Sync Opencode

---
name: sync-opencode
description: '[opencode] Use when running the opencode sync and verify pipeline (reconcile the recommended root opencode.json, generate the .opencode hooks bridge plugin, run tooling tests, verify drift).'
disable-model-invocation: true
---

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

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

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

**Missing/stale context route:** If `docs/project-config.json`, the docs index, `lessons.md`, `CLAUDE.md`, `AGENTS.md`, or any task-required reference doc is missing or stale, auto-run `$project-init` or the narrow setup route (`$project-config`, `$docs-init`, `$scan-all`, `$scan --target=<key>`, `$ai-context-refresh`) before ordinary project-specific work. A full `$sync-codex` run preflights `CLAUDE.md`; a completed `$ai-context-refresh` run may invoke the standalone runner with `--skip=claude-md` after final source edits. Markerless roots need AI smart-merge unless `portability.requireUniversalGuides: false` is explicit.

**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`, `configured styling reference`, `design-system/README.md`
- Spec authoring, `docs/specs/` pathing, or TC format: `feature-spec-reference.md`, `spec-system-reference.md`, `spec-principles.md`
- Behavior/public-contract changes or spec-test-code sync: `workflow-spec-test-code-cycle-reference.md` plus the spec docs above
- Derived spec indexes/ERDs/reimplementation guides: `spec-system-reference.md` and source Feature Specs under `docs/specs/`
- Integration test implementation/review: `integration-test-reference.md`
- E2E test implementation/review: `e2e-test-reference.md`
- Code review/audit work: `code-review-rules.md` plus domain docs above based on changed files

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

## Quick Summary

**Goal:** Reconcile the project-root `opencode.json` with the framework's recommended opencode defaults, compile `.claude/settings.json` into a portable opencode plugin (`.opencode/plugins/easy-claude-hooks.js`), and verify both. This runner is the **single and only** entrypoint for the opencode surface pipeline.

> **PORTABILITY CONTRACT — `.claude/` and `.opencode/` are portable, self-running and self-testing.**
> Copy them into ANY repository — a Python repo, a .NET repo, a repo with no `package.json` at all —
> and the sync / verify / test entrypoint still works, because it is a path INSIDE the bundle invoked
> with plain `node`. **Never** drive this framework through a host `package.json` script, and never
> document one: `npm run …` names a command that does not exist in most projects the bundle is copied
> into. The only external requirement is `node` >= 18 — no `node_modules`, no npm, no lockfile. The
> generated plugin imports only `node:` built-ins and spawns the canonical `.claude/hooks/*.cjs`
> scripts.
>
> Discover the roster instead of memorizing it:
> `node .claude/skills/sync-opencode/scripts/run-opencode-sync.mjs --list-stages`

**Summary:**

- opencode has **no shell-command hook system** — hooks are JavaScript plugin callbacks. This skill compiles `.claude/settings.json` into a bridge plugin whose runtime drives the original Claude hooks from opencode's plugin events.
- **Recommended opencode config is part of the framework.** `.opencode/opencode.recommended.json` is the single source of truth for the framework's opencode defaults; the `config` stage deep-merges it into the project-root `opencode.json` (recommended keys win, project-only keys survive).
- Scope is **hooks + recommended config + the sub-agent mirror**. opencode already auto-discovers skills from `.claude/skills` and `.agents/skills`, so there is **no skill mirroring** here (unlike `$sync-codex`). Sub-agents are NOT auto-discovered, so `.claude/agents/*.md` IS mirrored into `.opencode/agent/*.md` by the `agents` stage — that is what lets the workflow protocols dispatch the same specialists (`architect`, `code-reviewer`, `security-auditor`, …) on opencode.
- Keep `.claude` canonical: edit `.claude/settings.json` / `.claude/hooks/**` and re-run this pipeline; never hand-edit the generated `.opencode/plugins/easy-claude-hooks.js`.
- To change a default opencode setting, edit `.opencode/opencode.recommended.json`, then re-run this pipeline to propagate it into the root config of every project the `.opencode/` folder is copied into.
- The legacy hand-written `.opencode/plugins/notification.js` is superseded by the generated bridge; the runner backs it up under `tmp/opencode-legacy/` and removes it so notifications are not sent twice.

