Workflow Review Changes Loop

---
name: workflow-review-changes-loop
description: '[Workflow] Use when running /workflow-review-changes repeatedly until a complete pass applies zero fixes.'
---

> Codex compatibility note:
> - Invoke repository skills with `$skill-name` in Codex; this mirrored copy rewrites legacy Claude `/skill-name` references.
> - Task tracker mandate: BEFORE executing any workflow or skill step, create/update task tracking for all steps and keep it synchronized as progress changes.
> - User-question prompts mean to ask the user directly in Codex.
> - Ignore Claude-specific mode-switch instructions when they appear.
> - Strict execution contract: when a user explicitly invokes a skill, execute that skill protocol as written.
> - Subagent authorization: when a skill is user-invoked or AI-detected and its protocol requires subagents, that skill activation authorizes use of the required `spawn_agent` subagent(s) for that task.
> - Do not skip, reorder, or merge protocol steps unless the user explicitly approves the deviation first.
> - For workflow skills, execute each listed child-skill step explicitly and report step-by-step evidence.
> - If a required step/tool cannot run in this environment, stop and ask the user before adapting.
<!-- CODEX:PROJECT-REFERENCE-LOADING:START -->
## Codex Project-Reference Loading (Hook-Independent)

Claude and Codex use static project-reference loading as the authority; hooks may accelerate discovery but never replace the explicit read.
When coding, planning, debugging, testing, or reviewing, open project docs explicitly using this routing.

**Always read:**
- `docs/project-config.json` (project-specific paths, commands, modules, and workflow/test settings)
- `docs/project-reference/docs-index-reference.md` (routes to the full `docs/project-reference/*` catalog)
- `docs/project-reference/lessons.md` (always-on guardrails and anti-patterns)

**Missing/stale context route:** If `docs/project-config.json`, the docs index, `lessons.md`, `CLAUDE.md`, `AGENTS.md`, or any task-required reference doc is missing or stale, auto-run `$project-init` or the narrow setup route (`$project-config`, `$docs-init`, `$scan-all`, `$scan --target=<key>`, `$claude-md-init`) before ordinary project-specific work. If Codex mirrors or `AGENTS.md` are missing/stale, ask the user to run `$sync-codex`; do not auto-run it.

**Situation-based docs:**
- Project structure/architecture/tech-stack/deployment/setup (any layer — backend, frontend, or infra): `project-structure-reference.md`
- Backend/CQRS/API/domain/entity changes: `backend-patterns-reference.md`, `domain-entities-reference.md`
- Frontend/UI/styling/design-system: `frontend-patterns-reference.md`, `scss-styling-guide.md`, `design-system/README.md`
- Spec authoring, `docs/specs/` pathing, or TC format: `feature-spec-reference.md`, `spec-system-reference.md`, `spec-principles.md`
- Behavior/public-contract changes or spec-test-code sync: `workflow-spec-test-code-cycle-reference.md` plus the spec docs above
- Derived spec indexes/ERDs/reimplementation guides: `spec-system-reference.md` and source Feature Specs under `docs/specs/`
- Integration test implementation/review: `integration-test-reference.md`
- E2E test implementation/review: `e2e-test-reference.md`
- Code review/audit work: `code-review-rules.md` plus domain docs above based on changed files

Do not read all docs blindly. Start from `docs-index-reference.md`, then open only relevant files for the task.
<!-- CODEX:PROJECT-REFERENCE-LOADING:END -->

<!-- PROMPT-ENHANCE:STEP-TASK-ANCHOR:START -->

> **[BLOCKING]** Execute skill steps in declared order. NEVER skip, reorder, or merge steps without explicit user approval.
> **[BLOCKING]** Before each step or sub-skill call, update task tracking: set `in_progress` when step starts, set `completed` when step ends.
> **[BLOCKING]** Every completed/skipped step MUST include brief evidence or explicit skip reason.
> **[BLOCKING]** If Task tools are unavailable, create and maintain an equivalent step-by-step plan tracker with the same status transitions.

<!-- PROMPT-ENHANCE:STEP-TASK-ANCHOR:END -->

## Quick Summary

**Goal:** Converge a review scope to a **clean no-op pass** by re-running the ENTIRE `$workflow-review-changes` workflow INLINE, round after round, over a fixed scope combined with the fixes accumulated so far — stopping when a complete round applies **zero fixes**. **From round 2 the severity floor applies: LOW findings are no longer fixed, so a round that surfaces only LOW findings is a zero-fix round and ENDS the loop.**

**Summary:**

- **Steps (in order):** (0) resolve scope + Goal Contract → (0b) bind the convergence loop (protocol loop primary + optional `/goal` accelerator) → (1) round loop { run `$workflow-review-changes` INLINE → detect fixes-applied → log iteration } → (2) converge on a zero-fix round OR escalate on non-progress → (3) recap.
- **Convergence:** stop ONLY when a whole round applies **zero fixes** (its fix cycle skipped, working tree unchanged, reviews clean) — not merely one clean review.
- **Inline invariant:** run `$workflow-review-changes` via the skill invocation, NEVER the `spawn_agent` tool — it self-binds its own review-loop obligation (owning the session Stop hook for its `/goal` gate WHEN available), which a sub-agent cannot own or carry back to this loop.
- **Severity floor — from round 2, LOW stops blocking.** Round 1 fix every validated severity. **From round 2 tell the inner `$workflow-review-changes` to fix only CRITICAL/HIGH/MEDIUM and to defer LOW findings** — a round whose validated findings are ALL LOW therefore applies zero fixes and CONVERGES. Carry every deferred LOW into the recap and Goal Contract; NEVER re-tier a real CRITICAL/HIGH/MEDIUM down to LOW to reach the exit. Severity tiers per `SYNC:severity-rubric`.
- **Bounded:** round cap default 3; blocking findings not shrinking across 2 rounds, or cap hit with CRITICAL/HIGH/MEDIUM fixes still landing → **STOP & escalate** by asking the user directly.

**Why this skill exists (READ FIRST — it is the whole justification):** `$workflow-review-changes` already converges *internally* to a clean pass, BUT its inner loop does **NOT** re-run the 7 specialist reviewers from scratch. Per the `Conditional Inline Re-Review Protocol` in `.claude/skills/workflow-review-changes/SKILL.md`, the step-15 inline re-review re-runs only `$changes-review`'s own BE/FE/SCSS/UI dimensions plus *scoped* re-runs of the specific specialist that raised a finding — `$architecture-review`, `$performance-review`, `$security-review`, `$integration-test-review`, `$production-readiness-review`, `$domain-entities-review`, and `$ui-review` fire ONCE (steps 4–10). Only a **fresh full re-invocation** re-runs ALL specialists over the now-fixed code — catching **second-order defects the fixes themselves introduced** and killing whole-workflow confirmation bias. Without this outer loop those regressions ship unreviewed.

**Workflow:** resolve scope + Goal Contract → bind the convergence loop (protocol loop + optional `/goal` accelerator) → **round loop** { run `$workflow-review-changes` INLINE → detect fixes-applied → log iteration } → converge when a round applies zero fixes → recap.

