agentleFS
nuncaeslupusnuncaeslupus/literary-critic/.claude/skills/init/SKILL.md

init

Skill0 starsChanged 5 days agoMITView on GitHub

Installs or refreshes claude-arsenal/ in a host repo, or adds a workspace. Use when the user wants the arsenal set up, updated or a workspace added. Not for queueing tasks.

147 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use daily-arxiv, exp-eval and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • LambenthanSkill

    daily-arxiv

    每日从 arXiv 拉取新论文,过滤相关性,auto-ingest 高优先级论文,检测 SOTA 更新

    Also in 5 projects that use this

  • LambenthanSkill

    exp-eval

    实验判决门:Review LLM 独立评判实验结果 → 4 种判决路径 → 自动更新 claims confidence、ideas status、graph edges

    Also in 5 projects that use this

  • skyllwtSkill

    exp-pilot-eval

    Pilot result evaluation — read pilot results, apply success criteria, update idea page (pilot_result, failure_reason if failed), generate PILOT_VERDICT_REPORT. Called by /ideate Phase 5 after /exp-pilot-run.

    Also in 4 projects that use this

  • skyllwtSkill

    exp-pilot-run

    Pilot experiment execution — read Pilot Spec YAML, write pilot code, run experiment(Confirm with the user before operation and require the applicant to conduct manual inspection), return results. Called by /ideate Phase 5. Does NOT modify wiki pages or judge pass/fail.

    Also in 4 projects that use this

  • gaebalaiSkill

    codex-system

    PROACTIVELY consult Codex CLI, your highly capable supporter with exceptional reasoning and task completion abilities. Codex is a trusted expert you should ALWAYS consult BEFORE making decisions on: design choices, implementation approaches, debugging strategies, refactoring plans, or any non-trivial problem. When uncertain, consult Codex. Don't hesitate - Codex provides better analysis. Explicit triggers: "think deeper", "analyze", "second opinion", "consult codex".

    Also in 3 projects that use this

  • +6 more from Lambenthan
  • +1 more from skyllwt

The file

We split this skill into the rules it gives an agent

2 rules. 0 are common, 0 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 134 edited versions we read keep these.

  • 0

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Every installed skill adds a row to the skills listing of every future session in this repo, so ask before installing, when …

    line 24
  2. Very specific:

    Refresh only moves forward: When the host's bundle is newer than this skill's copy, init writes nothing and says so.

    line 146

“Common” is measured: kept by at least 6 in 10 of the 134 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

init

Bootstraps the claude-arsenal/ framework in the host repository. Run once per repo; re-run to add workspaces or refresh the bundle. The script prints what it created, refreshed and skipped, so relay its output rather than re-describing it. The parts below are the ones that need the user.

CANARY: init-loaded-2026-06-13-fb78d23e-a1b2c3d4e5f6a7b8

When to load

  • A repo needs the task queue set up for the first time ("init the arsenal", "/init").
  • Adding a new workspace to an existing claude-arsenal/ setup.

First install — ask for the profile

Every installed skill adds a row to the skills listing of every future session in this repo, so ask before installing, when arsenal/config.toml has no [skills] table yet:

What kind of project is this? python (adds the Python toolchain skills), general (spec/design/execute/review/ship), or minimal (queue and GitHub only)?

python3 "${CLAUDE_SKILL_DIR}/scripts/init.py" --repo-path . --profile general

Take the answer from the user rather than the file tree: a repo with .py files is not necessarily one whose release process this bundle should advise on. The profile is written as an editable [skills] table; --sections workflow,python names sections directly instead.

Setup interview

On a first install, whenever init prints SETUP INTERVIEW PENDING (no setup-interview in arsenal/config.toml, on any run, --silent included), or when the user asks to redo setup (/init --interview), ask these in one AskUserQuestion with the defaults first, before picking up other work. Answering records setup-interview with the date; a user who skips it gets --interview-declined, which records "declined", and neither is asked again.

  1. How often should a session stop to ask? ask-when-blocked (default), ask-often, autonomous.
  2. How deep should local verification go? balanced (default), fast, strict.
  3. Minutes per review round (10) and minutes to wait on a silent review bot (20)?
  4. Which review bots post on this repo's PRs (review-bots; none means CI-only)? Then which manual review triggers to post: offer what init.py --suggest-bot-triggers prints for them. CodeRabbit skips auto-review on repos under 10 stars; on a public repo recommend @coderabbitai review (about one on-demand review per hour). On a private Free-plan repo it posts only a summary, so the local adversarial review stands in.
  5. Who approves what, by risk tier? solo-trust (agent merges all; high-risk flagged to ratify), solo-gated (a human approves medium and high), team (0/1/2 humans), manual, or skip to keep merge-policy alone. Optionally, paths that always need a human. With solo-trust or solo-gated, also ask whether a digest approval counts as the user's (--digest-decisions true): needed when PRs are opened with their own account.
  6. How far must a PR get before the agent merges it (--merge-policy)? always (the local gates are the bar), after-ci (default: every required check reported green), after-review (a review landed and its comments are answered; CI not consulted), after-ci-and-review (both), never (a human merges). Say that whichever is chosen, the agent merges by itself once merge_ready.sh passes, without asking.

