| name | changes-review-loop |
| description | [Code Quality] Use when you need to combine /changes-review + /fix in a recursive loop — each round runs /changes-review (report-only) to surface validated findings over a fixed diff scope, then /fix to resolve them, then loops again with a FRESH full /changes-review over the CHANGED diff until one complete pass produces zero validated findings (nothing left to fix). |
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 (No Hooks)
Codex uses static project-reference loading instead of runtime-injected project docs.
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.
[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.
Quick Summary
Goal: Drive a diff scope to a clean zero-findings pass by pairing $changes-review with $fix in a recursive loop — each round runs $changes-review INLINE in report-only mode to surface validated findings over a fixed diff scope, then $fix to resolve them at the owning layer, then loops again with a FRESH full $changes-review over the CHANGED diff — stopping only when a complete $changes-review pass produces zero validated findings (nothing left to fix).
Summary:
- Each round =
$changes-review (report-only) + validate + $fix — the loop runs $changes-review as a review-PRODUCER that stops after its raw findings report (the documented $workflow-review-changes boundary: "stop after the report; parent step 2 owns validation", changes-review/SKILL.md:52,204); the loop then owns the validation gate ($why-review --validate-findings) and the dedicated $fix half, so $changes-review never self-validates or self-fixes; one without the other never converges.
- Steps (in order): (0) resolve diff scope + Goal Contract → (0b) bind the convergence loop (protocol loop primary + optional
/goal accelerator) → (1) round loop { run $changes-review report-only INLINE → $why-review --validate-findings on the report → run $fix on the VALIDATED findings at the owning layer → log iteration } → (2) converge on a zero-findings fresh review OR escalate on non-progress → (3) terminal $docs-update + recap.
- Convergence: stop ONLY when a fresh full
$changes-review over the CURRENT (post-fix) diff returns zero validated findings — not a stale clean report predating the last fix.
- Inline invariant: run
$changes-review and $why-review via the skill invocation, NEVER the spawn_agent tool — they self-bind their OWN review-loop obligations (and a session /goal gate WHEN available, changes-review/SKILL.md:197-223, why-review/SKILL.md:57-76), which a sub-agent cannot own or carry back to this loop. $changes-review's own dimensional reviewers (its Phase 0.7 sub-agents) stay sub-agents by its design, so context stays bounded.
- Apply ONLY validated findings: every finding is validated to the ≥85% survival bar via
$why-review --validate-findings before $fix touches it; the loop applies THOSE at the lowest owning layer (Entity > Service > Handler), routed by change type — NEVER unvalidated findings.
- Scope base is FIXED; the working tree grows. Recompute the diff scope each round (
branch-diff base ∪ current uncommitted changes) so convergence is measured against a stable subject as fixes accumulate.
- Bounded: round cap default 5; findings not shrinking across 2 rounds, or cap hit with findings still open → STOP & escalate by asking the user directly. Increasing findings → STOP (fixes regressing).
Why this skill exists (READ FIRST — it is the whole justification): standalone $changes-review COUPLES find + fix inside one skill invocation (its Phase 7 self-fix and Phase -1 self-recursive loop, changes-review/SKILL.md:53,197-223) — the review and the fix share one context, one lens, and one confirmation bias. This loop decouples them: it runs $changes-review purely as a finder (report-only, the documented $workflow-review-changes boundary where it stops after the report and the caller owns fixing, changes-review/SKILL.md:29,204,558,578), validates the findings, then hands them to a dedicated $fix half with its own intelligent routing at the lowest owning layer, and then re-runs a fresh full $changes-review over the changed diff. The value is the clean finder/fixer split plus the fresh-full re-review each round: it catches fix-induced regressions the coupled inner loop can rationalize away, and it lets $fix own the fix mechanics instead of the reviewer patching its own findings. Without this outer loop, "changes-review found issues, then something fixed them" ships those fixes without an independent fresh review.
