Agent Skills: workflow-review-changes

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:

• Backend/CQRS/API/domain/entity changes: backend-patterns-reference.md, domain-entities-reference.md, project-structure-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: Ensure changed work reaches clean review through validated findings, verified fixes, full re-review, and synchronized docs/tests — review all uncommitted changes, validate findings, fix only validated findings, then re-run $review-changes INLINE (only when $plan-execute actually changed files), repeating the plan→plan-execute→review-changes loop until a complete pass is clean.

Summary:

• Step 1 $review-changes runs FIRST and owns the baseline (surface analysis, integration-test/translation-sync gaps, UI review via internal $review-ui); step 2 $why-review validates those findings to drop false positives BEFORE the parallel batch fires — so steps 3–7 act only on warranted findings.
• Steps 3–7 ($review-architecture, $review-domain-entities [if entity files], $performance-review, $integration-test-review, $security-review) are read-only sub-agents: spawn ALL in ONE message and advance ONLY after every member returns (all-return barrier); the mutating $code-simplifier (step 8) waits until the barrier clears and self-reviews its own changes via $code-review.
• Fix cycle (steps 9–12 $plan→$plan-review→$plan-execute→$review-changes) runs ONLY when validated findings exist; the step-12 re-review runs ONLY if $plan-execute changed files, re-reading the full diff from scratch INLINE to counter orchestrator confirmation bias, and loops until a clean zero-finding pass (cap at 3 no-progress repeats of the same blocker → escalate via a direct user question).
• $docs-update (step 13) ALWAYS runs and triages internally; SPEC-STALE drift verdicts from step 1 flow here to update the Feature Spec first — the workflow is NOT clean while any behavior-vs-spec divergence stays unadjudicated (green tests do not normalize drift).

Sequence: $review-changes (owns UI review — invokes $review-ui internally when frontend changes) → $why-review (validate findings) → [parallel batch] $review-architecture + $review-domain-entities (if entity changes) + $performance-review + $integration-test-review + $security-review → $code-simplifier (self-reviews its own changes via $code-review) → $plan → $plan-review → $plan-execute → /review-changes (conditional inline re-review — only if $plan-execute changed files; loops $plan→/plan-execute→/review-changes until clean) → $docs-update → $workflow-end → $watzup

Key Rules:

• MUST ATTENTION define success criteria before execution and loop until observable verification passes.

• MUST ATTENTION when creating/reviewing specs or tests, name Business Intent / Invariant Guarded or the protected business intent/invariant and ensure the test would fail if that intent breaks.

• MUST ATTENTION carry every unresolved finding or unaccepted risk into validation/fix planning; do not close until fixed or explicitly accepted.

• MUST ATTENTION include unresolved risk register, generated mirror drift, and spec/test/docs drift in the fresh review prompt when relevant.

• MUST ATTENTION run $why-review at step 2 to validate the $review-changes findings BEFORE spawning the parallel reviewers — drop false positives early so the batch and fix cycle act only on warranted findings.

• After $plan-execute applies validated fixes (and ONLY if $plan-execute changed files) → re-run $review-changes INLINE over the current full diff from the first phase; re-read the diff from scratch to counter orchestrator confirmation bias

• Main-agent re-review (with knowledge of its own fixes) is NOT sufficient — orchestrator-level confirmation bias

• PASS = one complete review pass finds zero blocking issues after all validated fixes and verification are included

• Repeated blockers are tracked in conversation context; stop after 3 no-progress full invocations of the same blocker

First Principle — Easy to Change

The success metric of every coding decision is future change cost. DRY, SRP, abstraction, design patterns, naming, layering, tests — every technique exists to serve one goal: making the next change cheaper.

When evaluating code, a refactor, a test, or an abstraction, ask: does this make the next change cheaper or more expensive?

• Reject "best practices" that raise change cost (premature abstraction, speculative generality, leaky indirection, ceremony without payoff).
• Name the real enemies in findings: coupling, hidden state, duplicated knowledge, unclear intent, irreversible decisions exposed too early.
• A simpler design that is easy to change beats a sophisticated design that isn't.

Apply this lens before invoking any specific rule, pattern, or checklist below — if a downstream rule would raise change cost, this principle wins.

Mandatory Task Creation (ZERO TOLERANCE)

Create one task per row in the table below — source of truth is workflows.json → review-changes.sequence (currently 15 steps; verify count matches if you suspect drift):

