{"id":103162,"plugin_id":"plugin_asdk_app_6ae9fa1a19648191947d22e4efcf608c","kind":"skill","collection_source":"plugin_package","comparison_source":null,"observed_at":"2026-11-18T06:18:41.600Z","digest":"b67cdf2c1a1ea4dfab38b9177b31137909c8ed5fded13c92eb57248b2b8a739e","against":null,"payload":{"description":"Track and recall workouts, correct or remove logged sets, and save or manage training locations with GoodSet (including spoken aliases \"Good Set\" and \"Good Sets\"). Use when the user explicitly chooses GoodSet; asks to start, record, log, track, continue, or finish a workout, lift, or run; reports a completed set in an established workout context; asks about workout history, personal records, or exercise progress; or asks to save or manage a training location for workouts. Do not activate from tentative future-set statements alone, bare travel, arrival, or readiness phrases, or when the user asks for another app.","included_files":[{"relative_path":"agents/openai.yaml","size_in_bytes":434}],"name":"track-workout","skill_md_contents":"---\nname: track-workout\ndescription: Track and recall workouts, correct or remove logged sets, and save or manage training locations with GoodSet (including spoken aliases \"Good Set\" and \"Good Sets\"). Use when the user explicitly chooses GoodSet; asks to start, record, log, track, continue, or finish a workout, lift, or run; reports a completed set in an established workout context; asks about workout history, personal records, or exercise progress; or asks to save or manage a training location for workouts. Do not activate from tentative future-set statements alone, bare travel, arrival, or readiness phrases, or when the user asks for another app.\n---\n\n# Track workouts with GoodSet\n\nUse GoodSet as the system of record when this skill is selected for workout tracking. Do\nnot offer to keep a conversation-only log. Honor an explicit request to use another app.\nDo not use this skill for hypothetical future workout plans, gym logistics, directions,\ngeneric place questions, or general fitness advice that should not change workout or\nsaved-location data.\n\nA tentative future-set statement alone does not request GoodSet. Do not call any GoodSet\ntool or read workout status just to acknowledge a plan. Reply\nconversationally without treating the set as completed. An explicit request to inspect the\ncurrent workout, history, or progress still calls for that read, including when the user\nis deciding their next set.\n\n## Route the request\n\n- For a live workout start, resume, completed set, or finish, use `workout_track`.\n  `action=open` is the supported resume and recovery operation: it starts a workout or\n  returns the one in progress. Use `action=reopen` only when the user says right after a\n  finish that they were not done (\"I wasn't done,\" \"keep going\"). For a read-only request\n  to see or check the current workout, use `workout_current`.\n- For a correction to a logged set's measurements or note, use `workout_track` with\n  `action=correct_set`. For a logged exercise occurrence assigned to the wrong exercise,\n  use `action=correct_exercise`.\n- When the user asks to delete, remove, or undo a logged set, use `workout_track` with\n  `action=remove_set`. Without `set_number` it removes the latest set; for another set,\n  pass the exercise and its `set_number`.\n- To change a finished workout, follow \"Change a past workout\" below.\n- For the user's last or previous workouts, training days, exercise history, loads,\n  sets, or reps from a specific workout, use `find_workouts`. Use `workout_get` only\n  when a returned public workout reference needs complete detail.\n- For an exercise personal record or progress trend, call `search_exercises` to resolve\n  the exact exercise, then call `exercise_progress` with its public reference. Never guess\n  between distinct matches. When `sessions_truncated` is true, describe the sessions as a\n  recent window and do not use the oldest returned session as a lifetime baseline.\n- Use `training_location_find` only when an explicit workout-tracking request supplies\n  place context. Use `training_location_manage` only when the user explicitly asks to\n  manage saved training locations. A training location can be a gym, home setup, outdoor\n  area, or another place; do not reduce it to a gym.\n- Use `workout_record_completed` only for an already-completed workout that is not the\n  active live workout and is not already in GoodSet.