wutianxiang88-hash

Skill Trigger Lab — Testing skill for Claude Code

Testing community

Lint, score and live-test whether an agent SKILL.md actually gets triggered.

How to install Skill Trigger Lab

This entry records only its repository, not the path inside it, so there is no exact command to give. Open wutianxiang88-hash/skill-trigger-lab and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Skill Trigger Lab does

Lint, score and live-test whether an agent SKILL.md actually gets triggered. Deterministic offline frontmatter lint (the unquoted-colon silent-truncation trap, name/dir mismatch, missing Use-when scenarios) + trigger-collision check + seed eval plans + bring-your-own-runner accuracy. No model, no API key, CI-friendly.

Alternatives in Testing

Loki Spec Status — Check whether the spec has drifted from its lock using Loki's living-spec drift detection, and summarize the r 1.1k ★
Superpowers Evals — Behavioral eval lab (Quorum) for the superpowers project that drives real coding-agent CLIs (Claude, Codex, Ge 107 ★
Skill Trigger Test — PHẢI dùng khi: vừa tạo agent/skill mới, vừa sửa description frontmatter, nghi ngờ routing conflict giữa 2 agen 90 ★

README

skill-trigger-lab

**Lint, score and live-test whether an agent `SKILL.md` actually gets triggered.**

Every skill author hits the same wall: the skill installs fine, but the agent never loads it — or loads it on the wrong prompt. The official [anthropics/skills eval harness hit a 0% trigger rate](https://github.com/anthropics/skills/issues/556), and [claude-code #9716](https://github.com/anthropics/claude-code/issues/9716) is one long thread of "my skill won't activate". Most causes are deterministic and visible in the frontmatter *before* you ever call a model. This tool finds them offline, in milliseconds, with no API key — then, if you want, measures the real thing against your host.

$ pip install skill-trigger-lab   # or: uv tool run / npx-style one-off, see below
$ skill-trigger-lab lint my-skill/SKILL.md

## Routability score: 43 / 100 (grade D)

| sev | rule | field | problem |
|---|---|---|---|
| ✗ blocker | `colon-unquoted` | description | unquoted ': ' silently truncates the YAML value — the #1 documented cause of 0% triggers |
| ✗ high | `name-dir-mismatch` | name | name != directory; hosts register the directory and explicit invocation breaks |
| ! medium | `no-trigger-scenario` | description | says what the skill IS, never WHEN to use it |

What it does

Command	Job	Deterministic?
`lint`	14 frontmatter rules: fence position, BOM, unquoted `": "` truncation, name/dir mismatch, length caps, missing "Use when" scenario, unknown keys, empty body	✅ no model, no network
`collision`	overlap score against every installed `SKILL.md` in a directory — shows which sibling skill steals your traffic and the shared tokens	✅
`gen`	seed eval plan: should-trigger / should-not-trigger probe queries derived from the description	✅
`score`	one 0–100 routability grade (explicitly a heuristic on description quality, not a live routing measurement)	✅
`run`	live trigger accuracy: pipes each quer