Search

---
name: search
description: Search the FragCap network for capsules matching a query. Shows ranked results from the registry, then fetches full details or installs as a local skill on demand.
argument-hint: <query>
---

# Search Capsules

Search across all FragCap users for capsules relevant to the user's current problem. Uses the GitHub Pages registry (no rate limit, no auth required).

## Script Convention

    FRAGCAP_DATA="${CLAUDE_PLUGIN_DATA}" node "${CLAUDE_PLUGIN_ROOT}/scripts/<n>.mjs" [args]

If a script exits with a non-zero code or returns `{ error: "..." }`, **stop immediately and surface the error to the user**. Never silently retry with a different command or skip the step.

## When to Activate

- User runs `/fragcap:search <query>`
- User asks "has anyone solved this before" or similar

## Flow

1. **Get query**:
   - If `$ARGUMENTS` is provided, use it as the search query.
   - Otherwise, ask: "What problem are you working on?"

2. **Search** — run `search.mjs <query>`. It returns `{ results: [...], total_found, keywords }`.

3. **Display results** — show a compact numbered list with status, problem, tags, and author.
   - If no results: "No matching capsules found. Try broader keywords or different terms."
   - If `total_found` > 5: mention there are more results.

4. **Detail on demand** — present options inline matching the numbered list above, do not ask the user to type a gist ID:
   ```
   [1] View details  [2] View details  ...  [0] Done
   ```
   For each selected result, run `fetch-capsule.mjs <gist-id>` and present the full capsule content (problem, attempts, pitfalls, solution, snippet). The response also includes `raw_skill` — the full SKILL.md content.

5. **Install option** — after viewing details, offer:
   ```
   [1] Install as local skill  [2] Back to results
   ```
   If the user picks install, run `install-capsule.mjs <gist-id>`. This writes the SKILL.md to `.claude/skills/` in the current project, making it automatically available in future sessions.

6. **Apply** — if a capsule is relevant, suggest how it might apply to the user's current work.

## Error Handling

- If `search.mjs` fails: show the error message and stop.
- If `fetch-capsule.mjs` fails on a selected result: show the error and offer to try another result.

## Guidelines

- Keep the initial list concise — problem + tags only.
- Only fetch full details for results the user explicitly selects.
- Suggest more specific terms if results are poor.