| # | Task Subject | Conditional? | | --- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | 1 | [Workflow] $review-changes — Surface detection + dimensional review tasks (BE/FE/SCSS/Synthesis/General) + UI dimension via $review-ui (if frontend changes) + integration test sync check + multilingual translation sync check | No | | 2 | [Workflow] $why-review — Validate the $review-changes findings before parallel reviewers run (each finding warranted, evidence-backed, not a false positive) | No — FINDINGS-VALIDATION gate over the $review-changes findings; the fix plan's design is reviewed by $plan-review (step 10); if step 1 found zero issues, pass through with nothing to validate | | 3 | [Workflow] $review-architecture — Architecture compliance review ⚡ PARALLEL BATCH | No — run as sub-agent in parallel with steps 4/5/6/7 | | 4 | [Workflow] $review-domain-entities — DDD quality review of changed domain entity files ⚡ PARALLEL BATCH | Yes — skip if no domain entity files (Domain/, Entities/, ValueObjects/) in git diff | | 5 | [Workflow] $performance-review — Performance analysis ⚡ PARALLEL BATCH | No — run as sub-agent in parallel with steps 3/4/6/7 | | 6 | [Workflow] $integration-test-review — 7-gate test quality review + Gate 7 change coverage (every behavior change → covering test + spec TC) ⚡ PARALLEL BATCH | No — run as sub-agent in parallel with steps 3/4/5/7 | | 7 | [Workflow] $security-review — Security vulnerability review ⚡ PARALLEL BATCH | No — run as sub-agent in parallel with steps 3/4/5/6 | | 8 | [Workflow] $code-simplifier — Simplify and refine code (self-reviews its own changes via $code-review before returning) | No — runs AFTER parallel batch (modifies code; batch reviews pre-simplification state; simplifier owns review of its own output) | | 9 | [Workflow] $plan — Consolidate validated review findings into fix plan | Conditional — run ONLY if reviews surfaced validated findings to fix; skip if all reviews PASS | | 10 | [Workflow] $plan-review — Architecture/design review of fix plan (includes adversarial design-rationale pass + internal $why-review --validate-findings of its own findings) | Conditional — run ONLY if there is a fix plan (i.e. findings exist); skip if all reviews PASS | | 11 | [Workflow] $plan-execute — Implement fixes from plan | Conditional — run ONLY if there are validated findings to fix; skip if all reviews PASS | | 12 | [Workflow] $review-changes — Conditional inline re-review after $plan-execute (re-runs the review over the current diff); loop $plan→/plan-execute→/review-changes until clean | Skip if all reviews PASS, OR if $plan-execute applied no file changes | | 13 | [Workflow] $docs-update — Update impacted documentation | Always run — $docs-update triages internally (fast-exits when only config/tool files changed) | | 14 | [Workflow] $workflow-end — End workflow state (prints the concise change recap, then clears state) | No | | 15 | [Workflow] $watzup — Post-workflow summary and final $understand handoff | No |

UI review is owned by step 1. $review-ui is NOT a separate workflow step — $review-changes (step 1) invokes it internally (ui-ux-designer sub-agent) as its UI dimension whenever the diff contains frontend/UI files. Do NOT create a separate [Workflow] $review-ui task.

NEVER consolidate, rename, or omit steps. If reviews PASS, mark conditional tasks completed with note "Skipped — all reviews passed".

Integration Test Sync: The $review-changes skill (task #1) includes a mandatory integration test coverage check for changed command/query/handler files. When gaps are found, the skill uses a direct user question to surface them — NOT purely advisory. The user must explicitly choose to run $integration-test or confirm tests are already written. No silent skip.

Translation Sync: The $review-changes skill (task #1) includes a mandatory multilingual UI translation-sync check. When UI text changes in multilingual projects without locale updates, the skill uses a direct user question for an explicit user decision — NOT purely advisory.

Docs Update: $docs-update MUST run after EVERY review — it performs Phase 0 triage and fast-exits automatically when only non-business-code files changed (.claude/**, config). When business code is in the changeset, it WILL invoke: Phase 2 $spec (business feature doc update), Phase 2.5 $spec-index [mode=index] (derived bucket INDEX/ERD refresh — if docs/specs/ bucket maintains a derived index; note: dirs may be app buckets or flat system folders — probe ls docs/specs/{name}/ to find a specific service), Phase 3 $spec [mode=tests] (test spec sync), Phase 4 $spec [mode=sync] (§8 TCs ↔ integration test code). Never skip based on review PASS status alone.

Spec Drift Adjudication: The $review-changes skill (task #1) runs a mandatory spec-drift adjudication (SYNC:spec-drift-adjudication, per shared/sdd-artifact-contract.md → Drift Gates) for every behavior-changing file: it classifies each divergence between changed behavior and the canonical Feature Spec as CODE-WRONG (BLOCKING — fix the code/test against intended behavior), SPEC-STALE (the change is the new intent — the spec documents the old behavior), or AMBIGUOUS (escalate). The reviewer never silently picks a side. A SPEC-STALE verdict flows downstream: $docs-update (step 13) updates the Feature Spec FIRST via $spec [update], then re-syncs $spec [mode=tests]. The workflow is NOT clean while any behavior-vs-spec divergence remains unadjudicated — green tests do not normalize drift (green can encode the drift itself).

Spec enrichment per cycle (MANDATORY — closes the feedback loop): Every confirmed finding fixed in the loop (steps 9–12) that changed observable behavior MUST produce a new or updated §8 regression/preservation TC via $spec [mode=tests] before the workflow is clean — a code-only fix with no covering §8 TC is an INCOMPLETE cycle, not a clean pass. This applies to EVERY confirmed behavior-changing fix, not only SPEC-STALE drift verdicts or bugfix-workflow paths: a CODE-WRONG fix owes a regression TC describing the now-correct behavior; a behavior change owes a preservation/regression TC guarding the new behavior. So each recursive cycle ENRICHES the spec rather than only mutating code — the inline re-review (step 12) and the $workflow-end spec ↔ TDD-test sync gate both treat a behavior-changing fix that left no §8 TC as an open finding.

Parallel Review Phase (Steps 3–7) — EXECUTION PROTOCOL

Note: Steps 3–7 are ARCHITECTURAL/SECURITY reviewers (architecture compliance, DDD entities, performance, integration test quality, security vulnerabilities). They are separate from the DIMENSIONAL review (BE/FE/SCSS/Synthesis + UI via $review-ui) that runs inside Step 1 ($review-changes). Both operate in parallel — Steps 3–7 as explicit workflow parallel sub-agents; dimensional agents (including the UI dimension) inside Step 1 as its internal parallel batch. No overlap in responsibility. UI/frontend quality is NOT a step 3–7 reviewer — $review-changes (step 1) owns it and invokes $review-ui internally (ui-ux-designer sub-agent) only when the diff has files matching the project's configured frontend/UI file patterns.

