The Agent Skills Directory

[COMMAND_EXECUTION]: The orchestrator uses the date command (date +%s) to track execution time and manage the duration-based loop. Subagents are further instructed to perform Git commits (git commit) after successful refactoring iterations.
[DYNAMIC_EXECUTION]: The skill's safety protocol requires subagents to execute the project's existing test suite before and after making code changes. This involves running arbitrary code and scripts defined within the user's project environment (e.g., test runners, hooks, or test cases).
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze an entire codebase to identify specific code patterns. Maliciously crafted content within the analyzed files (such as comments, documentation, or string literals) could attempt to provide conflicting instructions to the subagents during the refactoring process.
Ingestion points: Project files within the codebase and the progress log file stored in /tmp.
Boundary markers: No explicit boundary markers or instruction-isolation techniques are defined for the subagents when reading the codebase.
Capability inventory: Subagents possess the capability to modify project files, execute test commands, and perform Git operations.
Sanitization: The skill does not implement sanitization or validation of the codebase content before it is processed by the LLM subagents.

anneal