agentleFS
arxdsilvaarxdsilva/ai-loops/skills/init/SKILL.md

init

Skill0 starsChanged 7 days agoMITView on GitHub

Set up loops in the current repository (any language pack) - detect the languages, create .loops/config.json, install AGENTS.md and CLAUDE.md pointers, and check the tools the detected packs need. Use when the user runs /loops:init or asks to set up loops.

145 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

5 rules. 0 are common, 2 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 133 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

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

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

    .loops/config.json: If absent, create it with these defaults (JSON has no comments.

    line 15
  2. Distinctive:

    If it exists, show it and ask before changing anything.

    line 33
  3. Very specific:

    AGENTS.md: If absent, copy $(loops path templates)/ROOT_AGENTS.md to ./AGENTS.md.

    line 35
  4. Distinctive:

    CLAUDE.md: If absent, create it with exactly:

    line 36
  5. Very specific:

    If present without @AGENTS.md, offer to append that line.

    line 42

“Common” is measured: kept by at least 6 in 10 of the 133 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

/loops:init

Idempotent setup for the current repo. Never overwrite an existing file without asking; show a diff or the proposed text and use AskUserQuestion.

  1. Preconditions. git rev-parse --show-toplevel (cd there). 1a. Languages. Run loops lang detect and show each detected pack with its root and variant (loops lang list shows every available pack).
    • Nothing detected, or a root is wrong (an unusual layout, a monorepo, a manifest deeper than detection looks): explain that languages in .loops/config.json pins packs and replaces detection, e.g. "languages": [{"id": "python", "root": "services/api", "variant": "uv"}, {"id": "go", "root": "."}] (root defaults to .; variant picks a build tool, loops lang show <id> --variant v). Offer to add it, then re-run loops lang detect.
    • A language with no pack: say that loops still runs the workflow there, but the gate checks nothing for it. Continue setup.
  2. .loops/config.json. If absent, create it with these defaults (JSON has no comments; explain each key to the user in your reply):
{
  "layers": {},
  "languages": [],
  "ask": [],
  "gate": { "skip": [], "vuln": false },
  "memory": { "workspace": "", "project": "" },
  "parallel": { "max_agents": 100, "agent_ram_gib": 1, "cpu_slots_per_core": 4, "max_test_runs": 0 }
}
  • languages: pinned packs (step 1a); [] means detect.
  • layers: object mapping layer keys (domain, application, ports, adapters, inbound, outbound, platform, compose, cmd, pkg) to lists of module-relative dir prefixes; omitted keys keep defaults. Example: {"domain":["internal/core"]}. Leave {} unless the repo uses non-standard dirs.
  • ask: the phases where the lead stops for the user (WORKFLOW §2.7): [] (the default) means it decides every phase and opens the PR itself; e.g. ["plan", "pr"] to approve the plan and the push, or "all". --ask on /loops:work, or the user's own words, override it per run. Older configs: "approval": "always" still adds plan; auto_pr is ignored (the PR opens unless pr is a checkpoint).
  • gate.skip: check names to skip, taken from the detected packs (loops lang show <id>; Go: build, gofmt, vet, staticcheck, arch, test, govulncheck). coverage and the race checks cannot be skipped. gate.vuln: include the packs' vulnerability checks (off by default).
  • memory: ai-memory scope; leave empty to derive from .ai-memory.toml or the git remote owner/repo.
  • parallel: concurrency. max_agents is the configured agent cap (default 100; the effective cap is also bounded by loops hw); agent_ram_gib and cpu_slots_per_core feed the hardware estimate; max_test_runs 0 means derive from the CPU count.
  • limits (opt-in, omitted by default; add only when the user wants caps): "limits": {"max_agents": 12, "budget": {"research": "200k", "implement": "2M"}, "agents": {"implement": 12}, "roles": {"implement": ["domain-engineer", "api-engineer"]}}. Nothing is limited and nothing asks without it. If it exists, show it and ask before changing anything.
  1. Git. Nothing to add: loops keeps .loops/ out of git itself with a .loops/.gitignore of * (run data, caches and settings are local). If the team wants to share the settings, tell the user git add -f .loops/config.json. If .loops/ files are already tracked from an older setup, suggest git rm -r --cached .loops (the user decides; never run it).
  2. AGENTS.md. If absent, copy $(loops path templates)/ROOT_AGENTS.md to ./AGENTS.md. If present, offer to append a "loops" section (the relevant headings from the template); do not replace.
  3. CLAUDE.md. If absent, create it with exactly:
# Claude

@AGENTS.md

If present without @AGENTS.md, offer to append that line. 6. Tool check. For each detected pack, loops lang show <id> lists its checks with tool, required, default and install. Run command -v for each distinct tool, plus gh (GitHub PRs) and glab (GitLab MRs), and print a table: tool, pack, present/missing, effect of missing:

  • a required check's tool: the gate fails (or uses the check's logged fallback);
  • an optional check's tool: the gate skips it and prints its install hint;
  • a default: off check (vulnerability scans): only matters when enabled. Also print each pack's race policy (race.mode and race.reason), since a policy the host cannot run fails the gate (e.g. Go's race detector needs cgo and a C compiler).
  • ai-memory: report whether MCP tools named mcp__plugin_ai-memory_ai-memory__* or mcp__ai-memory__* are available. Optional; without it memory steps are skipped.
  • loops version to confirm the binary.
  1. Validate: jq . .loops/config.json must succeed (loops warns on stderr when the file fails to parse).
  2. Report in at most 10 lines: languages and roots, files created, files left alone, tools missing, and next step /loops:work <ticket>.

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.