Skip to content

Releases: muippt/mu-skill-creator

Release list

v3.7.0 — AP-34 Gotchas-bloat + L1-8 check

v3.7.0 — AP-34 Gotchas-bloat + L1-8 check Latest
Latest

Choose a tag to compare

View all tags
@muippt muippt released this 27 Jul 07:05
v3.7.0
bdefb18

v3.7.0

New: AP-34 — Gotchas / troubleshooting bloat

A new anti-pattern for the case where a "Gotchas / 踩坑" section quietly grows into a per-incident log — each entry a full symptom → cause → fix narrative — until it dominates the main skill file. It is the structural sibling of AP-33 (version-history bloat): both push volatile, append-only detail into the file that should stay lean.

New: L1-8 automated check

The audit script now measures the Gotchas section: it counts ### entries and the section's share of total lines, and emits an info-level warning when there are more than 10 entries or the section exceeds 40% of the file. Like the L1-5 line-count check, this is advisory — never a hard fail.

Tiered remedy guidance

Each flagged Gotchas item is triaged three ways:

  • Delete — already covered by the workflow or a rule; redundant.
  • Uplift — an actionable conclusion that belongs as a step in the main workflow.
  • Sink — an implementation-specific pitfall that moves to references/troubleshooting.md, leaving only a short symptom → lookup index in the main file.

Rule of thumb: needed during normal execution → uplift; only needed when touching one specific thing → sink; stated elsewhere → delete.

Counts

  • Audit model: 52 → 53 items (27 auto-scannable)
  • Anti-pattern catalog: 33 → 34

Synced across SKILL.md, the audit script, quality-gates reference, both READMEs, and the landing page.

Assets 2
Loading

v3.6

v3.6

Choose a tag to compare

View all tags
@muippt muippt released this 20 Jul 03:33
v3.6
922a370

What's New

  • 52-item, 10-layer audit model for AI agent skill quality assurance
  • 3-layer architecture (L1/L2/L3) with context-aware loading
  • 33 anti-pattern catalog (AP-1 to AP-33), each traced to a real incident
  • 8-stage gated workflow with explicit entry/exit criteria
  • Automated audit script (skill-audit.sh) for batch scanning
  • Stall detection rules for iterative and scheduled skills
  • Security-first design with pre-publish credential and PII scanning

Quick Start

git clone https://github.com/muippt/mu-skill-creator.git
cp -r mu-skill-creator /path/to/your/agent/skills/
bash scripts/skill-audit.sh --verbose

Links

Loading