Back to catalog / Agent workflows
Schedule read-only repository reviews and verify finding continuity
Run recurring reviews with structured verdicts/findings and have each follow-up prove which prior issues remain.
Goal
Build a bounded, verifiable multi-cycle repository review.
Steps
- Seed a disposable synthetic repository with known defects and define a JSON Schema for verdict, finding IDs, and files. [scheduled]
- Limit
toolsto Read/Glob/Grep and bound unattended runs withpermission_mode="dontAsk",max_turns, andmax_budget_usd. [scheduled] - Resume with
resume=session_id, reread the current repository, and returnprevious_finding_ids,resolved, and surviving finding IDs. [scheduled]
How to verify
Against known fixed/new defects, verify verdict and retained/resolved/new finding IDs; confirm budget/turn failures emit no completion marker.
Troubleshooting
If the session is lost, create a fresh baseline; never present prior findings as a current repository scan.
Permissions
Prerequisites
- Python 3.11+, an Anthropic API key, Claude Agent SDK, and a local test repository.
Permissions
- Review only an approved repository; store transcripts/logs in restricted locations and exclude real credential files.
Limitations
The example uses local transcripts; protect and retain session files deliberately. Structured output still needs review, and read-only tools do not provide OS isolation.
Pinned upstream source
- Repository
- anthropics/anthropic-cookbook
- File path
- claude_agent_sdk/scheduled_repository_reviewer/scheduled_repository_reviewer.ipynb
- Pinned commit
9c1e493ddd2b82df399d2a61b9c806c539e18be6- Source checked
- 2026-10-14 · Reviewed against the pinned upstream source
Repository metrics
These values describe the upstream repository, not popularity or likes for an individual skill or playbook.
- Stars
- Not collected
- Forks
- Not collected
- Observed at
- Not collected
No metrics snapshot is available yet.
License: MIT
Main topic
Turn ecommerce campaign materials into a reviewable weekly brief