Bridges live capture data into Claude Code sessions. Exposes tools for Claude to check capture status and poll for new screenshots/transcripts from an active feedbacks recording session.
Four tools with clear verb-noun naming (feedbacks_status, feedbacks_poll, feedbacks_sessions, feedbacks_session). Descriptions are present and moderately detailed (100-200 chars), explaining WHAT each tool does and WHEN to use it. However, critical gaps: (1) feedbacks_poll and feedbacks_session lack explicit output schema documentation, the code builds complex nested structures (timeline segments with screenshots, transcripts, speech spans) but the tool definitions do not document the return type; (2) parameter descriptions are minimal, 'since' and 'session_id' in feedbacks_poll lack format/range guidance; (3) no error handling guidance in descriptions (e.g., what if the HTTP server is unreachable, or session_id is invalid?); (4) no mention of idempotency or side effects. The server instructions are comprehensive, but tool-level definitions are incomplete. Baseline: 194 chars avg description (these are ~120-150), 100% of A+ tools document return types (this server does not).
Poll for new live timeline events (screenshots and transcripts) from an active recording session. Pass latestSeqNum as since for incremental updates. Returns grouped timeline segments with screenshots as file paths and transcripts inline.
Get a specific saved session's full timeline with screenshots as file paths (read with the Read tool to view images) and transcripts inline, grouped by speech spans. Mirrors the session.md structure.
List all saved feedback sessions with dates, durations, and AI summaries. Returns metadata for each past session.
Check for an active recording session. Returns active status, session ID, start time, and latest sequence number for polling.
Output schemas not documented. feedbacks_poll and feedbacks_session return complex nested structures (timeline segments, screenshots, transcripts, speech spans) but tool definitions do not specify return type, field names, or structure. LLMs cannot plan downstream operations or extract data reliably.
Parameter descriptions lack format/range guidance. 'since' in feedbacks_poll is described as 'Sequence number' but does not specify valid range (0 - N?), whether it is inclusive/exclusive, or what happens if invalid. 'session_id' lacks format hints.
No error handling guidance. Descriptions do not explain what happens if the feedbacks HTTP server is unreachable, session_id is invalid, or no events exist. LLMs cannot recover from failures without explicit guidance.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | 2026-07-28+ | v2 |
No idempotency or side-effect declarations. Descriptions do not state whether tools are read-only (they are marked READ_ONLY in metadata, but not in descriptions). LLMs should know which calls are safe to retry.
Missing pagination guidance. feedbacks_sessions and feedbacks_session do not document result limits, pagination, or how to handle large session lists. If 100+ sessions exist, response could exhaust context window.