fix(skills): sharpen craft-skill audit checks

This commit is contained in:
2026-09-05 17:59:24 -04:00
parent 9b6dabd171
commit c004073137
2 changed files with 12 additions and 0 deletions

View File

@@ -35,6 +35,8 @@ For user-invoked skills, the description remains valid metadata.
### Context pointer ### Context pointer
A short reference that tells the agent when to load out-of-context material. A short reference that tells the agent when to load out-of-context material.
The pointer's wording controls whether the agent reaches the target.
Fix weak wording before moving the target inline.
The description is the top-level context pointer for a model-invoked skill. The description is the top-level context pointer for a model-invoked skill.
A link to a sibling reference file is a lower-level context pointer. A link to a sibling reference file is a lower-level context pointer.
@@ -154,4 +156,6 @@ Move reference down the information hierarchy.
An instruction does not change the agent's default behavior. An instruction does not change the agent's default behavior.
No-op instructions spend attention without adding control. No-op instructions spend attention without adding control.
Judge no-op instructions against behavior, not style.
Delete a no-op sentence instead of shortening it.
Cut no-op sentences unless they anchor a target-project convention. Cut no-op sentences unless they anchor a target-project convention.

View File

@@ -43,7 +43,10 @@ Finish when one branch matches the user's requested change.
Put active **steps** in `SKILL.md`. Put active **steps** in `SKILL.md`.
Keep shared **reference** near the steps that need it. Keep shared **reference** near the steps that need it.
Move large or branch-specific reference behind precise **context pointers**. Move large or branch-specific reference behind precise **context pointers**.
Front-load the **leading word** in descriptions and other context pointers.
Write pointer text for the condition that loads the target.
Give each step a checkable **completion criterion**. Give each step a checkable **completion criterion**.
Cut sentences that do not change agent behavior.
Use relative paths for all skill-local references. Use relative paths for all skill-local references.
Finish when each branch has ordered steps, resolved references, and checkable completion criteria. Finish when each branch has ordered steps, resolved references, and checkable completion criteria.
@@ -68,8 +71,11 @@ Finish when one branch matches the user's requested change.
3. Apply the checklist. 3. Apply the checklist.
Quote each passage that fails a glossary check. Quote each passage that fails a glossary check.
Check **invocation** against the intended harness. Check **invocation** against the intended harness.
Check **context pointers** for weak trigger wording.
Check **information hierarchy** for buried steps or disclosed required material. Check **information hierarchy** for buried steps or disclosed required material.
Check **completion criteria** for vague or weak stopping points. Check **completion criteria** for vague or weak stopping points.
Check **leading words** for repeated explanations they can replace.
Check **pruning** for sentences that explain default model behavior.
Check **duplication**, **sediment**, **sprawl**, and **no-op** instructions. Check **duplication**, **sediment**, **sprawl**, and **no-op** instructions.
Finish when every finding has quoted evidence or no finding remains. Finish when every finding has quoted evidence or no finding remains.
@@ -89,7 +95,9 @@ Finish when one branch matches the user's requested change.
Verify the name is portable and matches the directory. Verify the name is portable and matches the directory.
Verify the description is specific and under 1024 characters. Verify the description is specific and under 1024 characters.
Verify every skill-local reference resolves with a relative path. Verify every skill-local reference resolves with a relative path.
Verify context pointers name when to load referenced material.
Verify disclosed files exist. Verify disclosed files exist.
Verify each retained sentence changes behavior or records target-project policy.
Verify no stale harness-specific assumption remains. Verify no stale harness-specific assumption remains.
Finish when each check passes or the final report names checks that could not run. Finish when each check passes or the final report names checks that could not run.