What needs to change
Two related authoring gaps surfaced across the skill set. Both are judgment-heavy and per-skill, not a mechanical sweep; this issue tracks the class.
- Demonstrate, don't describe. Several skills are entirely abstract where a single worked example of the hard case would retire multiple rules. Candidates: a filled plan skeleton (
aiw-planning), a semantically-wrong-but-green output (aiw-verification), a drifted context line vs its correction (aiw-project-context-management), a baseline-anchored benchmark (aiw-performance-profiling), a filled contradiction block (aiw-failure-analysis), a concrete modality-conflict pair (aiw-ground-truth).
- Leading words. Each of these skills turns on a governing concept that is never coined and reused at the step that depends on it. Candidates: "coverage" (
aiw-verification), "trust collapse" reused at the audit (aiw-failure-analysis), "drift" (aiw-project-context-management), "baseline" (aiw-performance-profiling).
Each skill is its own decision.
Why / what triggered it
Surfaced by running aiw-prompt-smith across the skill set. An adjective leaves room the model fills with its average; one good example, and a named concept reused at the key step, steer better than more prose.
Affected files (each mirrored in .agents/skills/)
.claude/skills/aiw-planning/SKILL.md
.claude/skills/aiw-verification/SKILL.md
.claude/skills/aiw-failure-analysis/SKILL.md
.claude/skills/aiw-ground-truth/SKILL.md
.claude/skills/aiw-performance-profiling/SKILL.md
.claude/skills/aiw-project-context-management/SKILL.md
Acceptance criteria
- Where an example is added, it demonstrates the hard case and lets abstract rules shrink; net length should not grow materially (hold to the elegance check).
- Where a leading word is introduced, it is reused at the key step, not on every line.
- Changes stay behaviour-preserving; both skill directories identical; version bump with CHANGELOG entry; lite parity preserved.
Skills that may help
What needs to change
Two related authoring gaps surfaced across the skill set. Both are judgment-heavy and per-skill, not a mechanical sweep; this issue tracks the class.
aiw-planning), a semantically-wrong-but-green output (aiw-verification), a drifted context line vs its correction (aiw-project-context-management), a baseline-anchored benchmark (aiw-performance-profiling), a filled contradiction block (aiw-failure-analysis), a concrete modality-conflict pair (aiw-ground-truth).aiw-verification), "trust collapse" reused at the audit (aiw-failure-analysis), "drift" (aiw-project-context-management), "baseline" (aiw-performance-profiling).Each skill is its own decision.
Why / what triggered it
Surfaced by running
aiw-prompt-smithacross the skill set. An adjective leaves room the model fills with its average; one good example, and a named concept reused at the key step, steer better than more prose.Affected files (each mirrored in .agents/skills/)
.claude/skills/aiw-planning/SKILL.md.claude/skills/aiw-verification/SKILL.md.claude/skills/aiw-failure-analysis/SKILL.md.claude/skills/aiw-ground-truth/SKILL.md.claude/skills/aiw-performance-profiling/SKILL.md.claude/skills/aiw-project-context-management/SKILL.mdAcceptance criteria
Skills that may help
aiw-prompt-smith