**Key Rules:**

- **MUST run INLINE in the main session — NEVER dispatch `$workflow-review-changes` as a sub-agent.** It self-binds its own review-loop obligation (owning the session Stop hook for its `/goal` gate when available); as a sub-agent that in-session guarantee is silently lost (see the `[WORKFLOW-IN-WORKFLOW]` execution block in `workflow-review-changes/SKILL.md`). This loop skill therefore also runs inline.
- **Convergence = a whole round applied ZERO fixes** (its fix cycle, steps 12–15, was skipped because reviews passed clean). That, not "one clean review", ends the loop.
- **Scope base is FIXED across rounds; the working tree grows.** Recompute the scope each round as `branch-diff base` ∪ current uncommitted changes — the diff base never moves, so convergence is measured against a stable target.
- **Round cap (default 3)** and **findings-increasing → STOP & escalate** by asking the user directly. NEVER loop open-ended. Cap exhaustion escalates only when CRITICAL/HIGH/MEDIUM fixes are still landing — a LOW-only round converges via the severity floor.
- **The severity floor bounds ITERATION, never the standard.** It ends the loop; it never authorizes shipping a known CRITICAL/HIGH/MEDIUM, and it never applies to a binary gate inside the inner workflow (a failing test is a failure, not a LOW finding).

---

## First Principle — Convergence, Not Motion

> A round that changes files is progress **only if** the next round finds fewer things to fix.
> The loop exists to reach a fixed point (zero fixes), not to keep churning the diff.
> If findings stop shrinking, that is a signal to **escalate**, not to spin another round.

---

## Step 0 — Resolve Scope + Goal Contract (FIRST ACTION)

1. **Parse the review scope** from the user prompt into a stable, reusable scope string. It has two parts UNIONed:
   - **Branch-diff base** — a branch-to-branch or PR diff, e.g. `fix/timelog-org-business-timezone` into `develop`. Capture it as `git diff develop...HEAD` (three-dot: changes on the feature branch since it forked from `develop`) so the base is a **fixed merge-base**, not a moving target.
   - **Current changes** — the uncommitted working-tree changes (`git status --porcelain`, `git diff` + `git diff --staged`).
   - **Scope string (recompute each round):** `{branch-diff base} ∪ {current uncommitted changes}`. The base commit is fixed for the whole loop; the uncommitted set legitimately grows as fixes land.
   - If the prompt names no branch diff (pure "current changes" review), the scope is just the working-tree changes — the loop still applies.
2. **Resolve/create the Goal Contract** per `SYNC:goal-contract-satisfaction-loop` (`plans/goals/{YYMMDD-HHmm}-{slug}/goal.md`, template `.claude/templates/goal-contract-template.md`). Its single **required** Success Criterion:
   > *A complete `$workflow-review-changes` run over `{scope}` applies **zero fixes** (a clean no-op pass — no `$plan-execute` file changes; **round 1** no validated findings of any severity, **round 2+** no validated CRITICAL/HIGH/MEDIUM, with remaining LOW findings deferred rather than fixed).*
   Record the round cap (default 3), the severity floor (LOW non-blocking from round 2), and the scope string in **Constraints**.

## Step 0b — Bind the Convergence Loop (protocol-first; `/goal` is an optional accelerator)

The convergence loop is bound by TWO layers. The **protocol loop (Steps 1–2) is the BINDING mechanism** and MUST be self-driven by you, the running agent, on every host — with or without any command or hook. The **`/goal` command is an OPTIONAL accelerator** layered on top; it is never the primary mechanism, and its absence NEVER weakens the loop. This mirrors the project rule that hooks/trackers are accelerators only — correctness must not depend on them.

**1. Protocol loop — ALWAYS binding (hook/command-independent).** You are personally responsible for not stopping until the loop converges or bounded-escalates. This binds Claude, Codex, and Copilot equally, whether or not `/goal` exists:

> Repeatedly run `$workflow-review-changes` INLINE over `{scope}` (branch-diff base ∪ current uncommitted changes, recomputed each round). After each round, detect whether it applied any fix. From round 2 on, instruct the inner workflow to fix only CRITICAL/HIGH/MEDIUM validated findings and to defer LOW ones. Do NOT stop while the last round still applied fixes. Converge when a full round applies ZERO fixes (reviews clean, `$plan-execute` changed no files) — which from round 2 includes a round whose only validated findings were LOW. Cap at `{N=3}` rounds; if blocking findings do not shrink across 2 consecutive rounds, or the cap is hit with CRITICAL/HIGH/MEDIUM fixes still landing → STOP and escalate by asking the user directly. Never loop open-ended.

Treat this as a standing obligation you re-read at every Step 2 checkpoint — NOT a one-time note you can rationalize away after the first round. The Goal Contract's required Success Criterion (Step 0) is its durable, host-independent record.

**2. `/goal` command — invoke as an accelerator WHEN AVAILABLE.** If a `/goal` command exists and you are permitted to run it in this environment, ALSO invoke it (a real tool/command call, NOT a paraphrase, NOT a Goal Contract file substituted for it) with the SAME condition, so a session Stop hook mechanically enforces the loop:

```
/goal review-changes convergence loop: repeatedly run $workflow-review-changes INLINE over {scope} (branch-diff base ∪ current uncommitted changes, recomputed each round). After each round, detect whether it applied any fix; if fixes>0 → run another round; if a full round applied ZERO fixes (reviews clean, $plan-execute changed no files) → CONVERGED, clear the gate. From round 2 on, the inner workflow fixes only CRITICAL/HIGH/MEDIUM and defers LOW, so a LOW-only round is a zero-fix round and CONVERGES. Do NOT stop while the last round still applied fixes. Cap at {N=3} rounds; if blocking findings do not shrink across 2 consecutive rounds, or the round cap is hit with CRITICAL/HIGH/MEDIUM fixes still landing → STOP and escalate by asking the user directly. Never loop open-ended.
```

The `/goal` Stop hook blocks stopping until the condition holds and auto-clears when met — do not tell the user to clear it.

**If `/goal` is unavailable, unregistered, or not permitted** (e.g. Codex/Copilot, or a Claude run without the command): DO NOT error, DO NOT block, and DO NOT invent a stand-in gate. Record ONE line in the Goal Contract — `/goal accelerator unavailable — loop bound by protocol (Steps 1–2) + this Goal Contract` — and proceed. The protocol loop above plus the Goal Contract are the same gate, enforced by discipline instead of a hook.

> **Nested gates (by design, safe):** each inner `$workflow-review-changes` round self-binds its OWN Step 0 review-loop obligation (and installs its own `/goal` gate WHEN available) that clears when that round reaches its internal clean pass. This OUTER loop persists across rounds and **subsumes** the inner ones (a converged loop implies every inner round ended clean). All self-clear on satisfaction — no orphaned gate. Do NOT tell the user to clear either.

## Step 1 — Round Loop (run → detect → log)

For each round `R` (starting at 1), do ALL of:

1. **Snapshot before:** record the working-tree fingerprint — `git status --porcelain` + `git diff --stat` (or `git rev-parse` of `git stash create` for an exact hash). This is the fixes-applied baseline for the round.
2. **Run the workflow INLINE:** invoke `$workflow-review-changes` via the skill invocation (NEVER the `spawn_agent` tool) with the recomputed `{scope}` as its prompt. Let it run its full 20-step sequence including its own internal fix→re-review loop.
3. **Detect fixes-applied (objective):** compare the working tree after the round to the before-snapshot AND read the workflow's own result:
   - **Fixes applied (>0)** if the working tree changed during the round OR the workflow reported its fix cycle (steps 12–15) ran / `$plan-execute` modified files.
   - **Zero fixes** if the working tree is byte-identical to the before-snapshot AND the workflow reported either no validated findings, or (round 2+) validated findings that are ALL LOW with the fix phase explicitly skipped.
   - If a claimed LOW-only exit has a changed fingerprint, re-review; it is not a no-op convergence pass.
4. **Append an Iteration Log entry** to the Goal Contract: round number, files changed this round (`file:line`), fixes-applied count, the review verdict, and remaining gaps.

