The Agent Skills Directory

[COMMAND_EXECUTION]: The skill workflow involves executing local project-specific commands for testing, linting, building, and running migrations (e.g., 'tests, type checks, linters, formatters, build, migrations'). These are standard operations for a development-focused AI agent to verify that refactoring steps haven't introduced regressions.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads and processes project files which could contain malicious instructions.
Ingestion points: The agent reads repository files, architecture documentation, and current project state (Workflow step 1).
Boundary markers: No specific boundary markers or 'ignore embedded instructions' warnings are mandated when processing the codebase.
Capability inventory: The agent has the ability to write to the filesystem (refactoring code and progress files), execute arbitrary project commands (tests/builds), and perform Git commits.
Sanitization: No explicit sanitization or validation of the untrusted project content is mentioned.
[SAFE]: The use of a temporary file path in '/tmp' (e.g., '/tmp/refactor-{projectname}.md') for tracking state is a standard pattern for local development tools. While predictable paths in shared temporary directories can carry minor risks in multi-user environments, it is appropriate for this skill's use case.

architecture-satisfaction