femto-retry-reads

/femto:retry-reads

Recovers from read failures. Re-invokes the femto-reader subagent for each declared read whose state is missing or malformed, passing the original mission plus the validation error from the previous attempt.

How to invoke

1. Resolve the session_id (most recent .femto/session-*/, or ask the user).

2. Call femto_get_session_status({ session_id }) and read reads_status.

3. For each entry whose state is missing or malformed AND whose retry_count < 2:

   • Look up the read's original mission from the pack at ${CLAUDE_PLUGIN_ROOT}/packs/<pack_id>/pack.json under reads[id == <read-id>].

   • Find the output_path from the original femto_start_session response (or reconstruct: .femto/session-<session_id>/reads/<read-id>.md).

   • Spawn the femto-reader subagent via the Task tool with prompt:

     <original mission>
     
     Output path: <output_path>
     Output schema: required_sections=<...> min_length_chars=<...> require_source_citations=<true|false>
     
     Previous attempt failed validation: <validation_errors joined with '; '>.
     Re-extract and write to <output_path>, ensuring the declared output schema is satisfied.
     

4. After all retries finish, call femto_get_session_status again and report which reads now succeed and which remain failing.

Constraints

• 2 retries max per read. If retry_count >= 2, do NOT invoke; surface the failure to the user and recommend they revise the pack's mission text.
• Don't retry a read whose state is ok. Idempotency only on failures.
• Don't run reads in parallel via separate Task calls. They write to disjoint files but the host's permission UX is cleaner sequentially.

On exhaustion

If a required read is at retry_count == 2 and still missing/malformed, output:

[femto] read `<read-id>` exhausted retries (3 total attempts).
Validation errors: <list>
Recovery: revise the pack's mission text under packs/<pack_id>/pack.json#reads, then start a new session.