Steps 3–7 ($review-architecture, $review-domain-entities, $performance-review, $integration-test-review, $security-review) are read-only and independent — no shared mutable state, no ordering dependency between them. Run them as parallel sub-agents to preserve main session context budget and reduce wall-clock time.

Why parallel?

Each reviewer reads the git diff independently and analyzes one concern. Sequential execution would burn 50K+ tokens in the main session absorbing all five inline. The stepMeta in workflows.json marks all five as executionMode: subagent, contextBudget: high — dispatch each as a sub-agent per the model-driven advancement rule (no hook emits a 💡 [SUB-AGENT RECOMMENDED] hint).

UI review runs inside step 1, not here. $review-changes invokes $review-ui (ui-ux-designer sub-agent) as part of its own internal dimensional batch when frontend files changed — do NOT spawn a separate review-ui agent in this parallel phase.

Execution: spawn in one message

After steps 1 and 2 ($review-changes and $why-review) complete, spawn all active parallel reviewers in a single response with multiple spawn_agent tool calls:

spawn_agent(review-architecture, agent_type="architect", ...)           ← all in ONE message
spawn_agent(review-domain-entities, agent_type="code-reviewer", ...)    ← only if entity files in diff
spawn_agent(performance-review, agent_type="performance-optimizer", ...)
spawn_agent(integration-test-review, agent_type="integration-tester", ...)
spawn_agent(security-review, agent_type="security-auditor", ...)

Each sub-agent receives:

• The baseline summary from step 1 (what changed, integration test gaps found)
• Instruction to write report to plans/reports/{skill}-{date}-{slug}.md
• Full review protocols per SYNC:review-protocol-injection (verbatim in prompt — never by file reference)

State advancement after parallel batch (model-driven — PRIMARY)

Advancement here is model-driven — your responsibility against the task list, NOT a hook/tool signal. This is the same rule the universal context files carry ("Workflow Step Advancement & Parallel Phases" in CLAUDE.md / AGENTS.md), so the batch advances identically under Claude and Codex. The shared kernel is the canonical SYNC:parallel-phase-advancement block consolidated at the end of this skill — its barrier rule governs this batch: declare the group up-front; spawn ALL members in ONE message; advance ONLY after EVERY member returns (a skipped conditional member counts as "returned"); a sub-agent return advances a step IDENTICALLY to an inline call; defer the mutating $code-simplifier step until the barrier clears; hooks are accelerators only.

Applied to this workflow's batch — after ALL parallel reviewers (steps 3–7) have returned:

1. TaskUpdate step 3 → completed
2. TaskUpdate step 4 → completed (or "Skipped — no entity files" if the conditional review-domain-entities member did not run — a skipped conditional counts as "returned")
3. TaskUpdate step 5 → completed
4. TaskUpdate step 6 → completed
5. TaskUpdate step 7 → completed
6. Read all sub-agent report files; synthesize findings into a combined review summary
7. Proceed to step 8 ($code-simplifier) sequentially — only after the barrier above (it is a code-mutating step and must see the complete review snapshot)

Advancement here is model-driven. This sub-agent batch advances only after every member returns (the all-return barrier) — no step-tracking hook advances it. Claude and Codex both rely entirely on this rule.

Consolidation before $code-simplifier

Before running $code-simplifier, synthesize all parallel sub-agent findings:

• List all Critical/High/Medium/Low findings across all 5 reports (plus the UI-dimension findings folded into step 1's report when frontend files changed)
• Note any conflicts between reviewers (same file, different concerns)
• Pass this summary to $code-simplifier as context so simplification is informed by review findings

Surface Analysis from Step 1:

Step 1 ($review-changes) now emits a surface analysis summary in its report:

## Change Surface Analysis
BE files: {N}
FE-Logic files: {M}
SCSS files: {P}
Review Mode: [DIMENSIONAL | BE-ONLY | FE-ONLY | FE-SPLIT | TOOLING]

Include this surface analysis in the consolidation summary passed to $code-simplifier. This lets the simplifier focus attention on the dominant surface without re-analyzing the diff.

Dimensional agent reports (if mode = DIMENSIONAL):

• plans/reports/review-be-{date}.md — BE findings
• plans/reports/review-fe-logic-{date}.md — FE-Logic findings
• plans/reports/review-scss-{date}.md — SCSS findings (if spawned)
• plans/reports/synthesis-review-{date}.md — Cross-boundary findings

All four (plus the UI-dimension $review-ui findings when frontend files changed) feed into the consolidation summary alongside steps 3–7 architectural findings.

