[Documentation] Use when generating, smart-merging or refactoring the root AI context (CLAUDE.md, AGENTS.md) and its Claude/Codex projections. Config file only: project-config.
日本語の概要は準備中です。原文の説明を表示しています。
[Utilities] Use when asking about or configuring the .claude/.codex/.agents/.opencode skills framework: command discovery/help, runtime modes, scoped settings, commit reviews, hooks, workflows or skill visibility. Explains options; changes only requested settings.
インストールする前に、エージェントに与えられる指示の中身を確認できます。
Goal: Explain framework options and apply only requested settings at their correct scope, with validated saved and effective results.
Summary: Choose help/settings/workflow → discover exact keys and effective layers → answer questions read-only or apply the requested change → validate and read back → report the winning value/source.
Answer framework questions and manage requested settings without requiring the user to know JSON keys. This lightweight entry skill remains eligible for automatic selection when heavy skill auto-trigger is disabled. It does not authorize unrelated heavy skills, scans or workflows.
/framework-config [--mode=help|settings|workflow] <request> [options]
(Codex: $framework-config; OpenCode uses its native skill loader.) Infer the mode from natural
language when no flag is given; no arguments shows the three modes and a few common options.
| Mode | Purpose | Examples |
|---|---|---|
help | Command/category search, command details, task recommendations, framework usage | --mode=help, --mode=help review, --mode=help "debug a login error" |
settings | Show/explain/set/reset a framework preference | --mode=settings show, --mode=settings disable heavy auto-trigger --scope=checkout |
workflow | Show/set workflow route mode ask/auto/off, including session-only changes | --mode=workflow --show, --mode=workflow off --scope=session |
--scope=session|checkout|user|team wins; session scope is supported
only for workflow routing. User means every project; team means the configured project default.Run .claude/scripts/ck-help.py using the available Python interpreter (py -3 on Windows), with the
relevant English query (translate non-English search terms). It owns command/category search and task
recommendations. Respect its @CK_OUTPUT_TYPE marker: comprehensive-docs, category-guide,
command-details, search-results or task-recommendations. For an explicit catalog/document request,
show the complete relevant output then examples; for a focused question, answer the relevant part with
source pointers. Do not launch recommended heavy skills unless separately authorized.
The help backend's config/settings/options/switches/env queries use
node .claude/scripts/ck-config-help.cjs for schema-generated settings help. This is help, not mutation.
/plan → /plan --mode=execute is the planning flow; /feature-implement is standalone.
node .claude/scripts/workflow-mode.cjs --show
(add the actual --session=<id> when known). Report its source and missing-session note. Never
infer the current session's value from files alone./framework-config --mode=workflow ask|auto|off directive
(optional --scope=session) is applied by the route hook. Old workflow-mode: directives remain
compatible. If natural language did not form a directive, run the helper with the requested mode,
--set-session and the real session id. Never fabricate an id. Without an id/hook, follow the
requested mode in this conversation and disclose that runtime session persistence is unavailable.--scope=user (or legacy --save), run the helper <mode> --save; checkout (or
--save --local) adds --local. It refuses unless git reports the file ignored. A prompt directive
ending in --save also sets the session and saves the user preference, with independent outcomes.portability.workflowRouteMode.
Do not change a tracked file for a personal/session request. Resetting a session preference removes
the preference using the helper --reset-session and the real session id, then reads the effective mode again;
do not remove the session directory or other hook records.Resolve the requested preference, apply it only to the selected scope, and report the effective result.
From an updated checkout, use these requests in Claude/OpenCode; Codex uses $framework-config:
/framework-config --mode=settings show framework runtime
/framework-config --mode=settings quiet --scope=user
/framework-config --mode=settings off --scope=checkout
/framework-config --mode=settings on --scope=checkout
/framework-config --mode=settings reset framework runtime --scope=checkout
/framework-config --mode=settings apply framework runtime
/framework-config --mode=settings use lightweight commit reviews --scope=checkout
/framework-config --mode=settings choose commit reviews by risk --scope=user
/framework-config --mode=settings make <skill-name> explicit-only --scope=team
Replace <skill-name> with the registered skill name. Each line is a separate request, not a setup
sequence to execute together. Runtime requests use the CLI below, including apply to register and
reconcile this checkout without changing its preference. In --mode=workflow, off changes routing
only; it does not select frameworkRuntime.mode: off.
Work from the adopting project's root; use that project's copied framework, never an authoring-repo path.
node .claude/scripts/ck-config-help.cjs --json.