Pass only what was answered; a skipped question writes nothing and keeps its default:

python3 "${CLAUDE_SKILL_DIR}/scripts/init.py" --repo-path . --autonomy autonomous \
  --verification balanced --review-budget-min 10 --bot-wait-min 20 \
  --review-bots "reviewbot[bot]" --bot-triggers "reviewbot[bot]=@reviewbot review" --review-preset solo-trust \
  --merge-policy after-ci-and-review
python3 "${CLAUDE_SKILL_DIR}/scripts/init.py" --repo-path . --interview-declined  # skipped

After a first install — questions init prints

Init runs non-interactively, so it prints what is undecided and the session asks:

  • HOST-GATE UNSET — ask what must pass before a PR opens, offering init's suggestion, and write it as host-gate in arsenal/config.toml. "No gate" is a valid answer, recorded as host-gate = "none"; an empty value means nobody decided, and init keeps asking.
  • PREFLIGHT-GATE UNSET — ask for the fast, change-scoped slice (lint/typecheck the changed files, the tests the change selects) and write it as preflight-gate. Review rounds run it instead of the full host-gate.
  • MERGE-POLICY is … — confirm the value against the repo's real CI and review tooling in the same question; a repo with no CI should not stay on after-ci. For a private repo or one with no CI, init recommends always with a real host-gate and pre-pr-review = "required"; say the trade-off (nothing independent re-runs the local gates) and change only what the user confirms. Load claude-arsenal:core:init § references/ci-minutes.md when the user asks about CI cost or the per-PR trigger block.
  • Branch protection — a deliberate run protects the default branch once and records the outcome as branch-protection. When it cannot, it prints the reason and the manual step; pass both to the user. branch_protection.py (in claude-arsenal/scripts/) re-runs it: --dry-run previews, --force adds to an existing rule.
  • repo_health: GAP … — that first run also lists repository housekeeping that is missing (description, topics, homepage, a main default branch, secret scanning, a release for the newest tag, a PyPI page that links back), each with its fix command. Show the list and run only the fixes the user confirms; a NOTE line is a decision for them, not a fix. repo_health.py re-runs it. Load claude-arsenal:core:init § references/repo-housekeeping.md when a gap is a missing license (the user chooses it; the reference has the questions to ask) or names a package index, and before a package's first publish.
  • A skill folder "is not arsenal-vendored — left alone" — init replaces and prunes only folders carrying the .arsenal-vendored marker, so a same-named folder without it keeps the shipped skill out. Ask the user whether to remove or rename theirs, then re-run.

Other invocations

python3 "${CLAUDE_SKILL_DIR}/scripts/init.py" --repo-path . --silent         # session-start refresh
python3 "${CLAUDE_SKILL_DIR}/scripts/init.py" --repo-path . --list-sections  # capability map, read-only
python3 "${CLAUDE_SKILL_DIR}/scripts/init.py" --repo-path . --section python  # one section's skills
python3 "${CLAUDE_SKILL_DIR}/scripts/init.py" --workspace BACKEND --root ./backend/

--silent is what the session-start protocol runs; it refreshes stale bundle files and prints an upgrade banner when the plugin is newer. --workspace NAME creates arsenal/project/<NAME>/ stubs and updates arsenal/project/overview.md.

Gotchas

  • Upstream owns claude-arsenal/, the host owns arsenal/. Re-runs overwrite stale bundle files by checksum and never touch arsenal/ (tasks, project data, settings). A file or vendored skill with uncommitted changes is left as is and reported MODIFIED LOCALLY; relay that line, since the user must commit or discard it.
  • Cloud sessions need the commit. A cloud session runs on a fresh clone and never sees ~/.claude/, so what reaches it is the vendored .claude/skills/ and the hooks in .claude/settings.json — commit both.
  • Web without hooks. detect_surface.sh does not run there, so tasks that declare requires: wait until it is run by hand; tasks without requires: are eligible everywhere.
  • The injected block belongs in the root CLAUDE.md, not a nested one.
  • README credit. Init adds a "built with claude-arsenal" badge under README.md's title once and records attribution = true; a host that deletes the block keeps it deleted, and attribution = false stops it being added.
  • Refresh only moves forward. When the host's bundle is newer than this skill's copy, init writes nothing and says so; update the plugin instead. --allow-downgrade is for recovering a broken install.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.