefaultModel string
+	SessionsDir  string // optional: path to brain/sessions/ for history injection
 }
 
 type Skill struct {
@@ -63,6 +64,7 @@ func (s *Skill) Tools() []registry.ToolDef {
 					"test_path":    strProp,
 					"model":        strProp,
 					"test_cmd":     strProp,
+					"session_id":   strProp,
 				},
 			),
 		},
@@ -77,6 +79,7 @@ func (s *Skill) Tools() []registry.ToolDef {
 					"impl_path":    strProp,
 					"model":        strProp,
 					"test_cmd":     strProp,
+					"session_id":   strProp,
 				},
 			),
 		},

internal/skills/trainer/handlers.go

@@ -0,0 +1,80 @@
+// internal/skills/trainer/handlers.go
+package trainer
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+
+	iexec "github.com/mathiasbq/supervisor/internal/exec"
+	"github.com/mathiasbq/supervisor/internal/session"
+)
+
+type trainArgs struct {
+	SessionID string `json:"session_id"`
+	Model     string `json:"model"`
+}
+
+// Handle dispatches the MCP tool call to the trainer handler.
+func (s *Skill) Handle(ctx context.Context, tool string, args json.RawMessage) (json.RawMessage, error) {
+	if tool != "trainer" {
+		return nil, fmt.Errorf("unknown tool: %s", tool)
+	}
+	var a trainArgs
+	if err := json.Unmarshal(args, &a); err != nil {
+		return nil, fmt.Errorf("parse args: %w", err)
+	}
+	if a.SessionID == "" {
+		return nil, fmt.Errorf("session_id is required")
+	}
+	if s.cfg.ExecutorFn == nil {
+		return nil, fmt.Errorf("no executor configured")
+	}
+
+	model := a.Model
+	if model == "" {
+		model = s.cfg.DefaultModel
+	}
+
+	entries, err := session.Read(s.cfg.SessionsDir, a.SessionID)
+	if err != nil {
+		return nil, fmt.Errorf("read session log: %w", err)
+	}
+
+	// ── Step 1: Reader agent ─────────────────────────────────────────────────
+	history := session.FormatHistory(entries, "")
+	readerTask := fmt.Sprintf(
+		"role: reader\nsession_id: %s\nbrain_dir: %s\n\n%s",
+		a.SessionID, s.cfg.BrainDir, history,
+	)
+	readerResult, err := s.cfg.ExecutorFn(ctx, iexec.Request{
+		SkillPrompt: s.cfg.ReaderPrompt,
+		TaskPrompt:  readerTask,
+		Model:       model,
+		Tools:       "Read",
+	})
+	if err != nil {
+		return nil, fmt.Errorf("reader agent: %w", err)
+	}
+
+	// ── Step 2: Writer agent (receives reader candidates) ────────────────────
+	writerTask := fmt.Sprintf(
+		"role: writer\nsession_id: %s\nbrain_dir: %s\n\nreader_summary: %s\nreader_candidates:\n%s",
+		a.SessionID, s.cfg.BrainDir, readerResult.Message, readerResult.RunnerOutput,
+	)
+	writerResult, err := s.cfg.ExecutorFn(ctx, iexec.Request{
+		SkillPrompt: s.cfg.WriterPrompt,
+		TaskPrompt:  writerTask,
+		Model:       model,
+		Tools:       "Read,Write",
+	})
+	if err != nil {
+		return nil, fmt.Errorf("writer agent: %w", err)
+	}
+
+	b, err := json.Marshal(writerResult)
+	if err != nil {
+		return nil, fmt.Errorf("marshal result: %w", err)
+	}
+	return b, nil
+}

