Agent Self Improve
Diagnose recurring communication failures in agent conversations and stage the smallest durable corrections for approval.
When to use
Section titled “When to use”Diagnose recurring communication failures in supplied or available agent conversations and stage minimal guidance corrections for approval. Use only when explicitly requested.
Link this skill
Section titled “Link this skill”arcantry skills link agent-self-improve --scope userThe standard destination is ~/.agents/skills. Codex reads this universal Agent Skills location directly.
Claude compatibility
Section titled “Claude compatibility”arcantry skills link agent-self-improve --scope user --compat claudeThis keeps the universal link and adds a Claude alias to the same canonical package.
Examples
Section titled “Examples”Review a frustrating project task
Section titled “Review a frustrating project task”Prompt
Use the agent-self-improve skill to review this project’s recent tasks. Focus on why the agent kept asking questions the repository already answered.
Expected outcome
A diagnosis tied to conversation evidence and numbered, approval-gated edits for the narrowest guidance layer.
Evaluate a supplied transcript
Section titled “Evaluate a supplied transcript”Prompt
Use the agent-self-improve skill on this transcript and explain whether the failure came from execution, a skill, or project instructions.
Expected outcome
A scoped cause analysis that does not blame prompt wording or edit agent configuration without approval.