delete_avatar_check_in
ChatGPTDelete one saved check-in only on an explicit user request. Permanently removes its observations and recomputes later avatars so the deleted evidence is no longer retained. Use the exact checkInId and expectedRevision from the current context. Safe to retry the same ID. Does not delete workouts, body-weight logs, profile or other check-ins.
discard_workout
ChatGPTPermanently discard the shared active workout without saving it. Only call when the user explicitly asks to discard or cancel this workout. Completed workouts are unaffected.
finish_workout
ChatGPTFinish and save this workout to Cloudflare. Unlogged sets become skipped (0 reps). Only call when the user asks to finish/save. Safe to retry with the same workout ID after a network failure. After success, confirm the save and optionally one improvement supported by history in 1–2 short sentences; no lengthy recap unless asked.
get_avatar
ChatGPTReturn the illustrative front-and-side image and saved observations for a check-in. Omit checkInId for the latest saved check-in; use IDs from get_avatar_check_ins for history. Image sizes are fitted consistently; profile height is metadata, not a visual comparison of absolute height. This is not a body-composition assessment or future projection.
get_avatar_check_ins
ChatGPTRead saved photo-observation check-ins and derived avatar settings, newest photo date first. Includes uncertainty and decisions about retained or neutral shape controls. Paginate with nextOffset. Use get_avatar with a returned check-in ID to show its image. No photos are stored or returned.
get_avatar_context
ChatGPTRead the profile height, latest saved avatar, avatar revision and photo-observation rubric before interpreting a photo for a fitness-avatar check-in. Confidence describes evidence quality. Height belongs to Profile; do not request body weight or photo uploads through this tool. Does not save observations or change a workout.
get_button_coaching_status
ChatGPTRead active button coaching subscriptions and recent webhook timing for this OAuth connection. Does not read or change workouts. Accepted means ChatGPT received the event, not that a reply appeared. This status covers the OAuth connection and does not identify which conversation owns each subscription; match the host task to this conversation before treating monitoring as enabled. Callback URLs and signing secrets are never returned.
get_rest_timer
ChatGPTRead the remaining rest time for the shared active workout without starting or changing it. Use when asked how much rest is left. Returns a snapshot of the saved deadline for a text reply; it does not display a live countdown card or send notifications. The website shows the live countdown.
get_workout
ChatGPTRead the current 5×5 workout draft, exercises, warmup plans and completion, working weights, logged work sets and rest deadline for a conversational workout. Call first to get the current revision and workout ID. The browser and ChatGPT share one active workout, including warmups and rest. When opening/resuming a coaching session, also read get_workout_history (limit 6) once for context. Reply in 1–2 short conversational sentences, with at most one history observation. When the user is opening/resuming an active workout for coaching, follow the shared automatic button-coaching setup instructions once; a read-only state inspection alone must not subscribe. For saved sessions or the last workout date, use get_workout_history.
get_workout_history
ChatGPTRead saved workouts from the connected Cloudflare training log, newest first, including website workouts and imported history. Use this to answer when the user last trained or what they lifted, and once at the start of a coaching session (limit 6). For a coaching observation, reply in 1–2 conversational sentences, at most 35 words: choose one exercise or one pattern, not a list of every lift and weight. Only expand if the user explicitly asks for detail. Compare matching exercises and units; unknown plans are not failures. Do not claim an all-time record from this limited page. For the latest workout use limit 1. Dates are recorded local calendar dates. Excludes unfinished drafts. Does not change saved workouts.
log_set
ChatGPTLog a reported working set for an explicit exercise and 1-based set number. Assume five reps when the user reports the set done without specifying reps; omit reps to use that default. Preserve any explicitly reported rep count, including zero. Use null only to clear a set. Never infer that the user completed a set. Starts the saved rest interval and returns its deadline; clearing the latest logged set cancels its rest. Reopen on a revision conflict. On success, reply with a short acknowledgement and optionally one history-based comparison or cue (1–2 sentences, at most 35 words). Use already-read history, not an extra history call per set. Never invent improvement or change weights as part of encouragement.
log_warmup_set
ChatGPTMark an explicit exercise’s 1-based warmup set complete (done: true), or undo it (done: false). Read the warmup plan from get_workout or start_workout first. When the user reports the warmup done without a rep count, assume the prescribed five reps and mark it complete. If they report a different count, do not mark it complete; partial reps cannot be recorded here. Never use log_set for warmups. Completing all warmups for the exercise starts the saved rest interval; intermediate warmups do not. Undo cancels that exercise’s warmup rest only. Warmups never count as working sets or saved history. Reopen on a revision conflict. Acknowledge briefly.
preview_avatar_check_in
ChatGPTTurn visible observations from user-supplied photos into an illustrative front-and-side avatar image without saving. Follow get_avatar_context first. Record confidence per observation; use unknown/low or omit anything obscured. Side-depth observations need a side photo. Do not infer body-fat percentage, muscle mass, sex, race, age, health or a future body. Send only the structured observations and capture conditions: no photos, URLs, height, weight or manual avatar percentages. Low-confidence or unsuitable evidence retains the previous shape. This is a visual experiment using one adult base, not a validated body model.
save_avatar_check_in
ChatGPTSave a photo check-in only when the user asks to record it. Stores visible observations, per-observation confidence, capture conditions and server-derived avatar settings; never stores the photo. Use a fresh checkInId UUID and expectedRevision from get_avatar_context. Retry the identical arguments with the same ID after a network failure; changed payloads or stale revisions conflict. Inserting an older check-in recomputes subsequent derived avatars in date order. Does not log weight, edit Profile or change workouts.
set_weight
ChatGPTChange weight, in the workout’s displayed kg/lb unit, for this exercise’s unlogged sets only. Includes the bar. Only change it on a clear user request or acceptance of a suggestion. Confirm the change briefly.
start_workout
ChatGPTStart a workout using the same progression rules and history as the website, or reopen an active draft. Use the user’s local date. Omit routine to alternate A/B automatically. Review recent history for coaching, follow the returned planned weights, and introduce only the next step in 1–2 short sentences. After successfully starting/reopening the workout for coaching, follow the shared automatic button-coaching setup instructions to reuse or create this conversation's workout.set_logged event task. Do not dump the full workout unless asked.