**Workflow:**

1. **Config** — `node .claude/scripts/opencode/sync-config.mjs` deep-merges `.opencode/opencode.recommended.json` into the project-root `opencode.json`.
2. **Sync** — `node .claude/skills/sync-opencode/scripts/run-opencode-sync.mjs` regenerates the bridge plugin, the `.opencode/agent/*.md` mirror, and the sync report.
3. **Test** — the runner executes the opencode tooling tests (config + writer + generated-plugin runtime).
4. **Verify** — the runner re-merges/re-renders in memory with the REAL writers and byte-compares against the tracked files.
5. **Inspect** — on failure, re-run the failing stage with `--only=<stage> --verbose`.

**Key Rules:**

- MUST run stages in order — the orchestrator fails fast on the first non-zero exit
- NEVER hand-edit `.opencode/plugins/easy-claude-hooks.js`; regenerate it from `.claude/settings.json`
- **NEVER hand-edit a project-root `opencode.json` as the way to change framework defaults** — edit `.opencode/opencode.recommended.json` and re-run the pipeline
- `.opencode/opencode.recommended.json` MUST NOT be named `.opencode/opencode.json`; opencode auto-loads that path as project config
- The generated plugin and `.opencode/plugins/**` are the only opencode hook surface; no skill mirror is produced (skills are auto-discovered, sub-agents are mirrored into `.opencode/agent/**`)
- Only `node "$CLAUDE_PROJECT_DIR"/...` hook commands are compiled; other command shapes are reported as `unsupported-command-shape`
- Claude events opencode cannot reproduce are reported as `skipped-events` in the sync report — never silently dropped
- Idempotent — re-running the sync produces byte-identical plugin, config, and agent output

## Why a bridge plugin (not a config mirror)

> The `config` stage above reconciles framework-recommended opencode **defaults**; it is not a mirror of Claude settings. This section explains why the project's *hook* surface is generated as a plugin.

Codex supports lifecycle hooks and can be driven by `.codex/hooks.json`. opencode cannot: its only
extension point for lifecycle behavior is a JS/TS plugin under `.opencode/plugins/` (or an npm plugin).
So instead of transcribing commands into a JSON file, this pipeline emits a self-contained plugin that
spawns the canonical Claude hook scripts and translates both directions:

| Direction | Translation |
| --- | --- |
| opencode → Claude hook | event name, tool id → Claude matcher name, opencode args → `tool_input` field names |
| Claude hook → opencode | exit `0` = allow · exit `2` = block (throw) · stdout `hookSpecificOutput.updatedInput` = argument rewrite · stdout `hookSpecificOutput.additionalContext` = injected context · stdout `hookSpecificOutput.permissionDecision: "deny"` = block |

## Recommended opencode config (source of truth)

The framework ships recommended opencode defaults as a portable template and
reconciles them into whatever project the `.opencode/` folder is copied into.

| Item | Path |
| --- | --- |
| **Source of truth (edit this)** | `.opencode/opencode.recommended.json` |
| **Generated target (never hand-edit for defaults)** | `<project-root>/opencode.json` |
| Writer / verifier | `.claude/scripts/opencode/sync-config.mjs` (`--check` for verify) |

> **To change a default recommended opencode setting in the future, edit `.opencode/opencode.recommended.json`** and re-run `$sync-opencode`. Every project that receives the `.opencode/` folder then gets the updated default the next time the pipeline runs. The recommended file is deliberately NOT named `.opencode/opencode.json` because opencode auto-loads that path as project config — keeping the `.recommended.json` name makes it a template, not an active config.

**Merge semantics:** the writer deep-merges the recommended defaults into the existing root `opencode.json`. Recommended keys win at every leaf; object keys that exist only in the project survive untouched; arrays in the recommended file replace the project's array. A project with no root config receives the recommended defaults verbatim. A malformed existing root config is reported, never clobbered.

