メニュー
Analyze observational skill-effectiveness signals across scanned sessions. Use for exploratory monitoring and recommendation triage, not as a release gate.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
CONTRIBUTOR TOOL - Track CC changelog, extract new versions since last check, analyze impact on plugin (breaking changes, opportunities, deprecations). Run periodically or before releases. NOT part of the distributed plugin.
CONTRIBUTOR TOOL - Check the plugin against current cached Claude Code docs. Use before releases or after Claude docs changes to separate real schema drift from stale local assumptions. Not part of the distributed plugin.
Guide plugin development workflow for this repo. Use when editing shipped plugin files under plugins/ruby-grape-rails/, release/docs metadata, or contributor tooling under .claude/.
Initializing the Ruby/Rails/Grape plugin: writes stack notes (queues, ORM-per-package, layout) into CLAUDE.md. Triggers: "initialize plugin", "setup ruby plugin", "configure Claude for Rails".
Walking a user through the Ruby/Rails/Grape plugin commands, capabilities, and workflow. Tutorial-style intro for newcomers who want to learn what the plugin offers rather than tackle a specific task.
Running a project-wide audit of a Ruby/Rails/Grape codebase: architecture, security, performance, testing coverage, ops risk. Emits findings to .claude/audit/.
| name | skill-monitor |
| description | Analyze observational skill-effectiveness signals across scanned sessions. Use for exploratory monitoring and recommendation triage, not as a release gate. |
| argument-hint | [--skill NAME] [--improve] [--window 7d|30d|all] [--provider NAME] |
| disable-model-invocation | true |
Imperative-only. Exploratory — does NOT prove a skill is good or bad on its own.
.claude/session-metrics/metrics.jsonl (from /session-scan).
Before trusting the dashboard:
claude plugin validate --strict plugins/ruby-grape-railsmake eval or make eval-all/docs-check if Claude docs drift suspected/skill-monitor — corroborating observational input| Command | Behavior |
|---|---|
/skill-monitor | dashboard for all skills |
/skill-monitor --skill review | scoped to one skill |
/skill-monitor --improve | delegates to skill-effectiveness-analyzer |
/skill-monitor --window 30d | broader window |
/skill-monitor --provider claude-code | exact provider |
| Flag | Values |
|---|---|
--skill | NAME |
--improve | (no value, triggers analyzer mode) |
--window | 7d / 30d / all |
--provider | NAME |
Read .claude/session-metrics/metrics.jsonl. Extract
skill_effectiveness.
No skill_effectiveness data → tell contributor to rescan with current
scorer.
--provider present → restrict dashboard to that provider.
Useful aggregates:
Baseline comparison computed → present as heuristic context, NOT causal effect estimate.
Show:
Flagged skills are review candidates, NOT proven regressions.
--improve requested → delegate to skill-effectiveness-analyzer.
Analyzer requirements:
lab/eval, docs-check, previous session-analysis reportsWrite dashboard snapshots under .claude/skill-metrics/. Do NOT modify
historical snapshots.
metrics.jsonl from this skill.| Need | Reference |
|---|---|
| metric definitions + behavioral-proxy rubric | ${CLAUDE_SKILL_DIR}/references/effectiveness-metrics.md |
| improvement proposal template | ${CLAUDE_SKILL_DIR}/references/improvement-template.md |
Explicit confidence on observational findings. Direct language for HIGH-confidence (deterministic signals, large-N agreement across providers). LOW-confidence MUST be labeled — do NOT promote noise into confident findings via polite phrasing. Hedge qualifiers belong on genuinely-uncertain claims, NOT scattered across the report. State contradictions with contributor expectations directly.