The Agent Skills Directory

[SAFE]: The skill follows established software development patterns for code auditing and refactoring. It implements session-specific workspace boundaries (.workflow/scratch/) and requires user confirmation for knowledge persistence and code fixes unless the automated -y flag is explicitly provided.
[COMMAND_EXECUTION]: The skill uses the Bash tool to perform git history analysis and version control operations (log, blame, diff). These operations are restricted to the target file list and follow a predefined state machine logic.
[PROMPT_INJECTION]: The workflow involves reading and editing source code (Category 8: Indirect Prompt Injection). There is an inherent risk that untrusted content in the target files could influence the agent's behavior during the fix or generalization phases. However, the skill mitigates this through a structured review process and mandatory re-review of modified areas.
Ingestion points: Target source files identified during the intake phase (SKILL.md).
Boundary markers: None explicitly defined for untrusted file content.
Capability inventory: Includes Edit, Write, and Bash tools used for modifying the codebase (SKILL.md).
Sanitization: No explicit sanitization of input code is performed, relying instead on the agent's analytical capabilities and user review cycles.

odyssey-review-test-fix