What runs sequentially (never parallelize)

| Step | Why sequential | | ----------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | review-changes (#1) | Establishes baseline — must run first | | why-review (#2) | Validates the review-changes findings before the batch — gates which findings the batch and fix cycle act on | | code-simplifier (#8) | Modifies code — batch reviews pre-simplification state; self-reviews its own output via $code-review before returning | | plan → plan-review → plan-execute (#9–11) | Ordered validated fix-plan cycle — $plan consumes already-validated review findings; $plan-review reviews the fix plan's design (adversarial rationale pass + internal $why-review --validate-findings of its own findings) before $plan-execute implements it |

Conditional Inline Re-Review Protocol (CRITICAL)

Decision Logic

Reviews (steps 1-8) → ALL PASS (no findings)?
  YES → skip steps 9-12 ($plan//plan-review//plan-execute//review-changes), proceed to $docs-update (step 13) → $workflow-end → $watzup → DONE
  NO (findings exist) → $plan → $plan-review → $plan-execute → (if $plan-execute changed files) $review-changes INLINE re-review (step 12) → loop until clean
Note: $code-simplifier (step 8) self-reviews the code it changes via $code-review before returning — there is no separate workflow-level code-review step.
Note: $why-review runs ONCE (step 2) as a FINDINGS-VALIDATION gate over the $review-changes findings before the parallel batch. The fix plan's design rationale is reviewed by $plan-review (step 10) — which applies its own adversarial rationale pass and self-invokes $why-review --validate-findings on its own findings — so no separate post-plan-review $why-review step is needed.

Conditional Inline Re-Review Gate (Step 12) — After $plan-execute Applies Fixes

1. CONDITION (run only if $plan-execute changed files): Step 12 runs ONLY when $plan-execute actually modified files (validated fixes were applied). If $plan-execute made no file changes — nothing was wrong, or the plan resolved to no-ops — SKIP step 12 entirely and proceed to $docs-update.
2. DO re-run the $review-changes protocol INLINE in the main session over the current full diff. Create a fresh task breakdown, rerun blast radius, risk detection, surface categorization, diff collection, dimensional reviews, synthesis, and validation gates. (Inline by design for this workflow — cheaper than spawning a fresh sub-agent; accept the mild orchestrator-confirmation-bias tradeoff, and counter it by re-reading the diff from scratch.)
3. DO track re-review invocation count and repeated blockers in conversation context
4. DO integrate the inline $review-changes findings — MUST NOT filter, reinterpret, or override
5. IF the inline re-review returns PASS with zero findings → first confirm every confirmed behavior-changing fix applied in this loop produced a new/updated §8 regression/preservation TC via $spec [mode=tests] (per "Spec enrichment per cycle" above); a behavior-changing fix that left no covering §8 TC is an OPEN finding — re-enter the loop to add it before proceeding. Only once spec enrichment is complete → proceed through $docs-update → $workflow-end → $watzup → DONE
6. IF the inline re-review returns FAIL and the same blocker has not repeated 3 times → validate findings, run $plan + $plan-execute again, then re-run $review-changes (step 12)
7. IF the same validated blocker repeats across 3 invocations with no observable progress → STOP and escalate via a direct user question — do NOT silently loop or fall back to any prior protocol

Iteration Tracking (Conversation-Scoped)

Iteration count is tracked in conversation context only — no persistent files. Each new conversation starts fresh at round 0.

Rules:

• Repeated blocker cap — if the same validated finding repeats for 3 full invocations with no progress, STOP and escalate via a direct user question (manual review required)
• PASS = done — if no fix cycle happened, initial clean reviews/tests are enough; if a fix cycle happened, PASS requires a complete inline $review-changes re-review pass with zero findings
• Issue count increasing — if round N finds MORE issues than round N-1, STOP and escalate via a direct user question
• Goal Satisfaction FAIL = findings exist — a required saved criterion at FAIL in the Goal Satisfaction matrix enters the SAME loop as a code finding: validate the gap is real → $plan → $plan-execute → inline re-review of the affected criteria only. Workflow end requires every required criterion PASS or BLOCKED with a user-facing escalation reason; mark criteria BLOCKED (never silently drop them) when two consecutive iterations show no criterion progress.

Goal Contract propagation (workflow-owned): At workflow start, resolve the active Goal Contract per SYNC:goal-contract-satisfaction-loop (active plan goal.md → plans/goals/{YYMMDD-HHmm}-{slug}/goal.md). Pass the same goal file reference to every child step; step 1 $review-changes emits the Goal Satisfaction matrix against the SAME saved criteria. After each fix cycle, append an Iteration Log entry to the goal file with evidence references.

Flow Diagram

Main Session: Review → Validate findings → Plan → Fix ($plan-execute) → $review-changes re-review
                  │                                          │
                  │ (no issues)                              │ (only if $plan-execute changed files;
                  ↓                                          ↓  else skip to $docs-update)
            $docs-update                        $review-changes re-runs INLINE
            $workflow-end                       over the current full diff
            $watzup                                          │
            DONE ✓                                           ↓
                                                  Report → PASS? → DONE ✓
                                                         → FAIL? → Validate
                                                                 findings → Plan → Fix
                                                                 → $review-changes re-review

IMPORTANT MANDATORY Steps: $review-changes -> $why-review -> $review-architecture -> $review-domain-entities -> $performance-review -> $integration-test-review -> $security-review -> $code-simplifier -> $plan -> $plan-review -> $plan-execute -> $review-changes -> $docs-update -> $workflow-end -> $watzup

[STEP CONDITIONS] Not every step always runs — the bare list above is the canonical order; these are the run-conditions:

• Step 4 $review-domain-entities — only if domain entity files (Domain/, Entities/, ValueObjects/) are in the diff.
• Steps 9–11 $plan → $plan-review → $plan-execute — only if reviews surfaced validated findings to fix (i.e. there are findings / code changes to make). Skip all three when steps 1–8 PASS clean.
• Step 12 $review-changes (re-review) — only if $plan-execute actually changed files; re-runs INLINE and loops $plan→$plan-execute→$review-changes until a clean pass (3-repeat blocker cap).
• Steps 1–3, 5–8, 13–15 — always run.