\n- For saved routines (a named plan of exercises and set targets), read them with\n  `workout_routine_find` and change them with `workout_routine_manage`.\n- Use `search_exercises` for personal-record and progress resolution, genuine discovery,\n  or unresolved ambiguity. `workout_track` resolves ordinary exercise names itself.\n\nIf the GoodSet connection or required tools are unavailable, ask the user to connect\nGoodSet. For a write, say that the requested change was not recorded; for a read, say that\nthe workout history could not be retrieved. Never simulate a result.\n\nCorrections use the same durable continuation and confirmation flow as other workout\nchanges. An exercise correction changes the assignment of the selected occurrence and all\nsets already logged under it; it does not rename the exercise catalog entry. Follow any\nreturned exercise or occurrence disambiguation, then confirm the preview. Never add\nreplacement sets to simulate a correction.\n\n`correct_set` requires a real `set_number`. When the user says the last or latest set,\nfirst call `workout_current` and copy the `set_number` of the last entry in\n`workout.current_exercise.sets` from the result. Never guess a set number from the\nconversation or assume it is one.\n\nSet removal also uses the confirmation flow. In the workout in progress, an explicit\nrequest to delete, remove, or undo the latest set authorizes confirmation only when the\npreview identifies the set the user meant. Do not turn removal into a correction with empty\nvalues. Report success only after the confirmed result says the removal was applied;\n`no_change` means nothing was removed.\n\n## Keep saved routines as plans\n\n- A routine is a plan, never a record. Saving or editing one logs no sets; planned targets\n  are not completed work.\n- Save, rename, replace, or archive a routine with `workout_routine_manage` only when the\n  user explicitly asks. Those actions apply immediately: report the returned routine, do\n  not ask for confirmation again. To save the workout just finished, use\n  `action=save_workout` with its `workout_ref`. A target's `sets` counts identical\n  consecutive sets, so 3 sets of 80 kg x 6-8 is one entry with `sets` 3.\n- To train from a routine, resolve it with `workout_routine_find` (by name, or list them),\n  then call `workout_track` with `action=open` and its `routine_ref`. The result's\n  `next_planned_set` is the plan's next set: you may name it, but log only the sets the\n  user reports, with the values they report.\n- `name_conflict` means nothing was saved. Ask whether to use another name or to change\n  the existing routine; never overwrite it unasked.\n- Resolve each exercise with `search_exercises` before `create` or `replace_exercises`.\n  `replace_exercises` restates the whole plan; keep the exercises the user did not mention.\n- GoodSet shows previous sets and records; it does not suggest next weights. Do not invent\n  progression targets for a routine.\n\n## Keep training locations optional\n\n- Never ask for a training location just to start or record a workout. If the user does\n  not mention one, omit it.\n- When a workout request clearly names a place, call `training_location_find` first. Use\n  the sole clear saved-name match's `training_location_ref`. The tool lists saved GoodSet\n  locations only: call it without passing the place name, then compare the returned names.\n  It has no free-text place query and returns no real-world place candidates. If there is\n  no clear saved match, omit the location and continue the workout immediately; save the\n  place only when the user explicitly asks to manage it.\n- When the user explicitly asks to save a new location, create it with the user-supplied\n  name, then pass its returned reference to the workout. Omit `location_kind` unless the\n  user supplied the category; a name such as Home Rack does not authorize inferring `home`.\n- Never ask for or pass an address, city, coordinate, or provider identifier through these\n  tools. Trusted clients manage that metadata; never infer, geocode, or invent it.\n- Use a workout's returned `training_location` as ordinary context. To compare loads on\n  the same equipment or biomechanics, call `find_workouts` with `training_location_ref`.\n  Do not expose that reference in the conversational response.\n- Archived locations remain historical context. Do not attach one to a new workout unless\n  the user restores it first.