Workflow: resolve diff scope + Goal Contract → bind the convergence loop (protocol loop + optional /goal accelerator) → round loop { run $changes-review report-only INLINE → $why-review --validate-findings → run $fix on the validated findings at owning layer → log iteration } → converge when a fresh full review yields zero validated findings → terminal $docs-update + recap.
Key Rules:
- Each round pairs
$changes-review (find) + $fix (resolve). The loop runs $changes-review in report-only mode so it produces findings but does NOT self-validate or self-fix; the loop's $why-review --validate-findings gate and $fix are the halves that validate and land the change. A round is incomplete until BOTH have run (or the review returned zero findings).
- MUST run INLINE in the main session — NEVER dispatch
$changes-review or $why-review as a sub-agent. They self-bind their own review-loop obligations (and a session /goal gate when available, changes-review/SKILL.md:197-223); as a sub-agent that in-session guarantee is silently lost. This loop skill therefore also runs inline. ($changes-review's internal Phase 0.7 dimensional reviewers remain sub-agents by its own design — that is bounded and correct.)
- Report-only invocation is mandatory. Tell
$changes-review to run as a review producer: do its full dimensional review + Phase 6 $why-review --validate-findings gate, then STOP before Phase 7 self-fix / Phase 7.5 holistic / Phase 8 docs-update. The loop owns fixing (Step 1.3) and the terminal docs-update (Step 3). If a $changes-review invocation cannot be constrained to report-only in this environment and self-fixes anyway, fall back to detecting fixes-applied per round (like $workflow-review-changes-loop) and skip the redundant $fix half for that round — never double-fix.
- Convergence = a fresh full
$changes-review over the post-fix diff returns zero validated findings. A clean report produced BEFORE the latest fix landed does NOT count — re-review the changed diff.
$fix applies ONLY validated findings, at the lowest owning layer, routed by change type (code → $fix with its --target intelligent routing, or a direct edit at Entity/Service; spec-drift → $spec [update] + $spec [mode=tests]; docs → $docs-update; missing coverage → $integration-test; honor each finding's dual-feedback ledger, changes-review/SKILL.md:530). NEVER apply an unvalidated or demoted finding.
- The diff scope base is FIXED across rounds; its content changes as fixes land. Recompute
{scope} = branch-diff base ∪ current uncommitted changes each round so the base merge-base never moves and convergence is measured against a stable subject.
- Round cap (default 5) and findings-not-shrinking / increasing → STOP & escalate by asking the user directly. NEVER loop open-ended.
First Principle — Convergence, Not Motion
A round that changes the diff is progress only if the next fresh review finds fewer things to fix.
The loop exists to reach a fixed point (zero findings), not to keep editing the code.
If findings stop shrinking, that is a signal to escalate, not to spin another round.
Step 0 — Resolve Diff Scope + Goal Contract (FIRST ACTION)
- Parse the review scope from the user prompt into a stable, reusable scope string — exactly the diff kinds
$changes-review resolves (changes-review/SKILL.md:96-104). It has two parts UNIONed:
- Branch-diff base — a branch-to-branch or PR diff (e.g.
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. If the prompt names a commit range, capture it the same way.
- 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. NEVER silently convert the diff source type.
- 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 fresh full $changes-review over {scope} returns zero validated findings (no finding of any severity survives the loop's $why-review --validate-findings gate).