## Step 2 — Convergence & Escalation Gate

Evaluate after every round:

| Condition | Action |
| --- | --- |
| Round applied **ZERO fixes** (clean no-op pass) | **CONVERGED** → mark the required criterion PASS in the Goal Satisfaction matrix → clear the `/goal` gate → go to Step 3. |
| Round applied fixes AND round `< N` AND findings shrank vs prior round | Recompute `{scope}`, run round `R+1`. |
| Findings did **not shrink** across 2 consecutive rounds (same/increasing count) | **STOP & escalate** by asking the user directly — a non-converging loop is a signal, not a reason to spin. |
| Round `R ≥ 2` whose validated findings are **ALL LOW** (zero CRITICAL/HIGH/MEDIUM, so zero fixes applied) | **CONVERGED on the severity floor** → do NOT run another round for LOW alone → record every remaining LOW as a deferred finding in the recap + Goal Contract → mark the required criterion PASS → go to Step 3. |
| Round cap `N` hit with CRITICAL/HIGH/MEDIUM fixes still landing | **STOP & escalate** by asking the user directly — report the still-open findings; do not silently continue. (LOW-only at the cap converges via the severity-floor row above.) |

> **Increasing findings = STOP.** If round `R` surfaces MORE findings than round `R-1`, the fixes are regressing the code — STOP and escalate immediately (mirrors the **Issue count increasing** rule under **Iteration Tracking (Conversation-Scoped)** in `workflow-review-changes/SKILL.md`). Never trade one fix for two new findings across rounds.

## Step 3 — Recap

Emit a concise convergence recap: rounds run, total fixes applied per round (the shrinking sequence), the final clean-pass evidence, and the Goal Satisfaction matrix (required criterion PASS). Point to each round's report under `plans/reports/` and the Goal Contract Iteration Log. Do NOT commit or push unless the user explicitly asks.

---

## Convergence Detection — Why Two Conditions

A round counts as converged ONLY when **both** hold: (a) the working tree is unchanged by the round, AND (b) the reviews reported clean at that round's bar — no validated findings in round 1, no validated CRITICAL/HIGH/MEDIUM from round 2 (deferred LOWs listed, not fixed). Both are required because:

- Working-tree-unchanged alone is ambiguous — a round can make no changes because a finding was **unfixable/escalated**, not because it was clean. That is escalation, not convergence.
- Reviews-clean alone is insufficient — an orchestrator can rationalize a "clean" verdict; the objective `git`-diff comparison is the backstop that proves no fix actually landed.

When (a) is true but (b) is false → **escalate** (a real finding the loop cannot close). When (b) is true but (a) is false → the round DID fix things → run another round to re-prove clean.

---

**IMPORTANT MANDATORY sequence:** Step 0 (scope + Goal Contract) → Step 0b (bind the convergence loop: protocol loop primary + optional `/goal` accelerator) → Step 1 (round loop: run `$workflow-review-changes` INLINE → detect fixes → log) → Step 2 (converge on zero-fix round / escalate on non-progress) → Step 3 (recap).

<!-- SYNC:review-policy -->

> **Executable review policy — one predicate, one durable transition model.** Review skills and their tooling MUST use the canonical helper `.claude/scripts/lib/review-policy.cjs` (policy version 2) for round eligibility. The helper's `blockingFindings(round, findings, hardGates)` predicate returns every validated finding in round 1, and only CRITICAL/HIGH/MEDIUM findings in rounds 2–3; `NOT VERIFIABLE` is a separate unresolved-evidence state that remains blocking at every round. Failed binary gates are synthetic blocking findings at every round. `evaluateRound` retains round-2+ LOWs in `deferredLow` and never treats a LOW-only round as blocked after the floor applies. Severity is assigned before the predicate and never changed to obtain a PASS.
>
> **Round and minimum rules.** `MAX_ROUNDS` is 3 and is a ceiling, never a target. A clean review ends once `round >= minRounds`; the default minimum is 1. A caller may declare `minRounds` explicitly (for example, 2) when an independent second pass is a real requirement; the declaration is persisted and cannot be inferred from a round counter. A failing test-green, security-must-fix, required-artifact, or other binary gate is never waived by the severity floor.
>
> **Durable run record.** A review run MUST identify `runId`, target fingerprint, policy version, target revision, minimum/maximum rounds, completed rounds, full findings/gate evidence, interruption/resume metadata, and acceptance. Use the atomic, lock-serialized transitions in `review-policy.cjs`: `start`, `record`, `accept`, `interrupt`, `resume`, `invalidate`, and `check`. Repeating an identical completed round is idempotent and MUST NOT consume budget twice. A changed target fingerprint invalidates prior evidence and acceptance but MUST preserve the bounded round budget; stale evidence cannot be accepted. A policy-version change (including the round-2 LOW floor) invalidates old records; start a new run rather than interpreting old evidence under new semantics. Interrupted/resumed runs retain completed rounds and findings. The record is bookkeeping, not consent, native permission, or proof that a host actually performed the review.
>
> **CLI boundary.** The helper CLI accepts JSON on stdin and uses its own real clock; a supplied `now` is rejected. State directories must be absolute, non-root real directories, records are size-bounded, and malformed/locked state fails closed for the transition. Full reports remain on disk; an inline result envelope is only a transport summary. Any new review policy consumer must add a semantic fixture, boundary counter-cases, a seeded mutant, and a report with the target fingerprint and command exit status.

<!-- /SYNC:review-policy -->

<!-- SYNC:goal-contract-satisfaction-loop -->

> **Goal Contract Satisfaction Loop** — Persist the user goal in an external file, execute against it, and loop review/fix until every saved required criterion passes or a blocker escalates. Bounded closed loop — NEVER open-ended autonomous exploration.
>
> 1. **Resolve the active goal** (in order): active plan `goal.md` → `plans/goals/{YYMMDD-HHmm}-{slug}/goal.md` → create a new Goal Contract from the current user request (template: `.claude/templates/goal-contract-template.md`).
> 2. **Required sections:** Original Request, Purpose, Success Criteria (checkboxes; mark required vs optional), Constraints, Evidence Required, Iteration Log, Goal Satisfaction matrix.
> 3. **Before work:** read the active goal and map planned work to saved success criteria — execution serves the saved criteria, never chat memory alone.
> 4. **After execution/verification:** append an Iteration Log entry — result, evidence references (`file:line`, command output, report path), remaining gaps.
> 5. **Review gate:** emit a Goal Satisfaction matrix — `| Success Criterion | Evidence | Status |` with PASS/FAIL/BLOCKED. Overall PASS requires every required criterion PASS.
> 6. **Loop rule (retry):** required criterion FAIL → validate the gap is real → fix → re-review only the affected criteria. Stop cleanly when all required criteria PASS.
> 7. **Escalation rule (stop):** two consecutive iterations with no criterion progressing, or a blocker needing user input → mark the criterion BLOCKED with a user-facing reason and escalate. NEVER loop indefinitely.
> 8. **Skip rule:** tiny conversational tasks may skip the goal file ONLY with a recorded one-line reason. User-accepted gate skips are recorded in the goal file with reason and scope.
> 9. **Security:** NEVER store secrets, tokens, credentials, or private customer data in goal files — store evidence references and redact sensitive values.
>
> **Blocked until:** active goal resolved (or skip reason recorded) · saved success criteria read before edits · iteration evidence appended after execution · Goal Satisfaction matrix emitted before any PASS verdict.