[BLOCKING SEQUENCING] Step 1 $review-changes is SEQUENTIAL and MUST run FIRST — it produces the baseline (surface analysis + integration-test/translation gap detection) consumed by all downstream reviewers, AND owns the UI review (invokes $review-ui internally via a ui-ux-designer sub-agent when the diff has frontend/UI files). Step 2 $why-review is SEQUENTIAL and runs immediately after — it validates the $review-changes findings (drops false positives) before any parallel reviewer spawns. Steps 3–7 ($review-architecture, $review-domain-entities, $performance-review, $integration-test-review, $security-review) form a PARALLEL BATCH — spawn all in ONE message via specialized spawn_agent tool calls (architect, code-reviewer, performance-optimizer, integration-tester, security-auditor). Step 8 $code-simplifier is SEQUENTIAL and waits until ALL parallel batch sub-agents return + consolidation summary is built; it self-reviews the code it changes via $code-review (scoped to its own changed files) before returning, so there is no separate workflow-level code-review step. Steps 9+ proceed sequentially as listed.

[WORKFLOW-IN-WORKFLOW: MUST RUN AS SUB-AGENT when inside another workflow] This skill activates the full workflow-review-changes workflow (15 steps). When invoked as a step inside a parent workflow (e.g., workflow-feature, workflow-bugfix, workflow-refactor), it MUST execute via spawn_agent tool (agent_type: "code-reviewer") — NEVER as an inline skill invocation call. Inline execution absorbs 15 steps of context into the parent session.

Sub-agent prompt must include: current git diff, feature/task description, instruction to return SYNC:subagent-return-contract summary and write full findings to plans/reports/.

Standalone invocation (not inside a workflow): inline execution is fine — no sub-agent required.

[BLOCKING] Each step MUST invoke its skill invocation — marking a task completed without skill invocation is a workflow violation. NEVER batch-complete validation gates. [CONDITIONAL INLINE RE-REVIEW] After validated fixes in $plan-execute — and ONLY if $plan-execute changed files — re-run $review-changes INLINE (step 12) over the current full diff. If $plan-execute made no changes, skip step 12. Clean review passes with zero findings end the loop; repeated blockers stop after 3 no-progress invocations. [REPEATED BLOCKER CAP] Track re-review invocations in conversation context, not persistent files. After a fix cycle, PASS = a complete inline $review-changes re-review pass finds zero findings without more fixes; stop after the same blocker repeats 3 times with no progress.

Activate the workflow-review-changes workflow. Run $start-workflow workflow-review-changes with the user's prompt as context.

Parallel-Phase Advancement (model-driven) — How to run AND advance a declared parallel batch of workflow steps. Tool-agnostic: identical under Claude and Codex — neither depends on a hook. Mirrors the universal context-file rule ("Workflow Step Advancement & Parallel Phases" in CLAUDE.md / AGENTS.md).

1. Declare the group. Name the members of the parallel phase up-front — which steps run together, and mark any conditional member with its trigger.
2. Spawn ALL members in ONE message. Dispatch every member together (multiple spawn_agent/sub-agent calls in a single response) — never drip them one per turn.
3. Barrier — advance ONLY after EVERY member returns. A member is "returned" when its work completes inline OR its sub-agent returns; a conditional member whose trigger is absent counts as returned. Do NOT advance, and do NOT start the next step, until the whole group has returned.
4. A sub-agent return advances the step identically to an inline call. Advancement is YOUR judgment against the task list — never wait for a hook or tool event. Mark each member completed (or "Skipped — <reason>") as the batch resolves.
5. Mutating steps wait for the barrier. Never start a code-mutating step (e.g. code-simplifier) until the full batch has returned — it must act on the complete review snapshot, not a partial one.
6. Hooks are accelerators only. Any step-tracking hook may emit a "next step" hint as an optimization; correctness MUST NOT depend on it. Codex runs with no hooks and advances entirely by this rule.

Blocked until: - [ ] all members spawned in one message - [ ] every member returned (incl. skipped conditional) - [ ] each member marked completed/skipped - [ ] mutating step deferred until after the barrier.

End-to-Start Debugger Trace — For non-trivial bugs, failed verification, regression fixes, behavior-changing code, or unclear code flow, start from the observed final state and walk backward before proposing a fix.

1. Frame 0: observed end state — Name the exact user-visible output, failing assertion, log line, persisted value, API response, rendered UI, or aggregate bucket. Record the reader/query/renderer that produced it with file:line evidence.
2. Walk backward one hop at a time — Trace final reader -> projection/cache/storage -> writer -> consumer/handler/job -> producer/caller -> original trigger. At every hop record: input, transformation, output, owner, and evidence.
3. Enumerate all feeder paths — Find every upstream producer/caller/event/job that can write into the final path, including retry, async, cache, background, and alternate UI/API paths. Mark each path verified, ruled out, or still unknown.
4. Build the hypothesis matrix — For each plausible cause, list evidence for, evidence against, how to reproduce/verify, blast radius, and status (primary, contributing, ruled out, latent). Do not fix until competing causes are explicitly resolved or bounded.
5. Choose the owning fix layer — Identify the invariant owner and the lowest shared point that protects all downstream consumers. A fix at the symptom site is rejected unless the symptom site owns the invariant.
6. Prove convergence forward — After choosing the fix, walk start -> end again and show how the corrected state reaches the observed final output. Map each root cause to a fix part and each fix part to a test/proof.