### Compaction budget — 500K tokens

The framework targets the SAME 500K auto-compact budget on all three surfaces:
Claude Code (`env.CLAUDE_CODE_AUTO_COMPACT_WINDOW` in `.claude/settings.json`), Codex
(`model_auto_compact_token_limit` in `.codex/config.toml`) and opencode (the pinned
model's `limit.context` here).

opencode has no absolute compaction threshold — it compacts relative to the model's
declared window, so the window IS the knob. From `session/overflow.ts` (v1.18.31):

```text
usable = limit.input ? max(0, limit.input - (compaction.reserved ?? min(20_000, maxOutput)))
                     : max(0, limit.context - maxOutput)
maxOutput = min(limit.output, 32_000)          // OUTPUT_TOKEN_MAX
compaction happens once total tokens >= usable
```

So `limit.context: 500000` + `limit.output: 384000` compacts at **468,000** tokens.

Two traps this encodes, both verified against the released source:

- **`compaction.reserved` is inert here.** It is only read on the `limit.input` branch,
  and models.dev declares no `input` for this model. Raising it does nothing.
- **Do NOT express the budget as `limit.input`.** It is undocumented, and `limit.context`
  is what the rest of opencode reads as the window — the TUI context percentage, ACP usage
  reporting, and the `limits` handed to the AI SDK. Capping `input` while leaving `context`
  at 1M shows ~48% in the TUI at the moment it compacts.

**Copying the framework into a new project:** copy `.claude/` and `.opencode/` (including `.opencode/opencode.recommended.json`), then run `$sync-opencode` — it generates/updates that project's root `opencode.json`, bridge plugin, and reports.

## Hook mapping

| Claude hook event | opencode plugin surface | Notes |
| --- | --- | --- |
| `PreToolUse` | `tool.execute.before` | throw blocks the tool; `updatedInput` rewrites `output.args` |
| `PostToolUse` | `tool.execute.after` | `additionalContext` is appended to the tool output |
| `UserPromptSubmit` | `chat.message` | `additionalContext` is injected as a synthetic text part |
| `SessionStart` | `event:session.created` / `event:session.compacted` + `experimental.chat.system.transform` | `additionalContext` is injected into the system prompt |
| `SessionEnd` | `event:session.deleted` | |
| `Stop` | `event:session.idle` | |
| `Notification` | `event:question.asked` / `event:permission.asked` | matcher vocabulary `AskUserPrompt` / `permission_prompt` |
| `PermissionRequest` | `permission.ask` | blocked → `status: "deny"` |

**Tool id aliases** (opencode → Claude matcher names the hooks are written against):

| opencode tool | Claude matcher names |
| --- | --- |
| `bash` | `Bash` |
| `edit` | `Edit`, `MultiEdit` |
| `write` | `Write` |
| `read` | `Read` |
| `grep` | `Grep` |
| `glob` | `Glob` |
| `apply_patch` | `Edit`, `Write`, `MultiEdit`, `NotebookEdit` |
| `todowrite` | `TodoWrite`, task tracking, `TaskUpdate`, `update_plan` |
| `webfetch` / `websearch` | `WebFetch` / `WebSearch` |
| `question` | ask the user directly |
| `skill` | skill invocation |
| `<server>_<tool>` (MCP) | `mcp__<server>__*` |

## Known limitations

- **MCP argument visibility.** opencode registers MCP tools as `<server>_<tool>`, and Claude matchers like `mcp__github__*` are matched against that convention. However, opencode does **not** expose MCP tool arguments to `tool.execute.before`, so an MCP `PreToolUse` hook that inspects `tool_input` cannot see them on this host.
- **`apply_patch` has no `file_path`.** The bridge reports it as `Edit`/`Write`/`MultiEdit` and passes `patchText` through as `patch`; hooks that require `file_path` (doc-sync-gate) allow/ignore it exactly as they do on other hosts.
- **Session resume.** `SessionStart` runs on `session.created` and `session.compacted`; when opencode resumes a session without either event, the bridge runs `source: "startup"` once on the first `chat.message`.
- **Claude-only stdout fields** other than `updatedInput`, `additionalContext`, and `permissionDecision` are ignored by the bridge.

