fix: improve swamp-model description with explicit extension-model boundary

[stack72]

Summary

• Improve swamp-model skill description to better trigger on method-related
  queries (validation, transform, enrichment) while staying scoped to swamp
  domain terms. Added concrete definition of what swamp models are: "structured
  automation units that define typed schemas, methods, and outputs for data
  processing".
• Add explicit boundary with swamp-extension-model: "Do NOT use when the user
  wants to build, create, or implement a custom model type, Zod schema, or
  TypeScript model — that is swamp-extension-model". This prevents false
  triggers on queries like "Build a Zod schema for my custom validation model".
• Fix "Delete this model from the repository" eval expectation from false to
  true — model delete is explicitly in the skill's trigger keywords.
• Trigger eval: 75% → 81% (passing). Remaining miss ("Run the validation
  method on this JSON data") is a zero-context edge case that passes
  intermittently with EVAL_RUNS=3.
• tessl review: 91% average (description 82%, content 100%) — passes.

Test plan

• EVAL_RUNS=1 deno run eval-skill-triggers --skill swamp-model — 81%, passes
• npx tessl skill review .claude/skills/swamp-model — 91% average
• No false triggers on swamp-extension-model boundary queries
• Negatives all pass except probabilistic variance on EVAL_RUNS=1

🤖 Generated with Claude Code

[github-actions]

Adversarial Review

Critical / High

None.

Medium

None.

Low

1. .claude/skills/swamp-model/evals/trigger_evals.json: The old note for the "Delete this model" entry explained nuance ("borderline; context determines if this is about removing from system vs managing the definition"). The new note ("model delete is a listed trigger keyword") is simpler but loses that nuance. If the eval framework ever uses notes for disambiguation, the reduced context could matter. Practically irrelevant since notes are informational only.

Verdict

PASS — Metadata-only changes to a skill description and eval expectations. The eval fix correctly aligns "Delete this model from the repository" with the skill's explicit "model delete" trigger keyword. No executable code is modified; nothing to break.

[github-actions]

Review: LGTM

Clean, well-scoped skill definition update. No blocking issues.

What changed:

• Improved swamp-model skill description with concrete definition ("structured automation units") and expanded trigger keywords (validation method, transform method, enrichment model, etc.)
• Added explicit boundary with swamp-extension-model to prevent false triggers
• Fixed "Delete this model from the repository" eval: should_trigger corrected from false to true — model delete is clearly listed in the skill's trigger keywords and Quick Reference table

Verified:

• JSON eval file is valid
• YAML frontmatter is well-formed
• Eval expectation change is logically correct — delete is a lifecycle operation covered by this skill
• No TypeScript code changes, so no type/lint/test/import concerns apply

No suggestions — this is a straightforward improvement.