Record the round cap (default 5) 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 $changes-review report-only INLINE over {scope} (recomputed each round). After each review, validate its findings with $why-review --validate-findings, apply every VALIDATED finding's fix via $fix at its owning layer, then re-run a FRESH full $changes-review over the CHANGED diff. Do NOT stop while the last review still produced validated findings. Converge ONLY when a fresh full $changes-review returns zero validated findings. Cap at {N=5} rounds; if findings do not shrink across 2 consecutive rounds, findings increase, or the cap is hit with findings still open → 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 fix cycle. 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 changes-review convergence loop: repeatedly run $changes-review report-only INLINE over {scope} (recomputed each round). After each review, validate its findings with $why-review --validate-findings, apply every VALIDATED finding's fix via $fix at its owning layer, then re-run a FRESH full $changes-review over the CHANGED diff. If validated findings>0 → apply fixes and run another round; if a fresh full $changes-review returns zero validated findings → CONVERGED, run the terminal $docs-update and clear the gate. Do NOT stop while the last review still produced validated findings. Cap at {N=5} rounds; if findings do not shrink across 2 consecutive rounds, findings increase, or the cap is hit with findings still open → 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 $changes-review round would self-bind its OWN Phase -1 review-loop obligation — but the loop runs it in report-only mode, where Phase -1 is deferred to the caller (the $workflow-review-changes boundary, changes-review/SKILL.md:204), so no inner self-fix gate is installed; THIS outer loop owns the single convergence gate. Each inner $why-review --validate-findings self-clears when that round's findings are all adjudicated. All self-clear on satisfaction — no orphaned gate. Do NOT tell the user to clear either.
Step 1 — Round Loop ($changes-review → validate → $fix → log)
Each round couples the two halves — review to find, fix to resolve. For each round R (starting at 1), do ALL of:
- Snapshot before: record the working-tree fingerprint —
git status --porcelain + git diff --stat. This is the fixes-applied baseline for the round and the objective backstop for convergence detection (Step 2).
- Run
$changes-review report-only INLINE on the recomputed {scope} via the skill invocation (NEVER the spawn_agent tool). Direct it to run as a review PRODUCER: perform its full dimensional review and STOP after writing its findings report — do NOT run its Phase 6 validation, Phase 7 self-fix, Phase 7.5 holistic, or Phase 8 docs-update. This is exactly the $workflow-review-changes boundary where the caller (this loop) owns validation and fixing, so $changes-review hands the raw report back and stops (changes-review/SKILL.md:52,204,558,578). If the environment cannot constrain $changes-review to report-only and it self-fixes anyway, treat that self-fix as this round's fix half (detect fixes-applied vs the before-snapshot) and skip Steps 1.3–1.4 to avoid double-fixing.
- Validate the findings — THE LOOP OWNS THIS GATE. Run
$why-review --validate-findings <report-path> INLINE via the skill invocation over the report $changes-review just produced (plans/reports/code-review-*.md), so every finding is confirmed correct, proof-backed, reasonable, and best-practice (≥85% survival bar) before any fix. This is the parent-owned validation the report-only boundary defers to the caller (changes-review/SKILL.md:52). If the validated finding set is empty → this round converged; go to Step 2 (no fix half needed).
- Run
$fix on the validated findings (findings>0 only) — this is the half the loop owns. Resolve each validated finding at its owning layer: code → $fix (its --target intelligent routing) or a direct edit at the lowest layer (Entity > Service > Handler); spec-drift (CODE-WRONG/SPEC-STALE/SPEC-SILENT) → route per the finding's Spec Drift Adjudication verdict ($spec [update] + $spec [mode=tests]); missing coverage (GAP/SPEC-GAP) → $integration-test; docs → defer to the terminal $docs-update; behavior-changing → honor the finding's Dual-Feedback Ledger (spec verdict + test action, changes-review/SKILL.md:530). Fix ONLY validated findings — never an unvalidated or demoted one.
- Append an Iteration Log entry to the Goal Contract: round number, validated findings count, files/artifacts changed this round (
file:line), fixes applied, and remaining gaps.
Step 2 — Convergence & Escalation Gate
Evaluate after every round:
| Condition | Action |
|---|
Fresh full $changes-review returned zero validated findings AND the working tree is unchanged by that final review pass | CONVERGED → mark the required criterion PASS in the Goal Satisfaction matrix → clear the /goal gate → go to Step 3. |
Validated findings > 0 AND round < N AND findings shrank vs prior round | Apply the validated fixes (Step 1.4), recompute {scope}, then run round R+1 (fresh full re-review of the changed diff). |
| 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 cap N hit with findings still open | STOP & escalate by asking the user directly — report the still-open findings; do not silently continue. |
Increasing findings = STOP. If round R surfaces MORE validated findings than round R-1, the fixes are regressing the code — STOP and escalate immediately (mirrors changes-review/SKILL.md's regression stance). Never trade one fix for two new findings across rounds.