## Usage

```bash
# Discover the stage roster:
node .claude/skills/sync-opencode/scripts/run-opencode-sync.mjs --list-stages

# Full sync + test + verify:
node .claude/skills/sync-opencode/scripts/run-opencode-sync.mjs

# Stream live child output:
node .claude/skills/sync-opencode/scripts/run-opencode-sync.mjs --verbose

# Every read-only gate (no mutation):
node .claude/skills/sync-opencode/scripts/run-opencode-sync.mjs --verify-only

# Just regenerate the plugin:
node .claude/scripts/opencode/sync-hooks.mjs

# Just verify the tracked plugin is current:
node .claude/scripts/opencode/sync-hooks.mjs --check

# Just reconcile the recommended root opencode.json:
node .claude/scripts/opencode/sync-config.mjs

# Just verify the root opencode.json is current:
node .claude/scripts/opencode/sync-config.mjs --check

# Skip a stage while debugging:
node .claude/skills/sync-opencode/scripts/run-opencode-sync.mjs --skip=hooks
```

**Exit codes:** `0` all pass · `1` orchestrator failure · non-zero propagates from failing stage.

## Stages

7 stages, sequential — the complete opencode surface pipeline, owned entirely by this runner:

| # | Stage | Script | Effect |
| --- | --- | --- | --- |
| 1 | config | `.claude/scripts/opencode/sync-config.mjs` | Deep-merge `.opencode/opencode.recommended.json` into the project-root `opencode.json` |
| 2 | hooks | `.claude/scripts/opencode/sync-hooks.mjs` | Generate `.opencode/plugins/easy-claude-hooks.js` + `tmp/opencode-hooks.sync.report.json`; back up/remove legacy `notification.js` |
| 3 | agents | `.claude/scripts/opencode/sync-agents.mjs` | Mirror `.claude/agents/*.md` into `.opencode/agent/*.md` (`mode: subagent` + the canonical body verbatim) |
| 4 | tests | Runner discovers `.claude/scripts/opencode/tests/*.test.{mjs,cjs}` | Run opencode tooling tests; missing or empty discovery fails |
| 5 | verify-config | `.claude/scripts/opencode/sync-config.mjs --check` | Re-merge with the REAL writer and byte-compare with the project-root `opencode.json` |
| 6 | verify-hooks | `.claude/scripts/opencode/sync-hooks.mjs --check` | Re-render with the REAL writer and byte-compare with the tracked plugin |
| 7 | verify-agents | `.claude/scripts/opencode/sync-agents.mjs --check` | Re-render every agent with the REAL writer and byte-compare with the tracked `.opencode/agent/*.md` |

## Closing Reminders

**Protocols in force** (concise digest of the SYNC/shared blocks this skill carries) — **MUST ATTENTION** each canonical body below still binds:

- **AI Mistake Prevention:** verify generated content against evidence, trace downstream references, verify all affected outputs, re-read after context loss, surface ambiguity.
- **Critical Thinking:** traced `file:line` proof per claim, confidence >80% to act, never guess.

**MUST ATTENTION** keep the `$sync-opencode` skill user-invoked-only; no unrelated skill, agent, or workflow may auto-run the mutating pipeline.
**MUST ATTENTION** edit `.claude/settings.json` and `.claude/hooks/**` as the source, then regenerate; NEVER hand-edit `.opencode/plugins/easy-claude-hooks.js`
**MUST ATTENTION** the framework's default opencode settings live in `.opencode/opencode.recommended.json` — edit THAT file to change defaults, then re-run this pipeline; never treat a project-root `opencode.json` as the source
**MUST ATTENTION** the generated plugin must import only `node:` built-ins so `.claude`/`.opencode` stay portable into any project
**MUST ATTENTION** the `config` stage deep-merges (recommended wins, project-only keys survive) and never clobbers a malformed root config — it reports instead
**MUST ATTENTION** the runner auto-resolves the repo root from its own path — do not pass a cwd flag
**MUST ATTENTION** the legacy `notification.js` is backed up, never destroyed, before removal

