القائمة
Adopt Agentic Repos in a project, writing its per-repo config + rules layer by extracting the codebase's real conventions. Say "ar-install" or "make this repo agent-ready".
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Record an improvement, learning, fix, or refinement to the Agentic Repos framework. Invoke when a new pattern emerges, a bug is found in a skill/agent/rule, or a workflow refinement is identified during work. Handles version bumping, CHANGELOG updates, and cross-file consistency. Also user-invocable via "ar-add-improvement" or "update framework".
Install or update context-processing skills (discussion, email, meeting, specs, state) into a Docs Project. Separate from ar-taskflow - does not touch ar-taskflow skills or config.
Install or refresh the global Agentic Repos layer on this machine (ar-* skills, scripts, session hook, default-branch guard). Say "ar-install-tools" or "setup tools".
Reviews a PULL REQUEST of the Agentic Repos framework SOURCE repo itself for project-based noise. Its input is a framework PR, on trigger it FIRST asks which agentic-repos PR to review (framework PRs only; it refuses PRs from any other repo). Proves the PR introduces ZERO project-based noise into the project-agnostic framework, installed artifacts (rules/, skills/, agents/, templates/, setup.md) AND docs/. Catches a source project's identity (real names, packages, developer paths, tickets, people, dated incident notes), whole files bound to one project (a <project>-feedback.md mining doc, anything named after or dominated by a single product line, these belong in the source project, not the framework), stack idioms leaking into stack-agnostic skill/agent bodies, ungenericized business-domain examples, rationale/version-history bloat, mis-tiered rules, and version-file drift. Use whenever you want to review/self-review an agentic-repos PR for "does this leak or bind a project into the generic framework," befor
Refresh an adopted project's per-repo config + rules layer to the current framework version. Say "ar-upgrade" or "refresh the project rules".
Assess how agent-ready a repository is and raise it. Reads the actual codebase, extracts its real conventions into rule files, wires the config seam, and reports a readiness scorecard. Say "ar-agent-ready", "make this repo agent-ready", or "extract the rules for this project".
| name | ar-install |
| description | Adopt Agentic Repos in a project, writing its per-repo config + rules layer by extracting the codebase's real conventions. Say "ar-install" or "make this repo agent-ready". |
| disable-model-invocation | true |
Make a repo agent-ready by writing ONLY its per-project config + rules layer: config_hints.json, AGENTS.md, .claude/skill.config, an autonomous .claude/settings.json (permissions only), coding rules extracted from the actual codebase into standards_location, and PR/commit templates.
This command does NOT install the global procedure. The ar-* skills, the devkit agents (a_sag_*) and atomic skills (a_sk_*), the worktree helpers (a_g_worktree_*), the framework scripts, the SessionStart hook, and the default-branch guard are all installed once per machine by install.sh. This command reads that global layer. It never copies skills/agents into the target, never installs scripts, and never writes hooks (the global session hook and default-branch guard already cover every repo).
Follow setup.md, referenced below by section name.
.claude/config_hints.json).config_hints.json already exists, use ar-upgrade instead._install_config.json.standards_location, adapt framework rules on top.config_hints.json, AGENTS.md, .claude/skill.config, .claude/settings.json, CLAUDE.md, templates, optional ERD.All agents communicate through files on disk (_install_config.json, _codemap.md, _install_manifest.json), not conversation context.
Bash rule: run each command individually (no &&/; chaining), use absolute paths, no cd + relative.
Run setup.md -> Precheck: Global Layer Present. If the ar-* skills or agentic-devkit are missing, STOP and tell the user to run ./install.sh from the framework root first. Do not install the global layer from here.
Ask ONLY this single free-text question. No menu, no numbered list, no suggested paths:
What is the full path to your target project?
Validate the directory exists. Store as TARGET_PROJECT. Then check whether it is already adopted:
ls {TARGET_PROJECT}/.claude/config_hints.json 2>/dev/null
If it exists, tell the user this project is already adopted and to use ar-upgrade instead. Stop here.
Follow setup.md -> Step: Create Install Branch. Store BRANCH_NAME and DEFAULT_BRANCH for the summary.
Follow setup.md -> Step: Validate Prerequisites (gh check, read the canonical FRAMEWORK_VERSION).
Follow setup.md -> Step: Gather Project Configuration (project name, tracker, namespace(s), standards_location) and -> Stack Detection (Positive Evidence) for stack + applicable_rule_dirs.
Follow setup.md -> Step: Handle CLAUDE.md and Existing AI Files. Save any merge-worthy content for the Config Writer.
Write _install_config.json to the target root with everything gathered (target/framework paths, project name, tracker, namespace(s), standards_location, stack, applicable_rule_dirs, saved AI content, mode: "fresh", framework_version). Schema in setup.md -> File-Based Handoffs. Present the detected stack + applicable rules and continue.
Only for a true greenfield repo (no existing rules dir, no prior AI files, no .claude/). Follow setup.md -> Step: Greenfield Bootstrap. Skip otherwise.
This is what makes the repo agent-ready. Follow setup.md -> Rule Extraction (the heart of agent-ready) and Step: Extract and Adapt Coding Rules:
a_sag_codebase_explorer (Task) with the target path + _install_config.json, which writes _codemap.md.standards_location (cite real files, use the project's own idioms). Do not merely copy generic templates.requires: rules by probing for the symbol.Append every written rule file to _install_manifest.json.
Run the Config Writer (setup.md -> Config Writer), then settings, templates, and optional ERD:
config_hints.json: setup.md -> Step: Create config_hints.json (schema + test_command detection).AGENTS.md: setup.md -> Step: Create AGENTS.md (scanner-compat rules; build commands + structure from _codemap.md, rule list from _install_manifest.json)..claude/skill.config: setup.md -> Step: Create .claude/skill.config..claude/settings.json: setup.md -> Step: Write Autonomous Settings (autonomous permissions from the template, merged not clobbered; no hooks, the global layer owns those).CLAUDE.md: the standard @AGENTS.md pointer.setup.md -> Step: Setup Templates.setup.md -> Step: Generate ERD Documentation..gitignore / .dockerignore: setup.md -> Step: Update .gitignore / .dockerignore.Follow setup.md -> Step: Verification (contamination + reference check, fresh context, re-detect stack independently). Fix flagged files and re-run until PASS.
Extraction plus merging framework rules under any pre-existing project rules can leave overlap. Invoke ar-optimizer over standards_location to catch redundant or duplicated rules, alwaysApply overuse, and stale cross-references, then reconcile its findings. This is the same sweep ar-agent-ready runs, so adoption leaves no duplicated rules behind.
Follow setup.md -> Step: Summary to report what was written, the verification verdict, and next steps (ar-init-skills, ar-init-mcps, ar-taskflow, push + PR). Then run setup.md -> Cleanup to delete the temp handoff files.