checklists
暂无简介
- License 许可
- MIT
- 最近更新
- 2026-11-14
- Stars
- ★ 3
- Forks
- 0
02 资源文件 7
03 SKILL.md 正文
Checklist — skill
Audit a skill pack: SKILL.md (or skill.md, or the recipe file) and every
script, wrapper, schema, or helper it names. The skill profile weights
missing_feature highest: advertising a capability that does not exist is the
defining skill failure.
Look for
wrong_claim— the MD asserts a step, path, flag, tool, or outcome the code does not do, or does differently.missing_feature— a capability with no implementation: a script that is not in the repo, a stub, a dead link, an empty handler.- Commands that cannot run as printed. Shell snippets are the usual
offender:
$0inside a Markdown code block is the operator's shell, not the skill file; relative paths resolve against the caller's cwd, not the skill directory. bug/breaking_bug/security_issuein skill-adjacent code the MD implies is safe or correct.- Hollow absolutes — "always", "never", "automatically" — with no code behind them.
- Frontmatter vs body vs code. A
descriptionthat promises more than the body delivers is a contradiction. So is a declared tool the skill never uses, or a used tool it never declares. - Progressive-disclosure claims — files referenced for "more detail" that do not exist.
- Hollow tests and gates — a test whose name or docstring promises an
execution check while the body asserts a substring, or a CI step that greps
for a command instead of running it.
missing_feature, quoting the promise.
Do not invent missing files: if a referenced path is absent, quote the reference
from the MD and file missing_feature.
Use target: "skill" for MD quotes and target: "code" for script quotes.
Before you emit
- Sweep each quote across the tree (
rg -n --fixed-strings '<quote>'). One defect in one file is usually one defect in several. - Give same-cause findings a shared
clusterid. - List what you opened in
audit.files_read. Askillaudit with no implementation file in that list is stamped shallow. - Re-read the shared failure modes, especially hollow verification.
Emit
{ "version": 2, "target_kind": "skill", "target_ref": "<path-or-name>", "findings": [] }
04 同仓库更多 Skills
demo-deploy
Fixture skill whose documentation does not match its script.
bs-score
Use when auditing a repo, PR, branch, PR review, AI skill, agent config, or one or more prompts for claims vs truth and real defects: follow the canonical bs-score SKILL.md, emit findings JSON, then run bs-score (higher = more bullshit).
bs-score
Use when auditing a repo, PR, branch, PR review, AI skill, agent config, or one or more prompts for claims vs truth and real defects: follow the canonical bs-score SKILL.md, emit findings JSON, then run bs-score (higher = more bullshit).
bs-score
Use when auditing a repo, PR, branch, PR review, AI skill, agent config, or one or more prompts for claims vs truth and real defects: follow the canonical bs-score SKILL.md, emit findings JSON, then run bs-score (higher = more bullshit).
bs-score
Use when auditing a repo, PR, branch, PR review, AI skill, agent config, or one or more prompts for claims vs truth and real defects: follow the canonical bs-score SKILL.md, emit findings JSON, then run bs-score (higher = more bullshit).
bs-score
Use when auditing a repo, PR, branch, PR review, AI skill, agent config, or one or more prompts for claims vs truth and real defects: follow the canonical bs-score SKILL.md, emit findings JSON, then run bs-score (higher = more bullshit).