<!-- /SYNC:goal-contract-satisfaction-loop -->

<!-- SYNC:critical-thinking-mindset -->

> **Critical Thinking Mindset** — Apply critical thinking, sequential thinking. Every claim needs traced proof, confidence >80% to act.
> **Anti-hallucination:** Never present guess as fact — cite sources for every claim, admit uncertainty freely, self-check output for errors, cross-reference independently, stay skeptical of own confidence — certainty without evidence root of all hallucination.

<!-- /SYNC:critical-thinking-mindset -->

<!-- SYNC:ai-mistake-prevention -->

> **AI Mistake Prevention** — Failure modes to avoid on every task:
>
> **Re-read files after context changes.** Context compaction, resume, or long-running work can make memory stale; verify current files before acting.
> **Verify generated content against source evidence.** AI hallucinates APIs, names, claims, and document facts. Check the relevant source before documenting or referencing.
> **Check downstream references before deleting or renaming.** Removing an artifact can stale docs, generated mirrors, configs, and callers; map references first.
> **Trace the full impact chain after edits.** Changing a definition can miss derived outputs and consumers. Follow the affected chain before declaring done.
> **Verify ALL affected outputs, not just the first.** One green check is not all green checks; validate every output surface the change can affect.
> **Assume existing values are intentional — ask WHY before changing OR flagging one as a defect.** Before changing or reporting a constant, limit, flag, cutoff, wording, or pattern, read nearby context and history, the CALLER's ordering, and 2+ sibling call sites of the same convention. A doc stating WHAT without WHY is missing rationale, not proof of a missing guard.
> **Surface ambiguity before acting — don't pick silently.** Multiple valid interpretations require an explicit question or stated assumption with risk.
> **Assert the outcome your system owns, not the intermediate state your infrastructure owns.** When verifying async work, assert the final business state — never the delivery/retry bookkeeping held in shared infrastructure that any co-running process can write. Such a check passes when run alone and flakes the moment anything else shares that infrastructure.
> **Keep shared guidance role-relevant.** Universal guidance must help every receiving skill or agent; code-specific obligations belong only in code-specific protocols.

<!-- /SYNC:ai-mistake-prevention -->

<!-- SYNC:trade-off-interrogation-gate -->

> **Trade-Off Interrogation Gate** — ALWAYS ask these THREE questions before ANY verdict, score, finding, or recommendation — about the thing under review AND about every recommendation YOU make. — why: naming a benefit without its price is an endorsement, not a review; the costliest trade-offs are the ones nobody wrote down.
>
> 1. **Is there any trade-off?** Name what it SACRIFICES. "None" / "pure win" is an unfinished analysis, NOT an answer — to claim none, state which dimensions you checked and why each is unaffected: future change cost · complexity · performance/latency · memory/cost · coupling · reversibility · migration burden · operational load · blast radius · security posture · testability · team skill/ramp · delivery time · UX.
> 2. **Is it worth it?** Weigh gain against sacrifice EXPLICITLY — what is gained (with a metric) · what it costs · WHO pays · WHEN it comes due — then emit **WORTH IT / NOT WORTH IT / UNCLEAR**. "Better" with no metric and no cost FAILS this question. NOT WORTH IT → withdraw or replace the recommendation, never keep it as-is.
> 3. **Is the trade-off material enough to CONFIRM WITH THE USER?** A material trade-off is the user's call, never yours. **MATERIAL** when ANY holds: irreversible / one-way door (data migration, public contract, storage format, vendor lock-in) · cost shifted onto someone else (another team, ops/on-call, future maintainer, end user) · one quality attribute traded for another (correctness↔speed, security↔convenience, latency↔cost, simplicity↔flexibility) · a boundary crossed (client↔server tier, service contract, event contract, shared library) · a high-consequence path (auth, money, data integrity, breaking change, High/Medium residual risk) · the worth-it verdict is UNCLEAR.
>
> **MATERIAL → STOP and confirm by asking the user directly BEFORE the verdict stands** — state the trade-off, both options, what each sacrifices, and your recommendation. **NOT material →** record it inline with a one-line justification and proceed.
>
> **Non-asking execution contexts — ESCALATE BY HANDOFF, never by silence.** ask the user directly reaches only the main interactive agent: a sub-agent cannot ask the user, and a terminal/verdict-only mode asks nothing by design. When you are running in such a context, the obligation is **redirected, never waived** — do ALL of: (a) complete questions 1 and 2 normally; (b) decide materiality and record it in the Trade-Off Assessment row with `confirmed? = NO — cannot ask from this context`; (c) **name the unconfirmed MATERIAL trade-off explicitly in your returned summary/verdict so the CALLER (or parent orchestrator) escalates it by asking the user directly on your behalf** — a material trade-off mentioned only inside a report file on disk is NOT a handoff; (d) do not emit an unqualified PASS — mark the verdict as carrying an unconfirmed material trade-off, so the caller's gate stays closed until the user answers. The caller inherits the escalation duty the moment it reads your return.
>
> This carve-out is about **reachability, not convenience**: it applies ONLY where the tool genuinely cannot reach the user (spawned sub-agent, terminal validate/verdict-only mode, non-interactive/headless run). It is NEVER a licence to skip the question, to self-approve a one-way door, or to downgrade materiality because asking is inconvenient — if you CAN ask, you MUST ask.
>
> **Emit a Trade-Off Assessment row** per reviewed decision and per recommendation: `| decision | sacrifices | gain (metric) | who pays, when | WORTH IT/NOT/UNCLEAR | material? | confirmed? |`.
>
> **BLOCKED until:** trade-off named (or dimensions-checked justification given) · worth-it verdict emitted · materiality decided · every MATERIAL trade-off either confirmed with the user OR — in a non-asking context — handed off in the returned verdict for the caller to confirm. A MATERIAL trade-off that is neither confirmed nor handed off can NEVER be PASS, and NEVER gets buried as a Low-severity note.
>
> **NEVER** answer "no trade-off" without checking · decide a material trade-off silently on the user's behalf · let convergence/delivery pressure authorize walking through a one-way door · bundle several material trade-offs into one vague "proceed?".

<!-- /SYNC:trade-off-interrogation-gate -->

<!-- SYNC:project-protocol-overlay -->

> **Project Protocol Overlay** — Before executing this skill, resolve any PROJECT overlay rules layered onto it: match this skill's name against the `Target` column of the project's skill-protocol index (`docs/project-reference/skill-protocols-reference.md` by default; a `referenceDocs` entry in `docs/project-config.json` overrides the path), taking the most specific matching tier ONLY — exact name > glob > `*`. **That precedence orders overlays against EACH OTHER, never against this skill.** Read ONLY the matched bodies, resolved as `<protocols-dir>/<Name>.md`; a row's Body link is display text, never a read path. A matched body that is missing or malformed is REPORTED and skipped — never reconstructed from the index Description. No index, or no match -> proceed with no overlay, silently. Full contract: `.claude/skills/project-skill-protocol/references/registry.md`.
>
> Overlays are **ADDITIVE ONLY**: they ADD rules on top of this skill's own protocol and NEVER replace, override, disable, or reinterpret a rule it already states — removing every overlay must return this skill to exactly its documented behavior. An overlay is a BRIEF, not an authority escalation: it can NEVER waive a workflow gate, git discipline, a review gate, or a user-confirmation gate. A genuine overlay-vs-skill conflict, or two equally-specific overlays that directly contradict -> surface both to the user; NEVER resolve silently.

