| name | self-review |
| description | Run an agent self-review for skill and prompt improvements, including script extraction opportunities and token-conscious workflow upgrades. |
| argument-hint | Review scope, urgency, and output format |
| user-invocable | true |
| disable-model-invocation | false |
Self Review
Inherits from task-core-loop.
Added Context
Use this to periodically review and improve:
- skill design and overlap
- agent prompt clarity
- token-cost hotspots
- script-extraction opportunities for repeated terminal commands
- chat tool usage parity across prompt, runtime, MCP, and coding-agent surfaces
Prefer MCP tooling first for reads/writes where available; propose script hooks only for repeated operations that are not already covered by MCP tools.
Additional Procedure
- Review active skills for duplication and missing shared abstractions.
- Identify loops that should move from chat polling to script-driven waits.
- Propose prompt updates that reduce ambiguity and token use.
- For repeated manual content generation, propose hook-script outputs that feed MCP write tools.
- Treat chat tool usage parity as a first-class requirement: document any intentional mode differences and flag undocumented drift.
- Update requests pages under
Data/Pages/requests with:
- a master outstanding list
- grouped changes for small items
- dedicated pages for significant changes
- Update or create
/tasks records when a recommendation is significant enough to merit execution tracking.
- Mark each recommendation as now, next, or later with confidence.
Output
- Improvement shortlist with impact and effort.
- Proposed edits/scripts with evidence.
- Updated requests pages reflecting outstanding work.