Public agent skill · SKL-002
Overseer
Project context and file hygiene that stays clean, current, and recoverable.
01
The problem
What the system needed to solve.
Long-running agent work leaves stale instructions, duplicated context, temporary files, and documentation that no longer matches the repository.
Cleanup is risky when an agent cannot distinguish generated debris from user-owned work.
02
The approach
How I directed the work.
Designed a scan-first workflow that inventories changes before proposing cleanup.
Separated safe reconciliation from destructive actions and kept recoverability as a core constraint.
Added automated coverage around the workflow so hygiene work can be repeated without relying on memory.
03
Delivered and owned
What was built, and what I owned.
Measured or shipped
- Public Python project with a standard-library test suite.
- Documented scan, audit, and approval-gated clean modes.
- Project content is archived rather than permanently deleted by the workflow.
Owned surface
- Safety boundaries and cleanup model
- Documentation reconciliation rules
- Test cases and failure-path review
- Public documentation
Agent workflowsDocumentation systemsRepository hygieneAutomated tests
04
Continue through the ledger