**Anti-Rationalization:**

| Evasion | Rebuttal |
| --- | --- |
| "Just edit the .opencode plugin directly" | Next sync overwrites it. Edit `.claude/settings.json` and regenerate. |
| "Just edit the project-root opencode.json to change the defaults" | The next sync re-merges the recommended file and your default is lost, and no other project gets it. Edit `.opencode/opencode.recommended.json` and re-run. |
| "Make the recommended file `.opencode/opencode.json`" | opencode auto-loads that exact path as project config, so it would stop being a template. Keep the `.recommended.json` name. |
| "Skip the tests stage" | The tests exercise the generated plugin against real hook subprocesses; skipping ships an untested bridge. |
| "Mirror skills too for symmetry with sync-codex" | opencode already discovers `.claude/skills`; a mirror would duplicate and drift. Out of scope by design. |

> **[FAILS FAST]** First non-zero stage exit aborts the chain. Re-run the failing stage with `--only=<id> --verbose`.
> **[REPO ROOT]** The orchestrator auto-resolves the repo root from its own path.

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

> **AI Mistake Prevention** — Failure modes to avoid on every task:
>
> **Project applicability gate.** Before applying a stack, layer, style, tool, or architecture rule, read the project's config and relevant references, then check local implementations. Treat framework examples as examples; honor explicit N/A and do not require a technology or convention the project does not use.
> **ROOT-CAUSE GATE — INVESTIGATE FIRST.** Before applying any project-related correction, always use the project's root-cause investigation protocol and establish the cause; the failure site may be only a symptom.
> **FAILED-TEST GATE.** For any failed or unstable test, use the project's test-investigation protocol before editing source or tests; never change either side merely to force green.
> **Re-read files after context changes.** Context compaction, resume, or long-running work can make memory stale; verify current files before acting.
> **Verify generated content against source evidence.** AI hallucinates APIs, names, claims, and document facts. Check the relevant source before documenting or referencing.
> **Check downstream references before deleting or renaming.** Removing an artifact can stale docs, generated mirrors, configs, and callers; map references first.
> **Trace the full impact chain after edits.** Changing a definition can miss derived outputs and consumers. Follow the affected chain before declaring done.
> **Verify ALL affected outputs, not just the first.** One green check is not all green checks; validate every output surface the change can affect.
> **Assume existing values are intentional — ask WHY before changing OR flagging one as a defect.** Before changing or reporting a constant, limit, flag, cutoff, wording, or pattern, read nearby context and history, the CALLER's ordering, and 2+ sibling call sites of the same convention. A doc stating WHAT without WHY is missing rationale, not proof of a missing guard.
> **Surface ambiguity before acting — don't pick silently.** Multiple valid interpretations require an explicit question or stated assumption with risk.
> **Assert the outcome your system owns, not the intermediate state your infrastructure owns.** When verifying async work, assert the final business state — never the delivery/retry bookkeeping held in shared infrastructure that any co-running process can write. Such a check passes when run alone and flakes the moment anything else shares that infrastructure.
> **Store disposable generated output in the project workspace.** If an output can be regenerated and is not source code, a canonical source-of-truth, or an intentionally versioned projection, write it under the project-root `tmp/` or `temp/` directory (prefer `tmp/`), scoped to the run. This includes temporary state, integration/E2E results, reports, logs, screenshots, traces, videos, coverage, dumps, and candidate evidence. Never put these outputs in source, docs, the plans root (default `plans/`; a `docsRoots.plans.path` entry in `docs/project-config.json` overrides the path), the team-artifacts root (default `team-artifacts/`; `docsRoots.teamArtifacts.path` in the same config overrides the path), or mirror directories; the project-root `.gitignore` must ignore `/tmp/` and `/temp/` by default. Committed fixtures, accepted baselines, canonical specs/docs, and explicitly versioned generated mirrors remain at their declared owner paths.
> **Judge the environment before judging the code.** A bug report, failed test, error, or unexpected output is not proof of a code defect. Before and during adjudication, weigh environment causes as a competing hypothesis — setup, config, version and dependency state, service dependencies, stale artifacts or leftover state, and transient resource pressure (RAM, CPU, disk, handles, network). State the discriminator you ran; fix an environment cause in the environment, never by editing product code or weakening a test to absorb it.
> **Keep shared guidance role-relevant.** Universal guidance must help every receiving skill or agent; code-specific obligations belong only in code-specific protocols.