Step 3 — Terminal Docs-Update + Recap
- Terminal
$docs-update (MANDATORY once converged). The loop ran $changes-review in report-only mode, so its unconditional Phase 8 docs-update (changes-review/SKILL.md:55,68) was deferred to the caller. Once a fresh full $changes-review yields zero validated findings, invoke $docs-update INLINE over the full changeset as the terminal step so no stale docs survive.
- Recap. Emit a concise convergence recap: rounds run, validated findings per round (the shrinking sequence), the fixes applied at each round, the final zero-findings evidence, and the Goal Satisfaction matrix (required criterion PASS). Point to each round's
$changes-review report under plans/reports/ and the Goal Contract Iteration Log. Do NOT commit or push unless the user explicitly asks.
Convergence Detection — Why a Fresh Full Re-Review Is Required
A round converges ONLY when a $changes-review that ran over the current, post-fix diff returns zero validated findings. Both properties are required because:
- Fresh over the changed diff — a clean report from a review that predates the last fix proves nothing about the fix. Every applied fix invalidates the prior verdict (
changes-review/SKILL.md:72); the loop MUST re-review after fixing, never reuse a stale clean report.
- Zero validated findings — the loop's
$why-review --validate-findings gate already dropped inflated/unproven findings below the ≥85% bar; convergence rides on that validated set, so the loop never chases a nit the review itself would demote.
- Working-tree unchanged backstop — the objective
git-diff comparison (Step 1.1 snapshot) confirms the converging review actually landed no fix; a "clean" verdict that still mutated files means the round DID fix things → run another round to re-prove clean.
When validated findings remain but cannot be fixed (owner/product input needed) → escalate, do not loop. When a fix lands but the next fresh review still finds issues → run another round. Convergence is a fixed point, not a single clean read.
IMPORTANT MANDATORY sequence: Step 0 (resolve diff scope + Goal Contract) → Step 0b (bind the convergence loop: protocol loop primary + optional /goal accelerator) → Step 1 (round loop: run $changes-review report-only INLINE → $why-review --validate-findings → run $fix on validated findings → log) → Step 2 (converge on a zero-findings fresh review / escalate on non-progress) → Step 3 (terminal $docs-update + recap).
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.
- 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).
- Required sections: Original Request, Purpose, Success Criteria (checkboxes; mark required vs optional), Constraints, Evidence Required, Iteration Log, Goal Satisfaction matrix.
- Before work: read the active goal and map planned work to saved success criteria — execution serves the saved criteria, never chat memory alone.
- After execution/verification: append an Iteration Log entry — result, evidence references (
file:line, command output, report path), remaining gaps.
- Review gate: emit a Goal Satisfaction matrix —
| Success Criterion | Evidence | Status | with PASS/FAIL/BLOCKED. Overall PASS requires every required criterion PASS.
- 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.
- 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.
- 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.
- 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.
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.
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. Before changing a constant, limit, flag, wording, or pattern, read nearby context and history.
Surface ambiguity before acting — don't pick silently. Multiple valid interpretations require an explicit question or stated assumption with risk.
Keep shared guidance role-relevant. Universal guidance must help every receiving skill or agent; code-specific obligations belong only in code-specific protocols.
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.
- 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.
- 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.
- 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?".
- 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.
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.
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.
- 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.
Closing Reminders
IMPORTANT MUST ATTENTION Goal: Pair $changes-review + $fix in a recursive loop over a fixed diff scope — $changes-review (report-only) to find validated findings → $fix to resolve them → fresh full re-review of the CHANGED diff — until a complete $changes-review pass produces zero validated findings (nothing left to fix).