BLOCKED until: final state named · backward trace written · all feeder paths enumerated · hypothesis matrix completed · owning fix layer justified · forward convergence proof mapped to tests.

NEVER: Start at the first suspicious code path. Collapse multiple producers into one "flow". Treat duplicate symptoms as duplicate records without proving the read model. Skip ruled-out hypotheses.

Applicability in this workflow (reconciles with step 12): the canonical block below is the general fresh-context mechanism. In workflow-review-changes the step-12 post-fix re-review applies its principle — zero memory, re-read the full diff from scratch, no self-filtering — INLINE in the main session (the deliberate cost tradeoff documented at step 12), NOT via an isolated sub-agent. The isolated-sub-agent form below governs (a) the parallel dimensional reviewers in steps 3–7 (already sub-agents) and (b) the case where this entire workflow is invoked as a sub-agent inside a parent workflow. So "with isolated sub-agents where applicable" resolves to inline for the step-12 self-re-review — no contradiction.

Fresh Context Re-Review — Eliminate orchestrator confirmation bias after fixes by restarting the full review with isolated sub-agents where applicable.

Why: The main agent knows what it (or $feature-implement) just fixed and rationalizes findings accordingly. A fresh sub-agent has ZERO memory, re-reads from scratch, and catches what the main agent dismissed. Sub-agent bias is mitigated by (1) fresh context, (2) verbatim protocol injection, (3) main agent not filtering the report.

When: ONLY after a validated-finding fix cycle. A review round that finds zero issues ENDS the loop — do NOT spawn a confirmation sub-agent. A review round that finds issues triggers: validate findings → fix → full review restart from the first phase.

How:

1. Start a NEW full review invocation/task breakdown; when that protocol calls for agents, spawn NEW spawn_agent tool calls — use code-reviewer agent_type for code reviews, general-purpose for plan/doc/artifact reviews
2. Inject ALL required review protocols VERBATIM into the prompt — see SYNC:review-protocol-injection for the full list and template. Never reference protocols by file path; AI compliance drops behind file-read indirection (see SYNC:shared-protocol-duplication-policy)
3. Sub-agent re-reads ALL target files from scratch via its own tool calls — never pass file contents inline in the prompt
4. Sub-agent writes structured report to plans/reports/{review-type}-round{N}-{date}.md
5. Main agent reads the report, integrates findings into its own report, DOES NOT override or filter

Rules:

• SKIP fresh sub-agent when the prior full review found zero issues (no fixes = nothing new to verify)
• NEVER skip the full review restart after a fix cycle — every fix invalidates the prior verdict
• NEVER reuse a sub-agent across rounds — every fresh round spawns a NEW spawn_agent call
• Continue until a complete full review pass has zero findings; if the same blocker repeats 3 times with no progress, escalate via a direct user question
• Track iteration count and repeated blockers in conversation context (session-scoped, no persistent files)

Incremental Result Persistence — MANDATORY for all sub-agents or heavy inline steps processing >3 files.

1. Before starting: Create report file plans/reports/{skill}-{date}-{slug}.md
2. After each file/section reviewed: Append findings to report immediately — never hold in memory
3. Return to main agent: Summary only (per SYNC:subagent-return-contract) with Full report: path
4. Main agent: Reads report file only when resolving specific blockers

Why: Context cutoff mid-execution loses ALL in-memory findings. Each disk write survives compaction. Partial results are better than no results.

Report naming: plans/reports/{skill-name}-{YYMMDD}-{HHmm}-{slug}.md

Sub-Agent Return Contract — When this skill spawns a sub-agent, the sub-agent MUST return ONLY this structure. Main agent reads only this summary — NEVER requests full sub-agent output inline.

## Sub-Agent Result: [skill-name]

Status: ✅ PASS | ⚠️ PARTIAL | ❌ FAIL
Confidence: [0-100]%

### Findings (Critical/High only — max 10 bullets)

- [severity] [file:line] [finding]

### Actions Taken

- [file changed] [what changed]

### Blockers (if any)

- [blocker description]

Full report: plans/reports/[skill-name]-[date]-[slug].md

Main agent reads Full report file ONLY when: (a) resolving a specific blocker, or (b) building a fix plan. Sub-agent writes full report incrementally (per SYNC:incremental-persistence) — not held in memory.

Context budget — the return payload is a SUMMARY, not a transcript: ≤10 finding bullets, no raw file contents / full diffs / verbatim logs inline, no re-pasted source. Everything beyond the summary lives in the Full report on disk. A sub-agent that would exceed the summary shape MUST write the detail to its report and return only the pointer — the orchestrator's context is the scarce resource the whole map-reduce protects.

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.

Nested Task Expansion Contract — For workflow-step invocation, the [Workflow] ... row is only a parent container; the child skill still creates visible phase tasks.

1. Call the current task list first. If a matching active parent workflow row exists, set nested=true and record parentTaskId; otherwise run standalone.
2. Create one task per declared phase before phase work. When nested, prefix subjects [N.M] $skill-name — phase.
3. When nested, link the parent with TaskUpdate(parentTaskId, addBlockedBy: [childIds]).
4. Orchestrators must pre-expand a child skill's phase list and link the workflow row before invoking that child skill or sub-agent.
5. Mark exactly one child in_progress before work and completed immediately after evidence is written.
6. Complete the parent only after all child tasks are completed or explicitly cancelled with reason.

Blocked until: the current task list done, child phases created, parent linked when nested, first child marked in_progress.