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

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

> **Project Protocol Overlay** — Before executing this skill, resolve project overlays from the index at `<docsRoots.projectReference.path>/skill-protocols-reference.md` (default `docs/project-reference/`; `docsRoots.projectReference.path` in `docs/project-config.json` overrides it). A matching `referenceDocs[]` filename may relocate the index within that root; this registry is independent from task-specific reference-doc selection, so omitted or empty `referenceDocs` does not disable it. The index's `**Protocols directory:**` header selects a project-root-relative body directory (default `docs/project-protocols/`). Match this skill against the `Target` column and take the most specific tier ONLY — exact name > glob > `*`. **That precedence orders overlays against EACH OTHER, never against this skill.** Read only matched bodies derived as `<protocols-dir>/<Name>.md`; the row's Body link is display text, never a read path. Reject unsafe paths without reading. A matched body that is missing or malformed is REPORTED and skipped — never reconstructed from the index Description. An absent index or no match -> proceed with no overlay, silently. Full contract: `.claude/skills/project-skill-protocol/references/registry.md`.
>
> Overlays are **ADDITIVE ONLY**: they ADD rules on top of this skill's own protocol and NEVER replace, override, disable, or reinterpret a rule it already states — removing every overlay must return this skill to exactly its documented behavior. An overlay is a BRIEF, not an authority escalation: it can NEVER waive a workflow gate, git discipline, a review gate, or a user-confirmation gate. A genuine overlay-vs-skill conflict, or two equally-specific overlays that directly contradict -> surface both to the user; NEVER resolve silently.

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

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

**MUST ATTENTION** resolve this skill's overlays from the index at `<docsRoots.projectReference.path>/skill-protocols-reference.md` (default `docs/project-reference/`; `docsRoots.projectReference.path` in `docs/project-config.json` overrides it); an empty task `referenceDocs` does NOT disable this lookup. Read ONLY matched bodies from the directory the index header names (default `docs/project-protocols/`). Specificity (exact > glob > `*`) ranks overlays against EACH OTHER, never against the skill. Missing/malformed body → report and skip; no index or no match → proceed silently. Overlays are ADDITIVE ONLY — never an authority escalation or a gate waiver; equal-tier contradiction goes to the user.

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

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

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

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

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

**MUST ATTENTION** critical + sequential thinking: every claim carries traced evidence (`file:line` for code, source URL or artifact section otherwise); confidence >80% to act, <60% do NOT recommend. Never present a guess as fact; admit uncertainty and stay skeptical of your own confidence.

<!-- /SYNC:critical-thinking-mindset:reminder -->
<!-- SYNC:ai-mistake-prevention:reminder -->

**MUST ATTENTION** Check project config, relevant references, and local evidence before applying stack-specific conventions; honor explicit N/A. ROOT-CAUSE GATE: before any project-related correction, use the appropriate root-cause investigation protocol; failed/unstable tests require the test-investigation protocol before editing source/tests — never force green.

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

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

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

## 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. Resolve `specArtifacts` before selecting identity or carrier: use a valid profile, use strict-default TC/test identity only when the profile is absent, and block a malformed or unsupported declaration. Trace every requirement or invariant through decision, task, configured case/test identity and inspected assertion evidence, then carry it through source evidence and canonical docs/spec updates.
> 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** Resolve and validate `specArtifacts`: use valid native owner/case/variant identity and assertion-bearing evidence; use strict-default TC/TestSpec only when the profile is absent; block a malformed or unsupported declaration without fallback.
- **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, analyze the task graph (output dependencies, shared write targets) into ordered parallel waves per PARALLELIZE before starting any task, then keep it synchronized as each step starts/completes. Preserve fixed ordering when a skill or workflow explicitly fixes it.
- **MANDATORY** Pin `Original goal:` before the first action and keep `User prompts this session: P1…Pn` current; re-read both at every step, before delegation, and after compaction.
- **MANDATORY** Before claiming done, map the result to the original goal and every prompt (`P# → done | deferred | n/a`); never store secrets in them.
## [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`/a linter 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)