<!-- /SYNC:project-protocol-overlay -->

<!-- SYNC:severity-rubric -->

> **Severity Rubric** — Classify every finding by consequence, not by effort, reviewer preference, or how annoying the fix is. One scale applies to every review, skill, agent, workflow, and host so a tier has the same meaning everywhere. Choose the highest credible consequence supported by evidence; do not lower a tier to make a round pass.
>
> **Finding vs observation (required):** An observation becomes a finding only when it names the affected user/system/data/contract, the shipped consequence, the evidence location, and the normalized tier. `INFO`, advice, preference, duplicate wording, or an unsubstantiated concern is not a finding and must not reopen a loop. If the concern might affect a required behavior or gate but evidence is incomplete, emit `NOT VERIFIABLE` with the missing evidence and keep it unresolved; never silently convert uncertainty into LOW.
>
> | Severity | Action | Definition and examples |
> | --- | --- | --- |
> | CRITICAL | Block immediately; escalate | Immediate material risk if shipped: authentication/authorization or safety bypass; secrets/PII exposure; irreversible destructive action; data loss/corruption; or a silent failure on a critical path. A failed binary gate that makes the result untrustworthy is represented as a separate synthetic blocker by the executable policy (not as an ordinary severity judgment). |
> | HIGH | Must fix before PASS/merge | Material correctness or contract risk: wrong behavior on a supported path; violated business/data invariant; meaningful privacy or authority gap; breaking API/schema/compatibility change; likely harm to users/downstream systems; or a missing proof for a behavior-changing fix. |
> | MEDIUM | Must clear the current round; escalate if the fix needs an owner decision | Bounded but consequential risk: an edge case, resilience/observability/testability/maintainability gap, credible future defect, or local architectural drift whose impact is real but not immediate material loss. An explicit follow-up records the escalation/residual risk; it does not make an open MEDIUM a clean pass. |
> | LOW | Record and defer; never open another fix/re-review round from round 2 onward | Non-blocking polish with no credible present correctness, security, privacy, authority, availability, or data-integrity impact: wording/formatting, minor documentation or convention drift, optional defensive cleanup, or a cosmetic/refinement suggestion. |
>
> **Consequence decision tree (apply in order):** (1) Is a binary gate failed? Keep it as a separate hard blocker (the executable helper represents it as synthetic CRITICAL); do not use the ordinary severity label to hide what failed. Otherwise, would shipping permit immediate material security/safety/authority harm, irreversible destruction, data loss/corruption, or a critical-path silent failure? → **CRITICAL**. (2) Otherwise, does a supported path, invariant, public contract, privacy/authority boundary, compatibility promise, or behavior-changing proof fail with material user/downstream impact? → **HIGH**. (3) Otherwise, is there a bounded but consequential edge, resilience, observability, testability, maintainability, or architectural gap with a credible impact? → **MEDIUM**. (4) Otherwise, is the evidence sufficient to show only non-blocking polish with no credible present material impact? → **LOW**. (5) If the evidence needed to choose between steps 1–4 is missing, → **NOT VERIFIABLE**, not LOW. When multiple tiers fit, select the highest credible consequence; effort, implementation cost, reviewer discomfort, frequency alone, and proximity to the round cap never decide the tier.
>
> **Boundary examples (normalize before applying the round predicate):** an auth bypass, exposed secret/PII, destructive command without an authority gate, or failed required test/generation/parity gate is **CRITICAL**; a wrong supported response, broken invariant/API/schema, meaningful privacy/authority defect, or unproven behavior-changing fix is **HIGH**; a bounded retry/timeout/alert/testability gap or credible maintainability drift is **MEDIUM**; a typo, formatting inconsistency, optional cleanup, or cosmetic suggestion proven not to affect present behavior is **LOW**. A missing fact about any of those boundaries is **NOT VERIFIABLE** until evidence or an explicitly documented residual-risk decision exists.
>
> **Classification procedure (required for every finding):** (1) state the affected user, system, data, contract, or gate; (2) assess consequence if the issue ships; (3) assess exposure/likelihood and reversibility/detectability; (4) select the highest tier justified by those facts; (5) cite `file:line` or equivalent evidence and a confidence percentage. Effort, implementation cost, reviewer discomfort, and proximity to the round cap are never severity inputs. `NOT VERIFIABLE` is a pending evidence state, not one of the four tiers and never a LOW escape hatch: if the unresolved claim could affect required behavior, security, privacy, authority, availability, data integrity, or a binary gate, it remains an open evidence blocker until resolved or explicitly owner-accepted with documented residual risk. Classify an item LOW only when evidence supports the absence of credible present material impact.
>
> **Hard-gate rule:** Binary gates (tests, required artifacts, security must-fix checks, generated parity, policy compliance) are not ordinary severity-rated findings. The executable helper records a failed gate as a synthetic CRITICAL blocker solely so one predicate can carry it; the report must still name the gate and failure evidence. A failed gate blocks at every round, including when all ordinary findings are LOW; never disguise a failed gate as LOW.
>
> **Score-based skills** map their numeric scale onto these tiers — do not invent a parallel vocabulary:
>
> - **0-2 criterion scoring** (e.g. production-readiness-review): `0` = CRITICAL/HIGH (criterion unmet, blocks readiness), `1` = MEDIUM (partial, consequential gap), `2` = pass (no finding). If the criterion is only polish, use LOW rather than forcing a `0`.
> - **Two-axis scoring** (e.g. performance-review, impact × likelihood): high impact + high exposure → CRITICAL/HIGH; material impact with bounded exposure → HIGH/MEDIUM; low impact and low exposure → LOW. Record the axes and why the selected tier is the highest credible consequence.
> - **Scorecards / `/20` grades** (e.g. architecture-scalability-review): the aggregate score and verdict band are separate from finding severity. A sub-80 area is evidence to investigate, not an automatic CRITICAL/HIGH/MEDIUM/LOW label; classify each underlying gap by the consequence decision tree and keep advisory score deductions separate from blocking findings.
>
> **Domain-vocabulary normalization (mandatory):** Specialized skills may keep a local reporting vocabulary, but it MUST feed this same four-tier round predicate — never a second severity system:
>
> - `BLOCKED`, `HARD FAIL`, or `FAIL` is a blocking local verdict, not an automatic CRITICAL label. Classify the underlying consequence as CRITICAL when it is an immediate material risk or failed binary gate; otherwise classify it as HIGH or MEDIUM with evidence, while preserving the local block until the owning gate is satisfied.
> - `WARN` is not permission to ignore a finding. Map it to MEDIUM when the gap is consequential, to LOW only when evidence supports no credible present material impact, or upward to HIGH/CRITICAL when the consequence warrants it. `PASS`/compliant is not a finding.
> - UI `P0`/`P1`/`P2`/`P3`/`P4` map to CRITICAL/HIGH/MEDIUM/LOW/LOW respectively as a starting point; override upward only when the evidence shows a higher shipped consequence. A P0/P1 accessibility or task-completion floor remains a blocking gate even when a local UI report calls it a priority rather than a severity.
> - Numeric SRE/readiness or impact/likelihood scores are evidence inputs, not replacement tiers. Emit the score, the consequence, and the normalized CRITICAL/HIGH/MEDIUM/LOW tier together. `INFO`/advisory observations are not findings unless the evidence shows a material consequence.
>
> A finding's tier drives the gate: CRITICAL/HIGH/MEDIUM remain actionable and blocking under the round policy; LOW may be tracked as a follow-up and, from round 2, does not by itself justify another fix/re-review. An owner decision may explain or schedule an open MEDIUM but does not turn it into a clean pass; owner acceptance never makes a failed binary gate pass and must record scope, rationale, and residual risk.