Task Tracking & External Report Persistence — Bootstrap this before execution; then run project-reference doc prefetch before target/source work.

1. Create a small task breakdown before target file reads, grep, edits, or analysis. On context loss, inspect the current task list first.
2. Mark one task in_progress before work and completed immediately after evidence; never batch transitions.
3. For plan/review work, create plans/reports/{skill}-{YYMMDD}-{HHmm}-{slug}.md before first finding.
4. Append findings after each file/section/decision and synthesize from the report file at the end.
5. Final output cites Full report: plans/reports/{filename}.

Blocked until: task breakdown exists, report path declared for plan/review work, first finding persisted before the next finding.

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.

Project Reference Docs Gate — Run after task-tracking bootstrap and before target/source file reads, grep, edits, or analysis. Project docs override generic framework assumptions.

1. Identify scope: file types, domain area, and operation.
2. Required docs by trigger: always docs/project-reference/lessons.md; doc lookup docs-index-reference.md; review code-review-rules.md; backend/CQRS/API backend-patterns-reference.md; domain/entity domain-entities-reference.md; frontend/UI frontend-patterns-reference.md; styles/design scss-styling-guide.md + design-system/design-system-canonical.md; integration tests integration-test-reference.md; E2E e2e-test-reference.md; feature docs/specs feature-spec-reference.md + spec-system-reference.md + spec-principles.md; behavior/public-contract/spec-test-code sync workflow-spec-test-code-cycle-reference.md; derived spec index/ERD/reimplementation guides spec-system-reference.md + source Feature Specs under docs/specs/; architecture/new area project-structure-reference.md.
3. Read every required doc. 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 $project-init or the narrow lower-level route ($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.
4. Before target work, state: Reference docs read: ... | Not applicable: ....

Ready when: scope evaluated, required docs checked/read or setup route completed, lessons.md confirmed, citation emitted.

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 Bootstrap task tracking before target work; transition one task at a time.
• MANDATORY Persist plan/review findings to plans/reports/ incrementally and synthesize from disk.

• MANDATORY After task-tracking bootstrap and before target/source work, read required project-reference docs and cite Reference docs read: ....
• MANDATORY Always include lessons.md; project conventions override generic defaults.
• MANDATORY If project config, root instruction files, or any required reference doc is missing or stale, auto-run $project-init or the narrow lower-level route before ordinary project-specific work.

IMPORTANT MUST ATTENTION debugger trace gate: for non-trivial bug/fix/investigation/review work, start at the observed final output and trace backward through reader -> storage/projection -> writer -> consumer/job -> producer/trigger. Enumerate all feeder paths and hypotheses before fixing. BLOCKED until trace, hypothesis matrix, owning fix layer, and forward convergence proof exist.

• MANDATORY Parent workflow rows do not replace child phase tracking; expand phases and link the parent when nested.
• MANDATORY Orchestrators pre-expand child skill phases before invocation; use [N.M] $skill-name — phase prefixes and one-in_progress discipline.

• 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.

Prompt-Enhance Closing Anchors

IMPORTANT MUST ATTENTION follow declared step order for this skill; NEVER skip, reorder, or merge steps without explicit user approval IMPORTANT MUST ATTENTION for every step/sub-skill call: set in_progress before execution, set completed after execution IMPORTANT MUST ATTENTION every skipped step MUST include explicit reason; every completed step MUST include concise evidence IMPORTANT MUST ATTENTION if Task tools unavailable, maintain an equivalent step-by-step plan tracker with synchronized statuses

Closing Reminders

IMPORTANT MUST ATTENTION Goal: Ensure changed work reaches clean review through validated findings, verified fixes, full re-review, and synchronized docs/tests — review all uncommitted changes, validate findings, fix ONLY validated findings, then re-run $review-changes INLINE (only when $plan-execute changed files), looping plan→plan-execute→review-changes until one complete pass is clean.

MUST ATTENTION Protocols in force (concise digest of the SYNC/shared blocks this skill carries — each line is a signpost to its canonical body above; NEVER act on the digest alone, read the cited block):

• Parallel-Phase Advancement: spawn batch in one message; advance only after all-return barrier.
• End-to-Start Debugger Trace: trace observed end state backward before fixing.
• Fresh Context Re-Review: restart full review post-fix; zero-memory re-read counters confirmation bias.
• Incremental Persistence: append findings to report file per item; never hold in memory.
• Sub-Agent Return Contract: return only the summary shape; full report on disk.
• AI Mistake Prevention: verify generated content against evidence, trace downstream references, verify all affected outputs, re-read after context loss, surface ambiguity.
• Nested Task Creation: parent workflow row never replaces child phase tasks.
• Task Tracking & External Report: bootstrap task breakdown and report path before work.
• Critical Thinking: every claim needs traced proof; confidence >80% to act.
• Project Reference Docs: read required project-reference docs first; conventions override generic defaults.

IMPORTANT MUST ATTENTION run the sequence in order — step 1 $review-changes owns the baseline (surface analysis + UI review via internal $review-ui + integration-test/translation-sync/spec-drift gates), step 2 $why-review validates those findings BEFORE the parallel batch fires — so steps 3–7 act only on warranted findings — why: validating after fixing wastes the batch on false positives. IMPORTANT MUST ATTENTION spawn the steps 3–7 read-only reviewers ($review-architecture, $review-domain-entities [if entity files], $performance-review, $integration-test-review, $security-review) ALL in ONE message and advance ONLY after EVERY member returns (all-return barrier) — defer mutating $code-simplifier (step 8) until the barrier clears — why: a code-mutating step must see the complete review snapshot, not a partial one. IMPORTANT MUST ATTENTION every finding, recommendation, and verdict needs file:line proof or traced evidence + a confidence % — >80% act, 60–80% verify first, <60% DO NOT recommend; "Insufficient evidence" is valid output — why: speculation is forbidden output and silently encodes false positives into the fix plan.

