Three-layer evaluation engine (static + LLM judge + Monte Carlo)
with Elo ranking for Claude Code plugins. Includes CLI, Claude Code
plugin, and 69 tests. Calibrated against full 147-skill corpus.
Also improves 25 skills: trigger phrases, reference extraction.
Each skill's SKILL.md is trimmed to under 650 lines. Advanced content
moved to references/ directories with pointers back from SKILL.md:
- nodejs-backend-patterns: DI container, DB patterns, auth, caching,
API response format → references/advanced-patterns.md
- javascript-testing-patterns: integration tests, frontend/hook tests,
fixtures, snapshots, coverage, common patterns → references/advanced-testing-patterns.md
- modern-javascript-patterns: functional programming, classes, modules,
iterators/generators, operators, performance → references/advanced-patterns.md
- openapi-spec-generation: code-first (FastAPI, tsoa), validation/linting,
SDK generation → references/code-first-and-tooling.md
- tailwind-design-system: animations, dark mode, advanced v4 patterns,
migration checklist, best practices → references/advanced-patterns.md
- godot-gdscript-patterns: scene management, save system, performance
tips, best practices → references/advanced-patterns.md
- python-testing-patterns: trim SKILL.md from 1028 to 622 lines; move Patterns 6-10 (async, monkeypatch, tmp_path, conftest, property-based) plus database testing, CI/CD, and config to references/advanced-patterns.md
- python-packaging: trim SKILL.md from 866 to 510 lines; move advanced packaging patterns (namespace, C extensions, version management, testing, docs, distribution) to references/advanced-patterns.md
- python-performance-optimization: trim SKILL.md from 852 to 437 lines; move NumPy, caching, slots, multiprocessing, async I/O, DB optimization, and memory patterns to references/advanced-patterns.md
- uv-package-manager: trim SKILL.md from 814 to 345 lines; move Docker, lockfile workflows, performance, tool comparisons, common workflows, troubleshooting, and migration guides to references/advanced-patterns.md
The regex only matched "use when" but not "use this skill when" —
a common pattern in the codebase. Updated to match all trigger
variants: "use when", "use this skill when", "use proactively",
"trigger when".
Eliminates 17 false-positive MISSING_TRIGGER flags.
Layer 2 now runs via eval-judge agent using Max plan session auth
instead of Agent SDK subprocess calls. No API key required.
Also fixes composite scoring to exclude unmeasured dimensions.
Adds MonteCarloAnalyzer with SimResult/MonteCarloConfig dataclasses, run_simulation
async helper, and _compute_statistics using Wilson score CI, bootstrap CI,
Clopper-Pearson CI, and coefficient of variation. Wires MC layer into evaluate_skill
for Depth.DEEP and Depth.THOROUGH runs (50 and 100 runs respectively).
Add plugin.json, three commands (eval/compare/certify), eval-orchestrator agent,
and evaluation-methodology skill to wire plugin-eval into the claude-agents ecosystem.
Adds EvalEngine that coordinates static analysis, blends layer scores
across dimensions with renormalized weights, and produces PluginEvalResult
with composite score, badge, and per-dimension grades. Layer 2/3 stubs
ready for Tasks 8/9.
Match the model choice with the identical agent in cloud-infrastructure,
deployment-strategies, and full-stack-orchestration plugins.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Adds awareness of Oracle Cloud Infrastructure to any plugin that referenced
at least two of the major cloud vendors already. Skills updated to include
OCI services. Also updated some of the other cloud references.
Signed-off-by: Avi Miller <me@dje.li>
- Remove .mcp.json (not used in marketplace, add README instead)
- Add marketplace.json entry for plugin discovery
- Add README.md with MCP server setup guide, provider config, and troubleshooting
- Add tools: declaration to image-generator agent (functional fix)
- Move <example> blocks from YAML frontmatter to markdown body
- Remove unused tools: Read, Grep, Glob from prompt-crafter agent
- Remove redundant name: field from command frontmatter
- Use full MCP tool prefix (mcp__meigen__*) in commands
- Rewrite plugin.json description to factual style
- Pin npm version to meigen@1.2.5
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Remove references to non-existent resource files (references/, assets/,
scripts/, examples/) from 115 skill SKILL.md files. These sections
pointed to directories and files that were never created, causing
confusion when users install skills.
Also fix broken Code of Conduct links in issue templates to use
absolute GitHub URLs instead of relative paths that 404.
Improve frontmatter description with action-oriented trigger terms for
better skill matching. Add copy-paste artifact templates as a reference
file. Inspired by @fernandezbaptiste contribution in #437.
thc1006