<!-- /SYNC:severity-rubric -->


<!-- SYNC:goal-contract-satisfaction-loop:reminder -->

- **MANDATORY** Resolve the active Goal Contract BEFORE work (active plan `goal.md` → `plans/goals/{YYMMDD-HHmm}-{slug}/goal.md` → create from current request) and read saved success criteria before editing.
- **MANDATORY** Append iteration evidence after execution; emit a Goal Satisfaction matrix (PASS/FAIL/BLOCKED) before reporting PASS; loop on validated FAIL; escalate repeated no-progress or blockers. NEVER store secrets in goal files.

<!-- /SYNC:goal-contract-satisfaction-loop:reminder -->

<!-- SYNC:critical-thinking-mindset:reminder -->

**MUST ATTENTION** apply critical + sequential thinking — every claim needs appropriate traced evidence (`file:line` for repo/code claims; source URL or artifact section for research, product, content, and docs claims); confidence >80% to act, <60% DO NOT recommend. Anti-hallucination: never present guess as fact, admit uncertainty freely, cross-reference independently, stay skeptical of own confidence.

<!-- /SYNC:critical-thinking-mindset:reminder -->

<!-- SYNC:ai-mistake-prevention:reminder -->

**MUST ATTENTION** apply AI mistake prevention — verify generated content against evidence, trace downstream references before deleting or renaming, verify all affected outputs, re-read files after context loss, and surface ambiguity before acting.

<!-- /SYNC:ai-mistake-prevention:reminder -->

<!-- SYNC:trade-off-interrogation-gate:reminder -->

- **MANDATORY MUST ATTENTION ALWAYS ASK THE 3 TRADE-OFF QUESTIONS** — on the thing under review AND on every recommendation you make: (1) **is there any trade-off?** name what it SACRIFICES (change cost · complexity · perf · coupling · reversibility · migration · ops load · blast radius · security · testability · delivery time · UX) — "none"/"pure win" is an unfinished analysis, so state the dimensions checked; (2) **is it worth it?** gain (with a metric) vs cost, WHO pays, WHEN → emit **WORTH IT / NOT WORTH IT / UNCLEAR**; NOT WORTH IT → withdraw or replace it; (3) **is it material enough to confirm with the user?** irreversible/one-way door · cost shifted onto another team/ops/maintainer/user · one quality attribute traded for another · a tier/service/event/library boundary crossed · auth/money/data-integrity/breaking-change/High-or-Medium-risk path · verdict UNCLEAR → **STOP and confirm by asking the user directly BEFORE the verdict**.
- **MANDATORY** A MATERIAL trade-off with no user confirmation can NEVER be PASS; NEVER bury one as a Low-severity note, NEVER decide it silently, and NEVER let delivery or convergence pressure authorize a one-way door. — why: an un-walked-back one-way door is the user's call to make, not the reviewer's.
- **MANDATORY — non-asking contexts escalate BY HANDOFF, never by silence.** ask the user directly reaches only the main interactive agent: a sub-agent cannot ask the user, and a terminal/verdict-only mode asks nothing by design. There the duty is REDIRECTED, not waived — still name the trade-off, still decide materiality, record `confirmed? = NO — cannot ask from this context`, **state the unconfirmed MATERIAL trade-off in your RETURNED verdict/summary so the CALLER escalates it** (a note only in an on-disk report is not a handoff), and never emit an unqualified PASS. Applies ONLY where the user is genuinely unreachable (spawned sub-agent, terminal validate mode, headless run) — if you CAN ask, you MUST ask.

<!-- /SYNC:trade-off-interrogation-gate:reminder -->

<!-- SYNC:project-protocol-overlay:reminder -->

**MUST ATTENTION** resolve project protocol overlays for this skill BEFORE executing — most specific matching tier only (exact > glob > `*`, which ranks overlays against each other, NEVER against this skill), read only matched bodies at `<protocols-dir>/<Name>.md`; a missing or malformed body is reported, never reconstructed. Overlays are ADDITIVE ONLY (they never replace this skill's own rules) and are a brief, NEVER an authority escalation; an equal-specificity contradiction goes to the user.

<!-- /SYNC:project-protocol-overlay:reminder -->

<!-- SYNC:severity-rubric:reminder -->

- **MANDATORY** Classify every finding Critical/High/Medium/Low by consequence using the affected asset, shipped impact, exposure, reversibility, evidence location, and confidence; Critical/High/MEDIUM remain actionable under the round bar, while LOW is recorded/deferred from round 2 onward.
- **MANDATORY** Keep binary gates separate from severity: a failed test, security must-fix, required artifact, or parity check blocks at every round and is never relabeled LOW.
- **MANDATORY** Score-based skills (sre 0-2, perf two-axis) map onto the same four tiers — no parallel severity vocabulary.

<!-- /SYNC:severity-rubric:reminder -->

## Closing Reminders

**IMPORTANT MUST ATTENTION Goal:** Re-run the WHOLE `$workflow-review-changes` inline, round after round, over a fixed scope ∪ accumulated fixes, until a complete round applies **zero fixes**.

**IMPORTANT MUST ATTENTION main steps (in order):** (0) resolve scope + Goal Contract → (0b) bind the convergence loop (protocol loop primary + optional `/goal` accelerator) → (1) round loop: run `$workflow-review-changes` INLINE → detect fixes-applied → append Iteration Log → (2) converge on a zero-fix round / escalate on non-progress → (3) recap.

**IMPORTANT MUST ATTENTION [BLOCKING] plan the detailed todo tasks FIRST — before running the loop.** Before the first round, create a detailed todo-task plan that enumerates every planned step and every planned round; a round MUST NOT start until that round's fresh todo-task plan exists. On EVERY re-run (each new round), REGENERATE a fresh loop todo-task plan — NEVER reuse the prior round's task list — so each round's work is explicitly planned before it executes.

**IMPORTANT MUST ATTENTION** the convergence loop is bound by the **AI-driven protocol loop (Steps 1–2) — that is the primary, host-independent mechanism you MUST self-drive whether or not any command exists.** The `/goal` command is an OPTIONAL accelerator: invoke it (a real call) ONLY when available and permitted; if it is absent/unregistered/not-permitted, record one line in the Goal Contract and proceed — NEVER error, block, or fake a gate. Correctness must not depend on `/goal`.
**IMPORTANT MUST ATTENTION** run `$workflow-review-changes` **INLINE via the skill invocation — NEVER as a sub-agent** (it self-binds its own review-loop obligation + a session `/goal` gate when available).
**IMPORTANT MUST ATTENTION** convergence = a whole round applied **ZERO fixes** (fix cycle skipped, working tree unchanged, reviews clean at that round's bar) — not merely one clean review. **From round 2 the severity floor applies: LOW findings are deferred not fixed, so a LOW-only round is a zero-fix round and ENDS the loop.**
**IMPORTANT MUST ATTENTION** the outer loop's value is the **fresh full specialist sweep** the inner loop never re-runs (see the **Conditional Inline Re-Review Protocol** in `workflow-review-changes/SKILL.md`) — that is why this skill exists.
**IMPORTANT MUST ATTENTION** keep the diff **base fixed** across rounds; recompute `{scope}` = branch-diff base ∪ current uncommitted changes each round.
**IMPORTANT MUST ATTENTION** enforce the **round cap (default 3)**; blocking findings not shrinking across 2 rounds, or cap hit with CRITICAL/HIGH/MEDIUM fixes still landing → **STOP & escalate** by asking the user directly. NEVER loop open-ended.
**IMPORTANT MUST ATTENTION** do NOT commit or push unless the user explicitly asks.