IMPORTANT MUST ATTENTION main steps (in order): (0) resolve diff scope + Goal Contract → (0b) bind the convergence loop (protocol loop primary + optional /goal accelerator) → (1) round loop: run $changes-review report-only INLINE → $why-review --validate-findings → run $fix on VALIDATED findings at owning layer → append Iteration Log → (2) converge on a zero-findings fresh review / escalate on non-progress → (3) terminal $docs-update + 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 each round pairs $changes-review (find) + $fix (resolve); the loop runs $changes-review in report-only mode (changes-review/SKILL.md:29,204) so it never self-fixes, and the loop MUST run the $fix half — one without the other never converges.
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 $changes-review and $why-review INLINE via the skill invocation — NEVER as a sub-agent (they self-bind their own review-loop obligations + a session /goal gate when available, changes-review/SKILL.md:197-223). $changes-review's internal Phase 0.7 dimensional reviewers remain sub-agents by its own design.
IMPORTANT MUST ATTENTION convergence = a fresh full $changes-review over the post-fix diff returns zero validated findings (with the working-tree-unchanged backstop) — never a stale clean report predating the last fix.
IMPORTANT MUST ATTENTION apply ONLY validated findings (≥85% survival bar via $why-review --validate-findings) at the lowest owning layer (Entity > Service > Handler); NEVER apply an unvalidated or demoted finding.
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 5); findings not shrinking across 2 rounds, increasing, or cap hit with findings still open → STOP & escalate by asking the user directly. NEVER loop open-ended.
IMPORTANT MUST ATTENTION run the terminal $docs-update once the loop converges (deferred from $changes-review's report-only run); do NOT commit or push unless the user explicitly asks.
Hookless 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
[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. 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. Any supported AI tool may execute when this shared context and local docs are available.
- 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.
- ANALYZE: Choose the best option: execute directly, invoke a skill, activate a standard workflow, or compose a custom step combination.
- AUTO-SELECT: Pick the best option yourself. Do not ask the user to choose between direct execution, skill, standard workflow, or custom workflow.
- 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.
- CREATE TASKS: task tracking for ALL workflow/skill/custom steps before execution when the selected path has multiple steps.
- 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.
- Keep reusable AI-SDD principles in
.claude; put repository-specific paths, commands, owners, products, and formats in project config/reference docs.
- Preserve cycle:
spec -> plan -> tasks -> implement -> verify -> update spec/docs.
- Trace every requirement or invariant through decision, task, TC/test, source evidence, and docs/spec update.
- Treat code-to-spec extraction as reference-only until accepted by the canonical spec owner.
- Any supported AI tool may plan, implement, review, or verify with synced context; using multiple tools is optional.
- 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
- 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:
- Name the FAILURE MODE (reasoning/assumption failure), not symptom — "assumed API existed without reading source" not "used wrong enum value".
- Generality test: does this failure mode apply to ≥3 contexts/codebases? If not, abstract one level up.
- Write as a universal rule — strip project-specific names/paths/classes. Useful on any codebase.
- Consolidate: multiple mistakes sharing one failure mode → ONE lesson.
- Recurrence gate: "Would this recur in future session WITHOUT this reminder?" — No → skip
$learn.
- Auto-fix gate: "Could
$code-review/$code-simplifier/$security-review/$lint catch this?" — Yes → improve review skill instead.
- 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 → adjudicate WHO is at fault (source vs test) before forcing green. A green-again suite is not the goal; the correct verdict on what was actually wrong is. Root-cause first, then triangulate the failure against the governing spec (
docs/specs/** if one exists) AND the source: SOURCE-WRONG → fix code at the owning layer and keep/strengthen the test; TEST-WRONG → fix the stale assertion/setup at its root. NEVER weaken an assertion, add a skip, or relax a timeout to force green, and never change source to satisfy a broken test. Spec silent or ambiguous about which side is correct → STOP and ask the user.
- 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. Pattern-matching as "wrong" skips context. Before changing any constant/limit/flag: read comments, git blame, surrounding code.
- 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.
- 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.