メニュー
Applies 1/2/3-strike adjustment rules to plan.json based on a single workout's analysis dict. Mutates plan only when thresholds fire. Called after every workout log.
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
Renders today's workout card from plan.json and daily_state.json as a Markdown block. Use for every /run-today invocation. No writes — pure read-and-render.
Conversationally captures a completed workout (duration, distance, RPE, HR, splits, notes) and writes it to workouts.json. Triggers analyze-workout and adapt-plan downstream. Use for /run-log.
Inserts a post-race or injury-flag recovery block into plan.json: 4 days rest + 3 easy jog days (post-race), or 3 rest + 1 recovery (red-flag-fatigue), or 7 rest days (injury). Use after /run-race-recap or when adapt-plan returns red_flag_recovery.
Web-search-backed Q&A for novel coaching questions outside the plan's standard logic: heat acclimation, recovery from illness, pacing strategies for unique courses. Use for /run-research and when the Coach cannot answer from training data alone.
Aggregates the past 7 days of completed workouts: total mileage, quality session adherence, fatigue indicators (RPE creep, HR drift). Decides hold/advance/recovery for the next week. Use for /run-week.
Pulls recent run activities from the garmin-givemydata MCP server: triggers garmin_sync to refresh the local SQLite, queries activity IDs in the requested date range, fetches per-activity detail (with splits, HR zones, weather, running dynamics), normalizes to workouts.json schema via garmin_normalize.py, deduplicates by garmin_activity_id with preferred_source tiebreaker, and runs the same analyze-workout + adapt-plan pipeline as manual logging. Requires users.integrations.garmin.connected = true.
| name | adapt-plan |
| description | Applies 1/2/3-strike adjustment rules to plan.json based on a single workout's analysis dict. Mutates plan only when thresholds fire. Called after every workout log. |
Reads the analysis verdict from analyze-workout, updates the consecutive miss/ahead counters in daily_state.json, and applies the appropriate adaptation rule to plan.json when a threshold fires. The strike-rule logic is fully deterministic and implemented in scripts/adapt.py. Claude's role is to invoke the script, interpret the returned action, and — for vdot_bump — trigger compute-vdot to regenerate paces before re-resolving future plan entries.
analyze-workout returns a verdict, called by log-workout or fetch-strava-activity.adapt-plan returns red_flag_recovery and a recovery block must be inserted.Updated storage/plan.json (if action mutates the plan) and updated storage/daily_state.json (always). Returns the action string to the caller for display.
See body.md for full instructions and scripts/adapt.py for the rule engine.