<!-- CODEX:SYNC-PROMPT-PROTOCOLS:START -->
## Static Prompt Protocol Mirror (Auto-Synced)

Source: `.claude/.ck.json` + `.claude/skills/shared/sync-inline-versions.md` (`:full` blocks) + `.claude/scripts/lib/hookless-prompt-protocol.cjs` (legacy filename; static protocol composer)

## [WORKFLOW-EXECUTION-PROTOCOL] [BLOCKING] Workflow Execution Protocol — MANDATORY IMPORTANT MUST CRITICAL. Do not skip for any reason.

**Generic portability boundary:** Reusable skills and protocol text stay project-neutral; project-specific conventions are discovered from docs/project-config.json and docs/project-reference/. Apply shared AI-SDD from `shared/sdd-artifact-contract.md`. Read `docs/project-config.json` and `docs/project-reference/docs-index-reference.md`, then open the project reference docs named there immediately before the first target read, grep, edit, test, or analysis. For spec, test-case, behavior-change, public-contract, or `docs/specs/` work, route through the local spec docs named by the docs index: `feature-spec-reference.md`, `spec-system-reference.md`, `spec-principles.md`, and `workflow-spec-test-code-cycle-reference.md` when specs/tests/code must stay synchronized. If either file or a required reference doc is missing or stale, auto-run `$project-init` (or the narrow lower-level route such as `$project-config`, `$docs-init`, `$scan-all`, or `$scan --target=<key>`) before ordinary project-specific work. After compaction, resume, delegation, or a material context change, re-read the required docs and state `Reference docs read: ... | Not applicable: ...`; a hook reminder or prior conversation is not proof that the files are loaded. Any supported AI tool may execute when this shared context and local docs are available.

1. **DETECT:** If the prompt starts with an explicit slash skill/workflow command, execute it directly. Otherwise match the prompt against the workflow catalog and skill list.
2. **ANALYZE:** Choose the best option: execute directly, invoke a skill, activate a standard workflow, or compose a custom step combination.
3. **AUTO-SELECT:** Pick the best option yourself. Do not ask the user to choose between direct execution, skill, standard workflow, or custom workflow.
4. **ACTIVATE:** For a selected workflow, call `$start-workflow <workflowId>`; for a selected skill, invoke that skill; for a custom workflow, sequence custom steps directly; for direct execution, proceed with the task.
5. **CREATE TASKS:** task tracking for ALL workflow/skill/custom steps before execution when the selected path has multiple steps.
6. **PARALLELIZE:** Before executing the task list, tag each task `PAR` (independent inputs + write set disjoint from every other `PAR` task) or `SEQ` (name the blocking dependency), group `PAR` tasks into waves, declare the wave plan, and spawn each wave's sub-agents in ONE message — all-return barrier per wave, fan-out one level deep unless a sub-agent's own definition authorizes further fan-out. Sequential-by-default is a defect when tasks are independent; do not parallelize shared write targets, output-consuming tasks, trivial single-file work, ordering a skill or workflow explicitly fixes, or user-approval gates.
7. **EXECUTE:** Advance per the **Workflow Step Advancement & Parallel Phases** rule in your context instructions — model-driven; a sub-agent completion advances a step identically to an inline call; a parallel-phase group is an all-return barrier (advance only after ALL members return, never serialize it)
## Shared AI-SDD Protocol Markers

Source: `.claude/skills/shared/sync-inline-versions.md`

## SYNC:ai-sdd-artifact-contract

> **AI-SDD Artifact Contract** — Shared spec-driven development rules stay portable and source-owned.
>
> 1. Keep reusable AI-SDD principles in `.claude`; put repository-specific paths, commands, owners, products, and formats in project config/reference docs.
> 2. Preserve cycle: `spec -> plan -> tasks -> implement -> verify -> update spec/docs`.
> 3. Trace every requirement or invariant through decision, task, TC/test, source evidence, and docs/spec update.
> 4. Treat code-to-spec extraction as reference-only until accepted by the canonical spec owner.
> 5. Any supported AI tool may plan, implement, review, or verify with synced context; using multiple tools is optional.
> 6. Update `.claude` source first, then sync generated mirrors; do not manually edit `.agents`, `.codex`, or `AGENTS.md`. — why: mirrors are generated artifacts; hand-edits are overwritten on the next sync
> 7. If `docs/project-config.json`, root instruction files, or a required project-reference doc is missing or stale, auto-run `$project-init` or the narrow lower-level route before ordinary project-specific work.
>
> **Active reference:** `shared/sdd-artifact-contract.md` in the active skills root.

---

## SYNC:ai-sdd-artifact-contract:reminder

- **MANDATORY** Apply `shared/sdd-artifact-contract.md`; keep reusable AI-SDD in `.claude` and local rules in project docs.
- **MANDATORY** Code-to-spec extraction is reference-only until canonical acceptance; any supported AI tool may execute with synced context.
- **MANDATORY** Update `.claude` source before syncing generated mirrors; do not manually edit `.agents`, `.codex`, or `AGENTS.md`.
- **MANDATORY** Missing or stale project config, root instruction files, or required reference docs route project-specific work through `$project-init` or the narrow setup route automatically.
**[TASK-PLANNING] [MANDATORY]** BEFORE executing any workflow or skill step, create/update task tracking for all planned steps, then keep it synchronized as each step starts/completes.
## [LESSON-LEARNED-REMINDER] [BLOCKING] Task Planning & Continuous Improvement — MANDATORY. Do not skip.

Break work into small tasks (task tracking) before starting. Add final task: "Analyze AI mistakes & lessons learned".

**Extract lessons — ROOT CAUSE ONLY, not symptom fixes:**
1. Name the FAILURE MODE (reasoning/assumption failure), not symptom — "assumed API existed without reading source" not "used wrong enum value".
2. Generality test: does this failure mode apply to ≥3 contexts/codebases? If not, abstract one level up.
3. Write as a universal rule — strip project-specific names/paths/classes. Useful on any codebase.
4. Consolidate: multiple mistakes sharing one failure mode → ONE lesson.
5. **Recurrence gate:** "Would this recur in future session WITHOUT this reminder?" — No → skip `$learn`.
6. **Auto-fix gate:** "Could `$code-review`/`$code-simplifier`/`$security-review`/`$lint` catch this?" — Yes → improve review skill instead.
7. BOTH gates pass → ask user to run `$learn`.
**[CRITICAL-THINKING-MINDSET]** Apply critical thinking, sequential thinking. Every claim needs traced proof, confidence >80% to act.
**Anti-hallucination principle:** Never present guess as fact — cite sources for every claim, admit uncertainty freely, self-check output for errors, cross-reference independently, stay skeptical of own confidence — certainty without evidence root of all hallucination.
**AI Attention principle (Primacy-Recency):** Put the 3 most critical rules at both top and bottom of long prompts/protocols so instruction adherence survives long context windows.
**Goal-driven execution:** Define success criteria first, loop until verified, and stop only when observable checks pass.
**Tests verify intent:** Tests must protect business rules/invariants and fail when the protected intent breaks, not only mirror current behavior.
## Common AI Mistake Prevention (System Lessons)