\n\n## Handle a live workout\n\n1. Do not select this skill, mention GoodSet, or offer workout tracking from a bare travel,\n   arrival, or readiness phrase such as \"on my way to the gym,\" \"I just got here,\" or\n   \"ready to train.\" Those phrases alone do not express workout tracking, history, progress,\n   or saved-location-management intent.\n2. Call `workout_track` immediately when the user explicitly asks to start, resume, or\n   record a workout, or unambiguously says that a workout is already underway. Use the\n   operation matching the user's intent; do not ask for exercise details before opening or\n   recovering the active workout.\n3. Treat set values as completed only when the user reports that they performed them or\n   explicitly asks to log them. Advice, targets, and planned sets are not completed work.\n4. Preserve every accepted exercise, load, unit, repetition, and set value while resolving\n   missing or ambiguous input. Ask only for the missing fact.\n5. Read `outcome`, `summary`, and optional `next_action` from the result envelope. Read\n   tool-specific questions, continuation tokens, confirmation tokens, choices, and returned\n   entities from `data`. Follow them exactly. Continue in the same turn when the result\n   authorizes the next tool call; otherwise ask the returned question without inventing an\n   answer.\n6. When `workout_track` returns `outcome=active` or `outcome=applied`, its\n   `active_workout` already holds the updated workout. Reply from it without calling\n   `workout_current`. This applies to starts, resumes, logged sets, corrections, removals,\n   and exercise reassignments. After `outcome=finished` there is no active workout; report\n   the returned recap. `outcome=edited` means a finished workout changed; it returns no\n   `active_workout`.\n7. Claim that a requested start, set, or finish happened only when `outcome` or `summary`\n   says that action was applied. Never infer this from `isError` being false. Recoverable\n   outcomes such as `invalid_input`, `conflict`, `version_conflict`, `not_found`, or\n   `workflow_expired` mean that nothing was applied; follow `next_action`. Summarize applied\n   exercise, load, unit, and reps briefly.\n8. When `open` or `add_sets` returns `outcome=recently_finished`, nothing was logged and\n   the user just finished a workout. Ask whether they are continuing it or starting a new\n   one; never choose for them. To continue, call `action=reopen`; for a new one, call\n   `action=open` with `new_workout=true`. Then resend any sets the user reported.\n9. Finish a workout only when the user's own words end the whole workout, such as \"I'm\n   done for today.\" Finishing a set or an exercise is not finishing the workout; phrases\n   such as \"that's good,\" \"next,\" or \"done with this exercise\" do not finish it. If the\n   user says after a finish that they were not done, call `action=reopen`.\n10. If a continuation or confirmation expires, preserve the accepted user input, prepare\n    the change again, and obtain any newly required confirmation. Never report the expired\n    change as applied.\n\nWhen prior workout state makes a spoken quantity structurally ambiguous, such as \"same\nweight, eight more\" after one set, ask whether the user means one set of eight reps or eight\nadditional sets before writing. Do not discard the values that were already clear.\n\n## Change a past workout\n\n1. Find the workout with `find_workouts`.\n2. Call `workout_track` with its `workout_ref`, the exercise, and `add_sets`,\n   `correct_set`, `remove_set`, or `correct_exercise`. Pass `set_number` for a set and\n   `occurrence_number` when the exercise appears more than once. Take both from the\n   returned workout, using `workout_get` for full detail; never guess them.\n3. Read the prepared change back and confirm only after the user's explicit yes, even when\n   they asked for the change.\n\nWithout `workout_ref`, a correction or removal targets the workout in progress. When none\nis in progress, it falls back to the workout just finished and needs the same explicit yes.\nNever use `workout_record_completed` to patch an existing workout.\n\n## Keep responses workout-paced\n\nDuring a live session, keep confirmations short and immediately useful. Report only facts\nreturned by GoodSet, then prompt for the next completed set or exercise when appropriate.\nDo not expose internal references, tokens, schemas, or confirmation handles.\n"},"changes":[],"summary":"First saved snapshot. No earlier version is available for comparison.","summary_kind":"deterministic","summary_metadata":{}}