choo121600 opened a new pull request, #142: URL: https://github.com/apache/airflow-steward/pull/142
## Summary - Clears all 24 `skill-validator` SOFT warnings across 17 `SKILL.md` files (action-inventory / parenthetical-rationale / distinct-from / chain-handoff / criteria-source). - Honors the rule's *spirit*, not just its regex: sub-step enumerations (workflow steps, internal component lists, implementation actions) are moved out of frontmatter into the body where they belong. Top-level routing labels (PR dispositions, triage classes, sub-action names) stay in frontmatter as short slash-separated lists since they carry routing signal. - Preserves niche routing keywords (bubblewrap / socat / claude-code / claude-iso / settings.json / denial commands / CVE tool link / cve allocated / snapshot / locks / symlinks / hook / frontmatter / placeholder convention / prompt-injection defences / Privacy-LLM) as compact parenthetical hints in description so single-keyword user queries still route. - Net per-routing-call cost: **~−498 tokens** (−8.8% of skill frontmatter budget) vs HEAD. ## How `Distinct from` / `Counterpart to` clauses were handled Rewritten as concrete `Use after X` / `Skip when X` redirects so the LLM router gets the actionable routing decision instead of a comparison. The sibling skill name is preserved in every case, so cross-skill routing recall is unaffected. ## Verification | Check | Result | |---|---| | `skill-validate` | `OK (no violations)` (was: 24 SOFT warnings across 17 skills) | | `skill-validate --strict` against `HEAD` | `OK` — strict promotes trigger-preservation to hard-fail; passing proves no quoted phrase dropped from `when_to_use` | | Custom audit script | **76/76** quoted trigger phrases preserved verbatim (per-skill 1:1) | | Routing cue counts (`Invoke when` / `Skip when` / `Use after`) | preserved or increased on all 17 skills | | Metadata budget (1,536-char cap) | longest skill (`setup-steward`, 1,233 chars) well under cap | <img width="878" height="173" alt="image" src="https://github.com/user-attachments/assets/1c9d2dc0-3111-4917-9240-8385515a363a"; /> ## Files touched 17 `SKILL.md` files under `.claude/skills/`. Net **−22 lines** of frontmatter; bodies unchanged (existing detail files already covered the moved enumerations, so no body additions were needed). ## What this PR is *not* - No behavioural changes to any skill. - No body content rewrites. - No changes to `skill-validator` rules themselves — only the skills they flag. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