internal/skills/trainer/handlers_test.go

@@ -0,0 +1,82 @@
+// internal/skills/trainer/handlers_test.go
+package trainer_test
+
+import (
+	"context"
+	"encoding/json"
+	"testing"
+
+	iexec "github.com/mathiasbq/supervisor/internal/exec"
+	"github.com/mathiasbq/supervisor/internal/session"
+	"github.com/mathiasbq/supervisor/internal/skills/trainer"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestTrainerToolRegistered(t *testing.T) {
+	sk := trainer.New(trainer.Config{ReaderPrompt: "r", WriterPrompt: "w"})
+	names := make([]string, 0)
+	for _, tool := range sk.Tools() {
+		names = append(names, tool.Name)
+	}
+	assert.Contains(t, names, "trainer")
+}
+
+func TestTrainerRequiresSessionID(t *testing.T) {
+	sk := trainer.New(trainer.Config{ReaderPrompt: "r", WriterPrompt: "w"})
+	_, err := sk.Handle(context.Background(), "trainer", json.RawMessage(`{}`))
+	assert.ErrorContains(t, err, "session_id")
+}
+
+func TestTrainerCallsReaderThenWriter(t *testing.T) {
+	sessDir := t.TempDir()
+	require.NoError(t, session.Append(sessDir, "sess-1", session.Entry{
+		SessionID: "sess-1", Skill: "tdd", Phase: "red", FinalStatus: "pass",
+		Message: "wrote failing test", FilePath: "internal/foo/foo_test.go",
+	}))
+
+	callCount := 0
+	var readerTask, writerTask string
+
+	fakeFn := func(_ context.Context, req iexec.Request) (iexec.Result, error) {
+		callCount++
+		if callCount == 1 {
+			// reader call
+			readerTask = req.TaskPrompt
+			return iexec.Result{
+				Status: "pass", Phase: "trainer", Skill: "trainer",
+				RunnerOutput: `[{"type":"sft","moment":"first-pass clean TDD","score":4}]`,
+				Verified: true, ModelUsed: "self", Message: "1 sft candidate found",
+			}, nil
+		}
+		// writer call
+		writerTask = req.TaskPrompt
+		return iexec.Result{
+			Status: "pass", Phase: "trainer", Skill: "trainer",
+			FilePath: sessDir + "/training-data/sft/sess-1.jsonl",
+			Verified: true, ModelUsed: "self", Message: "1 sft pair written",
+		}, nil
+	}
+
+	sk := trainer.New(trainer.Config{
+		ReaderPrompt: "reader rules",
+		WriterPrompt: "writer rules",
+		ExecutorFn:   fakeFn,
+		SessionsDir:  sessDir,
+		BrainDir:     t.TempDir(),
+	})
+	out, err := sk.Handle(context.Background(), "trainer", json.RawMessage(`{"session_id":"sess-1"}`))
+	require.NoError(t, err)
+
+	assert.Equal(t, 2, callCount, "executor must be called exactly twice: reader then writer")
+	assert.Contains(t, readerTask, "role: reader")
+	assert.Contains(t, readerTask, "sess-1")
+	assert.Contains(t, readerTask, "wrote failing test") // session history in reader prompt
+	assert.Contains(t, writerTask, "role: writer")
+	assert.Contains(t, writerTask, "sft candidate") // reader output passed to writer
+
+	var result iexec.Result
+	require.NoError(t, json.Unmarshal(out, &result))
+	assert.Equal(t, "trainer", result.Phase)
+	assert.Equal(t, "pass", result.Status)
+}

internal/skills/trainer/skill.go

@@ -0,0 +1,53 @@
+// internal/skills/trainer/skill.go
+package trainer
+
+import (
+	"context"
+	"encoding/json"
+
+	iexec "github.com/mathiasbq/supervisor/internal/exec"
+	"github.com/mathiasbq/supervisor/internal/registry"
+)
+
+// ExecutorFn is the function signature for running a worker subprocess.
+type ExecutorFn func(ctx context.Context, req iexec.Request) (iexec.Result, error)
+
+// Config holds dependencies for the trainer skill.
+type Config struct {
+	ReaderPrompt string
+	WriterPrompt string
+	DefaultModel string
+	ExecutorFn   ExecutorFn
+	SessionsDir  string
+	BrainDir     string // root of brain/ directory; writer writes to BrainDir/training-data/
+}
+
+// Skill implements the trainer MCP tool.
+type Skill struct{ cfg Config }
+
+// New creates a new trainer Skill.
+func New(cfg Config) *Skill { return &Skill{cfg: cfg} }
+
+// Name returns the skill identifier.
+func (s *Skill) Name() string { return "trainer" }
+
+// Tools returns the MCP tool definitions for this skill.
+func (s *Skill) Tools() []registry.ToolDef {
+	schema := func(required []string, props map[string]any) json.RawMessage {
+		b, _ := json.Marshal(map[string]any{"type": "object", "required": required, "properties": props})
+		return b
+	}
+	return []registry.ToolDef{
+		{
+			Name:        "trainer",
+			Description: "Extract SFT and DPO training pairs from a session log. Runs a reader→writer chain: reader identifies learning moments, writer formats and writes pairs to brain/training-data/.",
+			InputSchema: schema(
+				[]string{"session_id"},
+				map[string]any{
+					"session_id": map[string]any{"type": "string"},
+					"model":      map[string]any{"type": "string"},
+				},
+			),
+		},
+	}
+}