Add fitness exercise finder skill

skills/find-fitness-exercises/SKILL.md

@@ -0,0 +1,42 @@
+---
+name: find-fitness-exercises
+description: Search the hasaneyldrm/exercises-dataset collection for fitness exercises by natural-language query, body part, equipment, target muscle, or supporting muscle, then return concise reference examples with multilingual steps, thumbnails, animation GIFs, source links, and attribution. Use when a user asks to find, compare, identify, or demonstrate exercises, including Chinese requests such as “找几个哑铃练胸动作” or “推荐徒手练背动作”.
+---
+
+# Find Fitness Exercises
+
+Use the bundled deterministic search script to ground exercise matches in the 1,324-record dataset. Do not invent an exercise, instruction, equipment requirement, or media URL.
+
+## Search workflow
+
+1. Extract the user's query, body-part, equipment, target-muscle, language, and result-count constraints. Default to the user's language and 5 results.
+2. Run `scripts/search_exercises.py` from this skill directory. Prefer JSON when composing a tailored answer:
+
+   ```bash
+   python scripts/search_exercises.py "哑铃练胸" --equipment 哑铃 --body-part 胸部 --lang zh --limit 5 --format json
+   ```
+
+   For a ready-to-present result, request Markdown:
+
+   ```bash
+   python scripts/search_exercises.py "bodyweight back" --equipment "body weight" --lang en --format markdown
+   ```
+
+3. Present the strongest matches. For each result, include the name, target, equipment, a brief match reason, 2–4 key steps when available, the animation or thumbnail link, and the dataset-provided attribution.
+4. State any inferred constraints. If results are weak or empty, relax only an inferred constraint, rerun once, and tell the user what changed.
+5. Keep dataset lookup separate from programming advice. Give sets, reps, load, progression, injury modifications, or a full workout only when requested and label them as general guidance.
+
+## Script options
+
+- Use `--body-part`, `--equipment`, and `--target` for hard filters. Chinese aliases such as `胸部`, `自重`, `哑铃`, `腹肌`, and `二头肌` are accepted.
+- Use `--lang en|es|it|tr|ru|zh|hi|pl|ko|fr` to select instruction language.
+- Use `--limit 1..20` and `--format json|markdown` to control output.
+- Use `--list-values` to inspect canonical filter values.
+- Use `--dataset PATH` for a custom local copy. Otherwise the script finds this repository's `data/exercises.json` or downloads the upstream file into a temporary cache.
+
+## Response guardrails
+
+- Preserve `© Gym visual — https://gymvisual.com/` on every response that displays or links media.
+- Link back to `https://github.com/hasaneyldrm/exercises-dataset` as the data source.
+- Treat the animations and instructions as references, not individualized medical advice. Tell the user to stop if an exercise causes pain and seek qualified help when injury, pregnancy, rehabilitation, or a medical condition is involved.
+- Read [references/dataset-notes.md](references/dataset-notes.md) when explaining coverage, fields, provenance, or licensing.

skills/find-fitness-exercises/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "Fitness Exercise Finder"
+  short_description: "Search exercises with filters and reference media"
+  default_prompt: "Use $find-fitness-exercises to find suitable exercises and show concise reference examples."

skills/find-fitness-exercises/references/dataset-notes.md

@@ -0,0 +1,20 @@
+# Dataset notes
+
+## Source and coverage
+
+- Source: `https://github.com/hasaneyldrm/exercises-dataset`
+- Default branch: `main`
+- Records: 1,324 exercises in `data/exercises.json`
+- Searchable metadata: `name`, `category`, `body_part`, `equipment`, `target`, `muscle_group`, and `secondary_muscles`
+- Instructions: English, Spanish, Italian, Turkish, Russian, Chinese, Hindi, Polish, Korean, and French
+- Reference media: one 180×180 thumbnail and one animation GIF per record
+
+The dataset does not contain difficulty ratings, contraindications, sets, reps, loads, or individualized programming. Do not present inferred values for those fields as dataset facts.
+
+## Licensing
+
+The repository's code, tooling, dataset structure, and instruction text/translations use the MIT License. The files under `images/` and `videos/` are excluded from that grant. They are owned by Gym visual, included upstream with permission, and must retain this attribution:
+
+`© Gym visual — https://gymvisual.com/`
+
+Linking to an upstream media file does not grant reuse rights. When asked about copying, redistributing, or commercial use of the media, direct the user to the repository `LICENSE`, `NOTICE.md`, and Gym visual's terms.