9 added, 22 removed. Audit A to A.
---
name: shellbrain-usage-review
- description: Use when a reviewer agent needs one fast cross-repo picture of how agents are using Shellbrain, where the product is working, where it is failing, and which expected capabilities are being skipped.
+ description: Review Shellbrain usage measurements across repositories and identify useful product follow-up.
---
# Shellbrain Usage Review
- Run the canonical cross-repo analytics report and synthesize it for review.
-
- ## Primary Command
-
- ```bash
- shellbrain admin analytics --days 2
- ```
-
- ## What To Extract
-
- - where Shellbrain is working well
- - where it is failing
- - what is failing and why
- - where working agents are skipping expected `recall`
- - where internal recall or knowledge-builder runs are failing
- - the top priorities that should drive product follow-up
+ Run `shellbrain admin analytics --days 2`.
- ## Operating Rules
+ The JSON report contains observed counts, command latency, empty retrieval results,
+ sync totals, and grouped errors with sample record IDs. It does not assign health
+ scores or product priorities.
- - treat the JSON report as the source of truth
- - do not reconstruct the report with ad hoc SQL unless the command is broken
- - summarize the strongest wins, the most important failures, and the highest-priority capability gaps first
- - use `repo_rollups` only as secondary supporting context
+ Use the measurements and supporting records to explain what works, what fails,
+ and which changes would help. Distinguish your interpretation from measured facts.
+ A successful call does not establish that the memory helped solve a task.
+ Do not infer missing recall from counts alone without checking task context.