- **Resolve project applicability before using framework examples.** Read the project config and relevant references, then inspect local evidence; honor explicit N/A and never impose a language, framework, architecture layer, styling method, tool, or runtime surface the project does not use.
- **ROOT-CAUSE GATE — INVESTIGATE FIRST.** Before applying any project-related correction, always use the project's root-cause investigation protocol and establish the cause; the failure site may be only a symptom.
- **FAILED-TEST GATE.** For any failed or unstable test, use the project's test-investigation protocol before editing source or tests; never change either side merely to force green.
- **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.
- **Ownership before action.** When investigating a failure, ask which part owns the behavior before changing anything. Trace the wrong state to the component responsible for its invariant, then make one authoritative correction there.
- **Test failure → record a provisional verdict before trace/edit, then investigate.** Use the full five-way taxonomy: SOURCE-WRONG (production violates intent), TEST-WRONG (assertion/setup is stale), TEST-NOT-OPTIMAL (valid but fragile or low-signal test), ENVIRONMENT-BLOCKED (external state prevents a verdict), or AMBIGUOUS (intent/evidence cannot choose safely). Then trace root cause and triangulate against the governing spec if one exists (the business spec root — default `docs/specs`; a `specRoots.business.path` entry in `docs/project-config.json` overrides the path) AND source. NEVER weaken an assertion, add a skip, relax a timeout, or change source merely to force green.
- **Grep ALL removed names after extraction/refactoring.** Primary file "done" ≠ secondary files clean. Grep entire scope for every removed symbol before declaring complete.
- **Assume existing values are intentional — ask WHY before changing OR flagging one as a defect.** Pattern-matching as "wrong" skips context. Before changing or reporting any constant/limit/flag/cutoff: read comments, git blame, the CALLER's ordering (the guarantee that makes the value correct usually lives in code running immediately BEFORE the cited line), and 2+ sibling call sites of the same convention. A doc stating WHAT without WHY is missing rationale, not proof of a missing guard — and in a validation pass, an accurate `file:line` citation proves the transcription, never the defect.
- **Verify ALL affected outputs, not just the first.** One build green ≠ all green. Multi-stack changes (backend/frontend/tests/docs) require verifying EVERY output.
- **Evaluate fit before copying a nearby pattern.** Closest example ≠ matching preconditions — verify the new context shares the same constraints, base classes, scope, lifetime.
- **Holistic analysis — resist the nearest-attention trap.** Do not dive into the first plausible cause. List every precondition (configuration, environment, inputs, dependencies, versions, permissions, and state). Verify each against evidence, not intuition. Ask "what would falsify this?" — if nothing, it is not a hypothesis. The most expensive failure is going deeper into an assumed area while the real issue sits in an unexamined condition.
- **Minimal changes — apply the relevance test.** Every change must trace to the reported problem; avoid unrelated cleanup. For review or enhancement work, announce improvements beyond the main request rather than silently expanding scope. Ask: "Would this change exist if I were not addressing this request?" — if not, remove it or disclose it.
- **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.
- **Why-Review adversarial mindset — apply when reviewing any plan, decision, or design.** Default SKEPTIC not VALIDATOR: steel-man a rejected alternative, invert each stated reason ("what does it sacrifice?"), stress-test top 2-3 assumptions, run pre-mortem ("ships, fails in 3 months — what breaks?"), surface 1-2 alternatives author missed. Section presence ≠ quality; quality = causal reasoning + concrete mitigations + evidence, not "it's better" or "monitor closely".
- **Front-load report-write in sub-agent prompts for large reviews.** Many-file sub-agents hit budget before final write — findings lost. Design prompts so: (1) report-write is first explicit deliverable, (2) append per-file/section (not batched), (3) scope bounded so reads don't exhaust budget. Truncated mid-sentence with no report file → spawn narrower scope, don't retry same prompt.
- **After context compaction, re-verify all prior phase outcomes before continuing.** Summaries describe intent, not environment state (git index, filesystem, processes). On resume, FIRST audit: git status, re-read modified files, verify filesystem. Every "completed" claim is an untested hypothesis until evidence confirms.
