Trackman session analyzer

Trackman Session Analyzer

Ingests the user's recent Trackman sessions, stores a deterministic analysis of each in the MCP's local store (capped at the last 30, newest first), and returns a summary of the most recent session with its stats normalized against prior sessions.

<HARD-RULE: RUN IN A FORK, NEVER ON THE MAIN THREAD> This is a data-collection skill. It makes many MCP calls and pulls large shot-level payloads that would bloat the main conversation. The main agent must NOT execute the steps below inline.

Instead, the main agent's ONLY job is to dispatch a single subagent (the Agent/Task tool, general-purpose type) whose prompt is "Follow the trackman-session-analyzer skill end to end and return only the final summary." The subagent does all the work in its own context and returns just the summary section. If you are the dispatched subagent, proceed with the workflow. </HARD-RULE>

Inputs / tools used (all from the golf-coach MCP)

auth, trackman(action="sessions"), session_analysis(action="analyze"), session_analysis(action="list"), session_analysis(action="get"). The heavy lifting (classify, metrics, course difficulty, normalization, used-vs-available clubs) is done deterministically inside session_analysis(action="analyze") — your job is orchestration and narration, not recomputing numbers.

Workflow (subagent)

1. Auth check. Call auth. If not authenticated, stop and report that the user needs to run golf-coach login — do not fabricate data.

2. Pull recent sessions. Call trackman(action="sessions", take=30) (newest first). This includes both practice activities and course rounds.

3. Find what's new. Call session_analysis(action="list") to get already-stored ids.

4. Analyze + store new sessions, OLDEST first. Reverse the fetched list and walk it oldest → newest. For each session whose id is not already stored, call session_analysis(action="analyze", activity_id). Oldest-first matters: each session is normalized against the sessions chronologically before it, so the history must already be stored when the newest session is analyzed. The store keeps only the last 30. Skip nothing silently — if a call errors, note it and continue.

5. Read back the latest. Call session_analysis(action="list") again, take latest_id, then session_analysis(action="get", activity_id=latest_id) for the full record.

6. Summarize the latest session using that record (see format below). Use the record's own fields — category, seriousness, is_improvement_attempt, reasons, metrics, normalized, clubs_used, clubs_unused, summary. For "normalized vs previous," report the normalized deltas/z-scores.

What the classification means (already computed for you)

• game — a played round (COURSE_PLAY / ON_COURSE / virtual). Reported with score-to-par, GIR, putts, driving, and a 0–1 course difficulty score.
• practice — a genuine attempt to improve: enough volume/duration/club variety, or an inherently focused kind (shot analysis, find-my-distance, performance center, map-my-bag, putting, combine). is_improvement_attempt is true.
• warmup — too short/few strokes to count as improvement (e.g. 8 balls over 5 minutes with one club). is_improvement_attempt is false. Do not credit these as training.

Output format (this is the skill's return value)

Return ONLY this summary — not the raw session payloads.

## Latest Trackman session — <date> (<category>)

<one-line headline from record.analysis.summary>

**Type:** <game | serious practice | warm-up>  ·  seriousness <0–1, if practice>
**Why classified this way:** <record.analysis.reasons>

**Key stats** (metric units — m/s, meters):
<for a game: to_par, GIR, putts, drive avg, course difficulty>
<for practice: stroke count, duration, clubs used, avg carry, notable per-club>
<if metrics.has_club_data: per-club **club delivery** from per_club[club].delivery —
  club path (+ club_path_std), face angle, face-to-path, attack angle, spin axis.
  Bay/sim sessions (shot analysis, sim, map-my-bag) carry this; range-radar
  sessions do not, so omit the line entirely when has_club_data is false.>

**Normalized vs your prior sessions** (n = <history size>):
<for each normalized metric: value, mean, delta, and whether better/worse>

**Clubs:** used <clubs_used>  ·  available-but-unused <clubs_unused>

**Recent habit check:** of the last <N> stored sessions, <X> were serious
practice, <Y> warm-ups, <Z> games.

Keep it tight and factual. This skill does not give coaching drills — hand the summary to the golf-coaching skill for that.