Index of every page on this site: llms.txt
All skills
Run it yourself

Picks the skill, or the short chain of skills, for a request that could match more than one or that needs several in order: a bug to diagnose and fix, a feature from spec to merge request, a release, a cleanup, a design. Use when it isn't clear which skill owns a request, or a request spans diagnosis, code, tests, review and shipping. Not needed when one skill plainly fits - load that one directly.

  • 1 file
  • 4.8 KB
  • Updated 3 weeks ago
  • GitHub
Use this Skill: https://skilld.dev/gh/avik-creator/agent-skills/which-skill

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈104 tokens for metadata: the name and description. ≈1.1k when used: this file.

Description uses 5.2% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈415 characters, or 5.2%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Which skill

Route the request to the fewest skills that do it, then follow those skills. This skill holds no rules of its own about the work. Why: a rule copied here drifts from the skill that owns it.

Scope

  • One skill: when one fits, run it. A chain is for work that really has several stages, never to look thorough.
  • Two fit equally: ask one question, then route.
  • Before routing: ask every blocking question in one turn. Never invent a fact. Put an unknown under Open questions and say what it blocks. Replies are 1–2 lines; a paragraph only when the user asks.

Pick

The request Skill
Build a whole feature after one round of questions: ask, prototype, build, prove, review, PR deliver
Write or change code, or a test that was asked for coding
Write tests for a change, fix a flaky test, "what could break here" testing
Manual test cases, QA coverage, how to test a flow by hand test-cases
Find why something fails: a crash, wrong result, red CI, a production error debug
Review a diff, a branch or a pull request, what a change could break, or a strict or thermo-nuclear quality review review
Refactor, clean up, deduplicate, or standardize an area refactor
Talk through an idea, a PRD, a spec, a system design, an architecture or scale review, a runbook, a README design-docs
A throwaway prototype to answer one design question prototype
Auth, input, uploads, secrets, a scanner finding, a threat model security
Logs, metrics, traces, alerts, an SLO, a slow endpoint observability
SQL, schema, migrations, transactions, pooling database
Prove a change works in the real app, or set up verification for a repo prove
Drive a real browser or app to check something automation
Finish a change into a PR or MR, a rollback plan, a release checklist, a tag ship
Write or update a PR or MR title or body, or draft review replies writing-pr
Create or edit a skill writing-skills

Alongside, not instead:

  • Load go, where installed, with any skill that touches .go files.
  • Load database when the work reaches SQL.
  • Load security when it crosses a trust boundary.
  • Load observability when it adds a path on-call must see.

Read the chosen skill and follow it.

Chains

Run the stages in order. Each skill's own stops still hold. Pause for the user's yes before any stage that writes code, pushes, or touches a shared system.

  • Bug: debug finds and proves the cause, and stops. After a yes, coding fixes it, with the reproduction test written red first through testing. Then review. Add test-cases only when no automated test can reach it.
  • Feature: design-docs when there is no agreed spec, then coding, then testing, then review. ship comes when the user wants it pushed. When the user wants all the questions first and the rest done with proof, deliver runs this chain with a prototype and prove.
  • Release: review, then ship. Ship's done gate, its interview, and writing-pr for the description all run inside ship. A migration follows database and ship's deploy notes.
  • Cleanup: refactor, which applies each approved step through coding.
  • Design: design-docs for the system design, then a spec in design-docs if one is still needed, then coding after approval. prototype first when one question needs trying before deciding.
  • Slow: observability measures it first. database for a slow query. Then coding for the change the measurement points to.
  • Live incident: debug, reading production only. A mitigation is a proposal for the user. The fix then follows the bug chain.

Limits that hold across chains

  • debug and refactor change no code until the user says yes.
  • review fixes proven findings only in the user's own work, and asks first before a schema, API, security or user-visible change. Someone else's PR is report-only.
  • ship opens the PR or MR and stops. The user merges, deploys and tags.
  • design-docs doesn't write code.
  • No skill pushes to the default branch or writes to production.

Replies

Talk to the user in simple English: short sentences, everyday words, and a technical term only where it is needed, explained the first time it appears.

Name the skill or chain in one line, then run the first stage.

Source: SKILL.md on GitHub

No rule matched.

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No third-party reports yet.

Signed by skilld at a0324a3. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 4 days ago.

Activeupdated 3 weeks ago

README badge

README badge for avik-creator/agent-skills/which-skill