MUST ATTENTION break work into small todo tasks using task tracking BEFORE starting — create ALL 15 tasks immediately (source of truth = workflows.json → review-changes.sequence); mark one in_progress, mark completed immediately after each step's evidence; on context loss call the current task list first — never duplicate. MUST ATTENTION grep 3+ existing patterns and read the target files BEFORE proposing any fix; cite file:line evidence in the fix plan — local conventions override generic framework defaults — why: closest example ≠ matching preconditions, verify shared base classes/scope/lifetime before copying. MUST ATTENTION after fixes in $plan-execute (and ONLY if $plan-execute changed files), re-run $review-changes INLINE over the current full diff from Phase 0; re-read the diff from scratch to counter orchestrator confirmation bias — why: the main agent rationalizes findings about its own fixes; loop $plan→$plan-execute→$review-changes until clean. MUST ATTENTION track full re-review invocations and repeated blockers in conversation context (session-scoped, no persistent files) — stop after the same blocker repeats 3 times with no progress and escalate via a direct user question; STOP and escalate if round N finds MORE issues than round N-1 — never silently loop. MUST ATTENTION PASS means one complete review pass finds zero blocking issues after all validated fixes and verification are included; a behavior-changing fix that left no covering §8 regression/preservation TC is an OPEN finding, NOT a clean pass — green tests do not normalize spec drift. MUST ATTENTION skip steps 9–12 ONLY when all reviews PASS with zero findings (no fixes needed); mark conditional tasks completed with note "Skipped — all reviews passed" — NEVER consolidate, rename, or omit steps. MUST ATTENTION adjudicate every behavior-vs-spec divergence in step 1 as CODE-WRONG (BLOCKING) / SPEC-STALE (spec is stale, $docs-update fixes spec first) / AMBIGUOUS (escalate) — NEVER silently pick a side; the workflow is NOT clean while any divergence stays unadjudicated. IMPORTANT MUST ATTENTION each step MUST invoke its skill invocation — marking a task completed without invocation is a workflow violation; NEVER batch-complete validation gates — why: a skipped gate ships unreviewed work. IMPORTANT MUST ATTENTION treat integration-test coverage gaps and multilingual UI translation gaps as mandatory a direct user question user-decision gates — surface them, never silently pass when tests or locale updates are missing. IMPORTANT MUST ATTENTION $why-review runs ONCE at step 2 as a FINDINGS-VALIDATION gate (drops false positives before the parallel batch); the fix-plan rationale check is owned by $plan-review (step 10), which self-invokes $why-review --validate-findings internally — no separate explicit step needed. IMPORTANT MUST ATTENTION when invoked as a step inside a parent workflow, run this whole 15-step workflow via spawn_agent (agent_type: "code-reviewer"), NEVER inline — why: inline execution absorbs 15 steps of context into the parent session. IMPORTANT MUST ATTENTION apply critical + sequential thinking — keep the SKEPTIC default when reviewing: steel-man rejected alternatives, invert each stated reason, stress-test top assumptions; section presence ≠ quality — why: certainty without evidence is the root of hallucination. IMPORTANT MUST ATTENTION Easy to Change is the success metric — every finding/test/refactor must answer "does this make the next change cheaper?"; name the real enemies (coupling, hidden state, duplicated knowledge, unclear intent) — reject best practices that raise change cost.

Anti-Rationalization:

| Evasion | Rebuttal | | ----------------------------------------------- | ------------------------------------------------------------------------------------------------------------------ | | "Reviews look clean, skip $why-review" | Step 2 validates findings BEFORE the batch — run it; a false positive entering the fix plan wastes 5 reviewers. | | "I already know what I fixed, skip re-review" | Orchestrator confirmation bias — re-read the full diff from scratch INLINE; main-agent self-review is NOT enough. | | "Tests are green, the spec drift is fine" | Green can encode the drift itself — adjudicate CODE-WRONG / SPEC-STALE; not clean until every divergence resolved. | | "Mark the step done, the skill obviously ran" | Marking completed without invoking the skill invocation is a workflow violation — show the invocation evidence. | | "Same blocker again, one more loop will fix it" | Cap at 3 no-progress repeats → escalate via a direct user question; if issues increase round-over-round, STOP now. | | "Fix at the crash site, it's faster" | Trace caller (wrong data) vs callee (wrong handling); fix at the responsible layer, never patch the symptom site. |

IMPORTANT MUST ATTENTION step 1 $review-changes runs FIRST and owns the baseline; step 2 $why-review validates findings BEFORE the steps 3–7 parallel batch — spawn the batch in ONE message, advance only after the all-return barrier, defer $code-simplifier until it clears. IMPORTANT MUST ATTENTION every finding/verdict needs file:line evidence + confidence (>80% act, <60% DO NOT recommend); grep 3+ patterns and read target files before any fix — no speculation. IMPORTANT MUST ATTENTION after $plan-execute changes files, re-run $review-changes INLINE from scratch and loop until ONE clean zero-finding pass — a behavior change with no covering §8 TC is an OPEN finding; cap repeated blockers at 3 → escalate.

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.

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. 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.
• 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.