gtapps/capture-activity-rpe

Capture Activity RPE

Records the operator's perceived effort and subjective notes for the most recent synced activity.

Self-triggers on RPE-shaped replies while state/strava-pending-rpe.json is fresh. The skill's description is intentionally narrower than claude-code-hermit:channel-responder's so the harness picks it for these specific messages; any non-matching message exits silently and the normal channel-responder flow proceeds. Re-checks allowed_users itself.

Steps

1. Read .claude-code-hermit/state/strava-pending-rpe.json. If the file is absent: exit silently. If synced_at is more than 24 hours ago: run rm .claude-code-hermit/state/strava-pending-rpe.json and exit silently (reply window closed).

2. Authorization. Read .claude-code-hermit/config.json → channels.<channel>.allowed_users for the inbound channel (extract the channel name from the inbound message metadata):

   • If allowed_users exists and the sender's platform user ID is not in it: exit silently. No response, no log, no state write.
   • If allowed_users is absent: accept (backwards-compatible).
   • If allowed_users is []: reject all — exit silently.

3. Parse the inbound message body (case-insensitive, leading/trailing whitespace stripped):

   ^(?:RPE\s*:?\s*)?(\d{1,2})(?:\s*/\s*10)?(?:[\s,:]+(.+))?$
   

   This matches: 7, 7/10, RPE 7, RPE: 7, RPE:7, 7 heavy legs, RPE 7, heavy legs, RPE: 7 heavy legs.

   • Extract rpe (group 1) and notes (group 2). If group 2 is absent or an empty string after trim, treat notes as null.
   • Validate rpe is an integer from 1 to 10 inclusive.
   • If parse fails or validation fails: exit silently. (Bare yes/no/y/n cannot match \d{1,2} so they are naturally excluded; the micro-approval branch in channel-responder handles them.)

4. Persist the entry via the script (atomic upsert into state/activity-notes.json, keyed by <pending.activity_id>, overwriting silently):

   bun ${CLAUDE_PLUGIN_ROOT}/scripts/fitness-lab.ts rpe <pending.activity_id> <rpe> [notes...]
   

   Pass the parsed rpe; append the notes words only when group 2 matched (omit entirely when notes are null). The script writes {rpe, notes, recorded_at} — notes is always present, null when none were captured — and stamps recorded_at itself. No Strava call is made (the activity ID is supplied, not latest), so this never touches the network.

5. Run rm .claude-code-hermit/state/strava-pending-rpe.json so the same sync cannot bind twice.

6. Reply via the channel's reply tool with {chat_id, text}. allowed-tools only lists Discord (mcp__plugin_discord_discord__reply); on non-Discord channels this step will fail with a permission error, but the RPE is already persisted in step 4. Extend allowed-tools when adding Telegram or iMessage.

   Got it — RPE <rpe>/10 saved for <pending.name>.
   

7. Refresh the deep-dive. If pending.sport is Run, re-invoke /claude-code-fitness-hermit:activity-deep-dive <pending.activity_id>. The fitness-brief evening run generated this activity's deep-dive before any RPE existed, so re-running now folds the RPE in — the skill reads state/activity-notes.json at its step 3b and overwrites compiled/activity-<id>-*.md. For non-Run sports the brief writes no deep-dive, so skip.