The Agent Skills Directory

[SAFE]: The skill implements a legitimate workflow for autonomous improvement loops. It focuses on establishing clear completion criteria and verifiable checks, which is a security best practice for autonomous agents.
[COMMAND_EXECUTION]: The skill workflow involves executing standard development commands (such as npm test or yamllint) during the verification phase to confirm if targets are met. These operations are within the expected scope of a development-focused skill.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data provided by the user (described in Phase 1) and uses it within a loop that can execute shell commands (Phase 4).
Ingestion points: User-provided code, text, or file paths received in Phase 1 of SKILL.md.
Boundary markers: None explicitly defined in the instructions to separate user data from agent instructions.
Capability inventory: Execution of shell commands for verification (e.g., running tests, word counts) as described in SKILL.md Phase 4b.
Sanitization: No specific sanitization or filtering of external content is mentioned.
[SAFE]: The skill records its progress in improve-history.md. This is a transparent local logging mechanism and does not constitute a persistence or exfiltration risk.

improve