- **OOM/memory: check row count before row size.** Triage: (1) Unbounded query — no DB filter for trigger? Push filter to DB; eliminates OOM. (2) Large rows? Projection reduces proportionally. Row reduction > projection in ROI.
- **Assert the outcome your system OWNS, never the intermediate state your INFRASTRUCTURE owns.** When testing anything asynchronous (queue/broker delivery, retries, background jobs, caches, replication), assert the final business/entity state. NEVER assert the delivery bookkeeping — consume/send status, attempt counts, last-error, row existence or counts in a broker, scheduler, or outbox/inbox table. That bookkeeping lives in shared infrastructure that ANY co-running process (a peer worker, a second replica, a leftover local container) can write, usually under a deterministic shared key, so the assertion silently tests the developer's environment instead of the system: green when run alone, flaky the instant anything else shares that broker + database. Gate question for every assertion: "would this hold no matter WHICH process did the work?" — if no, assert the converged data state instead. Corollary: process-local fault injection and in-process telemetry cannot gate work any process may perform — use them as stress amplifiers (arm → bounded window → disarm → assert convergence), never as preconditions.
- **Store disposable generated output in the project workspace.** If an output can be regenerated and is not source code, a canonical source-of-truth, or an intentionally versioned projection, write it under the project-root `tmp/` or `temp/` directory (prefer `tmp/`), scoped to the run. This includes temporary state, integration/E2E results, reports, logs, screenshots, traces, videos, coverage, dumps, and candidate evidence. Never put these outputs in source, docs, the plans root (default `plans/`; a `docsRoots.plans.path` entry in `docs/project-config.json` overrides the path), the team-artifacts root (default `team-artifacts/`; `docsRoots.teamArtifacts.path` in the same config overrides the path), or mirror directories; the project-root `.gitignore` must ignore `/tmp/` and `/temp/` by default. Committed fixtures, accepted baselines, canonical specs/docs, and explicitly versioned generated mirrors remain at their declared owner paths.
- **Judge the environment before judging the code — a competing hypothesis, not a fallback.** A bug, failed test, error, or odd output is NOT proof of a code defect. Before deep tracing and before any verdict, sweep environment preconditions (toolchain/dependency/lockfile state, stale build or cache artifacts, env vars and config profile, service dependencies up-migrated-seeded, ports/network/clock, OS-path/locale, permissions and locks, leftover processes/containers/test data) AND transient resource pressure (RAM/OOM, CPU saturation under parallel workers, disk/temp exhaustion, handle and connection-pool limits, network flakiness, a timeout that is really slowness). Tell-tale shape: non-deterministic, timing-dependent, passes alone but fails in parallel, fails only on one machine or only on CI, or an error naming resources rather than business rules. Cite the discriminator you ran (clean environment? did code on the failing path change since it last passed? one machine or all? concurrency 1 or a clean rebuild?) — a verdict without one is a guess, for code as much as for the environment. Fix an environment cause in the environment or setup; NEVER edit product code or weaken/skip a test to absorb it, and a failure that vanishes on retry stays unexplained until its mechanism is named. — why: forcing green against an environment fault hides the real defect and permanently rots the test.
- **Cross-platform execution is a required contract.** Before authoring or changing a tool, script, process launcher, path assertion, or filesystem test, name the supported Windows, macOS, and Linux behaviors. Use platform-neutral Node APIs and literal argv vectors; never infer shell, temporary-path, executable-extension, ACL, or symlink semantics from the current host. Canonicalize existing paths before identity, hashing, or equality checks; test native Windows and POSIX seams when behavior differs; keep CI platform matrices authoritative. Preserve fail-closed security boundaries — repair the fixture or platform branch, never weaken the guard just to make one OS green.
- **Keep domain concepts out of generic/shared/infrastructure layers.** Reusable layer (shared library, framework, infra module) must reference NO consumer-specific domain concept — tenant/customer/product IDs, business entities, feature rules. Leak compiles + runs → passes review silently while coupling the "reusable" layer to one consumer. Keep shared type domain-free; push domain fields/logic down into the consumer via subclass/composition. — why: a layer coupled to one consumer's domain is no longer reusable.

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