For project options, run node .claude/skills/project-config/scripts/project-config-help.cjs --search=<term>
or --section=<section>. Read only the owning documentation relevant to the question..claude/scripts/ck-help.py with the available Python
interpreter (py -3 on Windows) and the relevant English search terms, or read .claude/docs/README.md
and its linked topic. Present the relevant answer and usable examples; no need to start another skill..claude/scripts/lib/workflow-routing-config.cjs to resolve paths and effective runtime choices.
Its resolveProjectConfigPath(root) honors portability.projectConfigPath in .claude/.ck.json;
resolveUserConfigPath() resolves the actual home; resolveLocalOverridePath(root) resolves the checkout.
For skill activation, resolveSkillAutoTrigger({rootDir: root}) returns enabled, source and paths.
For workflow routing, use resolveWorkflowRouteMode and report any session-specific limitation.~/.claude/.ck.json;
checkout → .claude/.ck.local.json. Verify checkout-local config is git-ignored before writing it;
if not, add only its ignore entry to the project's ignore file as part of the requested personal setup.validateConfig(candidate) from
.claude/hooks/lib/project-config-schema.cjs for team config, or validateCkConfig(candidate) from
.claude/hooks/lib/ck-config-schema.cjs for personal config. Report validation errors without saving.
Use normal host file tools or a small properly quoted script; never interpolate user text into shell code.Use the platform-neutral Node CLI from an updated checkout on Windows, macOS or Linux:
node .claude/scripts/framework-mode.cjs show --json
node .claude/scripts/framework-mode.cjs quiet --scope=user
node .claude/scripts/framework-mode.cjs off --scope=checkout
node .claude/scripts/framework-mode.cjs on --scope=checkout
node .claude/scripts/framework-mode.cjs reset --scope=checkout
node .claude/scripts/framework-mode.cjs apply
show --json inspects all layers and the winning mode without mutation. Execute only the requested
change with node .claude/scripts/framework-mode.cjs <mode> --scope=<scope>:
on: normal framework hooks.quiet: no hook-injected instructions, routing, ledgers or automatic context preparation. Commit review and doc checks, formatting and notifications stay active.off: all shipped framework hooks return before parsing input or doing work; the CLI also disables this checkout's framework skills in Claude and Codex native discovery.apply: reconcile native skill visibility with the effective mode without changing the preference; register this checkout for later global changes.reset: remove only this layer's frameworkRuntime.mode, exposing the next layer.Scopes are checkout (default, ignored .claude/.ck.local.json), user (every project using this version, ~/.claude/.ck.json) and team (configured project config). The same Node command works on Windows, macOS and Linux. A missing team config still needs a valid project identity. The CLI validates and reads back changes, refuses malformed existing settings and unignored checkout settings, and reports a masking environment override.
Precedence: default on → team → user → checkout → CK_FRAMEWORK_MODE. No mirror regeneration is needed for the mode change. Use the CLI rather than editing JSON for a complete toggle: it writes personal native overrides outside Git, preserving the team profile and previous settings. Off also removes explicit Claude/Codex skill invocation; restore with this Node command. On/quiet/reset restore the previous profile. Host tool permissions and repository Git/CI hooks remain separate. OpenCode currently receives hook control only.
User-scope native changes cover registered existing checkouts; run apply once in each new checkout after its mirrors exist. Older framework copies must be updated before they honor this setting. Direct JSON/environment changes need apply for native visibility, as do newly added skills or a changed Codex home. The recovery registry is ~/.claude/framework-runtime-projects.json; do not delete it while disabled. Conflicting native skill settings fail before the preference is saved. Do not remove user settings to silence a refusal. Report missing checkouts, a masking preference, recoveryPending and any unsupported host explicitly. Read .claude/docs/configuration/README.md#framework-runtime-mode for recovery and scope limits.
Native settings require fresh Claude/Codex sessions; already delivered context cannot be withdrawn. Read .claude/docs/configuration/native-session-smoke.md when verifying native discovery and explicit invocation. Enabled skills retain their own checks, including commit/PR reviews.
“Turn off hook context injection” maps to quiet; “turn all machinery off” maps to off. Do not change a global preference without a request for user/global scope. Keep CLAUDE.md and AGENTS.md project-only; never put a framework-mode notice or fallback instructions into them.