- **Re-read files after context compaction.** Edit requires prior Read in same context; compaction wipes read state. Re-read before editing.
- **Grep for old terms after bulk replacements.** AI over-trusts find/replace completeness. Grep full repo after bulk edits for missed refs in docs/configs/catalogs.
- **Check downstream references before deleting.** Deletions cascade doc/code staleness. Map referencing files before removal.
- **After memory loss, check existing state before creating new.** Compaction wipes prior-work memory. Query current state to resume — never blindly duplicate.
- **Verify AI-generated content against actual code.** AI hallucinates APIs, class names, method signatures. Grep to confirm existence before documenting/referencing.
- **Trace full dependency chain after edits.** Changing a definition misses downstream consumers. Trace the full chain.
- **When renaming, grep ALL consumer file types.** Some file types silently ignore missing refs (no compile error). Search code, templates, configs, generated files.
- **Trace ALL code paths when verifying correctness.** Code existing ≠ code executing. Trace early exits, error branches, conditional skips — not just happy path.
- **Update docs that embed canonical data when source changes.** Docs inlining derived data (workflows, schemas, configs) go stale silently. Update all embedding docs alongside source.
- **Verify sub-agent results after context recovery.** Background agents may finish while parent compacted — grep-verify output, don't trust assumed completion.
- **Cross-check full target list against sub-agent assignments.** Parallel sub-agents by category miss boundary items. Reconcile union of assignments against target list before proceeding.
- **Sub-agents inherit knowledge only from their agent .md definition — use custom agent types, not built-in Explore.** Tool adoption = permission + knowledge + enforcement (numbered workflow step).
- **Persist sub-agent findings incrementally, not as a final batch.** Long sub-agents hit cutoffs before final write — findings lost. Instruct append-per-section to report file.
- **When debugging, ask "whose responsibility?" before fixing.** Trace caller (wrong data) vs callee (wrong handling). Fix at responsible layer — never patch symptom site.
- **Test failure → record a provisional verdict before trace/edit, then investigate.** Use the full five-way taxonomy: SOURCE-WRONG (production violates intent), TEST-WRONG (assertion/setup is stale), TEST-NOT-OPTIMAL (valid but fragile or low-signal test), ENVIRONMENT-BLOCKED (external state prevents a verdict), or AMBIGUOUS (intent/evidence cannot choose safely). Then trace root cause and triangulate against the governing spec (`docs/specs/**` if one exists) AND source. NEVER weaken an assertion, add a skip, relax a timeout, or change source merely to force green.
- **Grep ALL removed names after extraction/refactoring.** Primary file "done" ≠ secondary files clean. Grep entire scope for every removed symbol before declaring complete.
- **Assume existing values are intentional — ask WHY before changing OR flagging one as a defect.** Pattern-matching as "wrong" skips context. Before changing or reporting any constant/limit/flag/cutoff: read comments, git blame, the CALLER's ordering (the guarantee that makes the value correct usually lives in code running immediately BEFORE the cited line), and 2+ sibling call sites of the same convention. A doc stating WHAT without WHY is missing rationale, not proof of a missing guard — and in a validation pass, an accurate `file:line` citation proves the transcription, never the defect.
- **Verify ALL affected outputs, not just the first.** One build green ≠ all green. Multi-stack changes (backend/frontend/tests/docs) require verifying EVERY output.
- **Evaluate fit before copying a nearby pattern.** Closest example ≠ matching preconditions — verify the new context shares the same constraints, base classes, scope, lifetime.
- **Holistic-first debugging — resist nearest-attention trap.** Don't dive into first plausible cause. List EVERY precondition (config, env vars, paths, DB, endpoints, creds, versions, DI, data). Verify each against evidence (grep/query — not reasoning). Ask "what would falsify this?" — if nothing, it's not a hypothesis. Most expensive failure: going deeper in "obvious" layer while bug sits in layer never questioned.
- **Surgical changes — apply the diff test (context-aware).** Two modes: (1) Bug fix → every line traces to the bug; no restyling; orphan cleanup only for imports YOUR changes made unused. (2) Review/enhancement → implement improvements AND announce as "Enhancement beyond main request: [what]". Never silently scope-creep. Diff test: "Would this line exist if I wasn't asked to do X?" — if no, delete or announce.
- **Surface ambiguity before coding — don't pick silently.** Multiple valid interpretations → present each with effort: "[Request] could mean (1) [N h], (2) [N h]. Which matters?" List scope/format/volume/constraints assumptions first. If simpler path exists, say so. Never silently pick.
- **[MANDATORY FIRST ACTION] ALWAYS activate a suitable skill or workflow BEFORE responding.** Match task against workflow catalog + skill list; invoke via skill invocation or `$start-workflow <workflowId>`. NEVER answer or write code before checking. Skip = protocol violation.
- **Why-Review adversarial mindset — apply when reviewing any plan, decision, or design.** Default SKEPTIC not VALIDATOR: steel-man a rejected alternative, invert each stated reason ("what does it sacrifice?"), stress-test top 2-3 assumptions, run pre-mortem ("ships, fails in 3 months — what breaks?"), surface 1-2 alternatives author missed. Section presence ≠ quality; quality = causal reasoning + concrete mitigations + evidence, not "it's better" or "monitor closely".
- **Front-load report-write in sub-agent prompts for large reviews.** Many-file sub-agents hit budget before final write — findings lost. Design prompts so: (1) report-write is first explicit deliverable, (2) append per-file/section (not batched), (3) scope bounded so reads don't exhaust budget. Truncated mid-sentence with no report file → spawn narrower scope, don't retry same prompt.
- **After context compaction, re-verify all prior phase outcomes before continuing.** Summaries describe intent, not environment state (git index, filesystem, processes). On resume, FIRST audit: git status, re-read modified files, verify filesystem. Every "completed" claim is an untested hypothesis until evidence confirms.
- **OOM/memory: check row count before row size.** Triage: (1) Unbounded query — no DB filter for trigger? Push filter to DB; eliminates OOM. (2) Large rows? Projection reduces proportionally. Row reduction > projection in ROI.
- **Assert the outcome your system OWNS, never the intermediate state your INFRASTRUCTURE owns.** When testing anything asynchronous (queue/broker delivery, retries, background jobs, caches, replication), assert the final business/entity state. NEVER assert the delivery bookkeeping — consume/send status, attempt counts, last-error, row existence or counts in a broker, scheduler, or outbox/inbox table. That bookkeeping lives in shared infrastructure that ANY co-running process (a peer worker, a second replica, a leftover local container) can write, usually under a deterministic shared key, so the assertion silently tests the developer's environment instead of the system: green when run alone, flaky the instant anything else shares that broker + database. Gate question for every assertion: "would this hold no matter WHICH process did the work?" — if no, assert the converged data state instead. Corollary: process-local fault injection and in-process telemetry cannot gate work any process may perform — use them as stress amplifiers (arm → bounded window → disarm → assert convergence), never as preconditions.
- **Keep domain concepts out of generic/shared/infrastructure layers.** Reusable layer (shared library, framework, infra module) must reference NO consumer-specific domain concept — tenant/customer/product IDs, business entities, feature rules. Leak compiles + runs → passes review silently while coupling the "reusable" layer to one consumer. Keep shared type domain-free; push domain fields/logic down into the consumer via subclass/composition. — why: a layer coupled to one consumer's domain is no longer reusable.

<!-- CODEX:SYNC-PROMPT-PROTOCOLS:END -->