agentleFS
leekleek/agent-skills/skills/which-skill/SKILL.md

which-skill

Skill5 starsChanged 10 days agoMITView on GitHub

Not sure which skill to reach for? Name the situation and get routed to one.

12 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use to-issues and to-prd.

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.
  • GoogleSkill

    to-issues

    Break a plan, spec, or PRD into independently-grabbable issues on the project issue tracker using tracer-bullet vertical slices.

    Also in 3 projects that use this

  • GoogleSkill

    to-prd

    Turn the current conversation into a PRD and publish it to the project issue tracker — no interview, just synthesis of what you've already discussed.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

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

  • 0

    Common

    Most of the 10 edited versions we read keep these.

  • 3

    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. Distinctive:

    If the situation is genuinely ambiguous, ask one question that separates the branches, not a menu of everything.

    line 14
  2. Very specific:

    Test a web app from a URL and a prompt → browser-test.

    line 44
  3. Very specific:

    Write or rewrite a PR body → pr (the agent also reaches for it on its own).

    line 46
  4. Distinctive:

    Start a fresh session: when the next task is genuinely independent.

    line 73
  5. Very specific:

    handoff: when the next agent needs what this session learned.

    line 74
  6. Distinctive:

    Delegate to a subagent: when the expensive part is legwork whose *conclusion* is all you need back.

    line 75

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

Which Skill

The user does not know which skill fits. Read their situation, then name one skill and say why in a sentence. Recommend, never invoke: most of the skills below are user-invoked, so only the user can start them.

If the situation is genuinely ambiguous, ask one question that separates the branches, not a menu of everything.

The rows below are for orientation only. Before saying what a skill does beyond its row, or that a step of it can be skipped, read that installed skill's SKILL.md.

The pipeline

Most work runs one straight line (decide → spec → tickets → build) over .scratch/<slug>/. Route by where the user actually is:

They have…They want…Route to
a vague ideait stress-testedgrill-with-docs (default in an existing codebase), or grill-me for a plain interview
work too big for one sessiona map of the decisionswayfinder
an open question needing sourcescited findingsresearch
sources to watchnew chatter written upchatter-scout
a question only someone else can answera document to send themto-questionnaire
a settled conversationit written downto-spec
a specbuild ticket filesto-tickets
an inbound issue or external PRit categorised and briefedtriage
one ticketit built end to endimplement
a specthe whole thing built fast, in parallel, this sessionimplement-spec (implement <spec> all for one-by-one, no subagents)
a map, spec, or tickets, and no wish to babysitit driven to doneautopilot

Before any of it, once per repo: setup writes the tracker, status vocabulary, and domain doc layout.

Steps only a human can do (a third-party dashboard, credentials, CI secrets, a one-off cutover) → wizard scripts the walkthrough.

After the code exists

  • Review the diff → code-review. Want four independent models on it instead of one → panel.
  • Prove it in the running app → verify.
  • Test a web app from a URL and a prompt → browser-test.
  • Ship the session's work → commit.
  • Write or rewrite a PR body → pr (the agent also reaches for it on its own).
  • Review comments waiting on a PR → resolve-review-comments.
  • A PR to take all the way to merged (reviews, checks, fixes, merge) → triage-github-pr.
  • Merge or rebase blocked → resolving-merge-conflicts.
  • About to close the session and want to know what is still open → closeout.
  • A step that must wait for a deploy, merge, or commit you are about to start (a backfill after deploy) → start it, then run-after.

Something is wrong

  • Broken, throwing, failing, or slow → diagnosing-bugs.
  • A branch's CI is red → fix-ci-failures.
  • A backlog of production or review-bot issues → fix-nightwatch-issues, fix-posthog-issues, or fix-github-issues for the source they come from.
  • The shape of the code is the problem, not one bug → improve-codebase-architecture to find the opportunity, architecture-satisfaction to run the refactor loop, codebase-design for the vocabulary either one uses.
  • Suspect a specific module smells → code-smells-audit.
  • Nothing known is wrong, but you want the next unnoticed problem found, proven, and fixed as a PR → sweep.
  • Unsure whether a state model or a UI direction holds up → prototype.
  • Want a page held to your UI rules, or a new rule recorded → ui-design.

Repo upkeep

housekeeper (low-risk cleanup), repository-cleanup (branches, PRs, stashes, worktrees), scratch-status (what is still open in .scratch/, read-only), scratch-cleanup (file research into docs/, remove finished efforts), dependency-audit (Composer or npm audit), upgrade-plan (plan one library or framework upgrade), prune-tests (delete low-signal tests, one PR per area), retro (look back over this session or recent logs for environment improvements), domain-modeling (terms and ADRs), laravel-herd-worktrees (worktree plus .test URL).

Running out of context

Order matters: the cheapest move that preserves the primary source wins:

  1. Keep going while the work still fits. Most sessions never need any of the rest.
  2. Start a fresh session when the next task is genuinely independent. Nothing to carry, nothing to lose.
  3. handoff when the next agent needs what this session learned. It writes the document; you start clean with a primary source on disk.
  4. Delegate to a subagent when the expensive part is legwork whose conclusion is all you need back.
  5. Compact last. Summarising a grilling or diagnosis session replaces a primary source with a confidently-wrong paraphrase. Prefer handoff, which writes the record down before the context goes.

When nothing fits

Say so plainly and answer the question directly. Reaching for the nearest skill because one must be named is worse than none: it spends a whole workflow on the wrong shape of problem.

Two skills exist for the meta-work rather than the work: writing-for-agents (writing any document an agent consumes) and teach (learn a concept inside this workspace). wait-what is the one to reach for when this answer did not land.

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.