| Request | Exact setting | Behavior |
|---|---|---|
| “Set framework runtime to on / quiet / off” | frameworkRuntime.mode: "on" / "quiet" / "off" | Use the runtime CLI to reconcile hooks and native discovery at the requested scope; default on |
| “Don't automatically start heavy skills” | portability.skillAutoTrigger: false | Ask once to run a suitable matched skill or skip and execute directly; named requests, required calls and authorized workflow steps remain allowed |
| “Restore automatic skills” | portability.skillAutoTrigger: true, or reset if requested | Later valid preference wins; default true |
| “Use lightweight commit reviews” | commit.reviewMode: "lightweight" | Default: automatically run commit-review --fix-loop at every commit risk tier, at most two rounds; no review-choice question |
| “Choose commit reviews by risk” | commit.reviewMode: "risk-based" | Low skip / Medium why-review / High changes-review; PR whole-branch reviews always retain this policy |
| “Workflows should ask / recommend the full workflow first / only run explicitly” | portability.workflowRouteMode: "ask" / "auto" / "off" | Separate workflow routing preference; ask and auto both ask before a self-selected workflow starts |
| “Make these skills explicit-only for the team” | skillProfile.commandOnly in project config | Merge only the requested skill names and regenerate native host projections; profile guards still apply |
Skill-auto-trigger precedence: default → team → user → checkout → CK_SKILL_AUTO_TRIGGER.
Commit review precedence: default lightweight → team → user → checkout → CK_COMMIT_REVIEW_MODE.
After showing, setting or resetting commit.reviewMode, run node .claude/scripts/lib/commit-review-policy.cjs
from the requested checkout root and report its winning mode/source. Tests, exact-candidate review
receipts, required project checks and explicit Git authority remain unchanged; invalid settings fail closed.
Native manual-only selection is team-scoped: for a requested team profile change, merge the named
skills into or out of skillProfile.commandOnly in the configured project config, preserving its
preset and unrelated list entries. Do not save this team-only profile in personal runtime overrides.
Validate with resolveProfile(root, candidate) from .claude/scripts/sync-skill-profile.cjs and the
team schema before saving. Report warnings; do not save if validation throws or refusals is nonempty.
Then run node .claude/scripts/sync-skill-profile.cjs and
node .claude/skills/sync-codex/scripts/run-codex-sync.mjs to generate host projections. Keep required
dependencies callable; never set allowHidingCalledSkills just to silence a refusal. Removing the
owned list entries and syncing restores invocation defaults. Do not hand-edit generated host files.
commit, pull-request and this configuration/help entry remain eligible under restricted selection.
An explicitly requested workflow authorizes its scoped planned skill calls, including later/resumed steps
and required nested calls. Commit and pull-request select tests and review under the commit skill's
decision policy without a test/review question. commit.reviewMode selects commit review depth;
skill auto-trigger and runtime preferences never approve a skip. Required skills selected by that
policy remain callable even when auto-trigger is disabled.
For a suitable unrequested skill match, false asks once to run that skill or skip and execute directly,
then waits for the answer. Confirmation authorizes its scoped dependencies; Skip keeps direct execution
and required checks without re-asking for the same task. No match proceeds directly without a question.
Named requests and already authorized steps need no skill-choice question.
This runtime switch is model guidance, not a hard permission boundary.
Examples of automatic matches: “How do these .claude skills work?”, “What can I configure in this .codex framework?”, “Disable heavy skill auto-trigger for me”, “Explain framework hooks”, and “Which setting controls workflows?”. No file is changed by the questions.
Goal: Explain framework options and apply only requested settings at their correct scope, with validated saved and effective results. MUST ATTENTION Route: choose mode → discover keys/layers → answer or apply the requested change → validate/read back → report effective value/source. Keep questions read-only, preserve unrelated keys, and route project facts through project-config. Missing session identity limits persistence; configuration does not authorize heavy tasks or waive their gates.
まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
[Documentation] Use when generating, smart-merging or refactoring the root AI context (CLAUDE.md, AGENTS.md) and its Claude/Codex projections. Config file only: project-config.
日本語の概要は準備中です。原文の説明を表示しています。
[Code Quality] Use when a workflow step or the user asks for AI code or plan review: LLM calls, prompts, agents, RAG, evals and guardrails. --mode={code|plan}; --report-only is read-only.
日本語の概要は準備中です。原文の説明を表示しています。
[Architecture] Use when a workflow step or the user asks for architecture design, compliance review, scalability grading or whole-project audit via --mode={design|review|scalability|full}.
日本語の概要は準備中です。原文の説明を表示しています。
[Content] Use when a workflow step or the user asks for product/feature ideation with ranked opportunities and validation tests. --mode={roadmap|scope}.
日本語の概要は準備中です。原文の説明を表示しています。
[Content] Use when a workflow step or the user asks for business viability: Business Model Canvas, financials, risks and go-to-market recommendation.
日本語の概要は準備中です。原文の説明を表示しています。
[Code Quality] Use when a workflow step or the user asks for a focused code-change review: staged, unstaged, branch diff or simple pull request. --fix-loop fixes and re-reviews. Complex pull request: workflow-review-changes; rationale: why-review.
日本語の概要は準備中です。原文の説明を表示しています。