The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data that could potentially contain malicious instructions. Ingestion points: All code comments and markdown documentation files in the repository worktree (SKILL.md). Boundary markers: The skill does not define explicit delimiters for ingested text but requires findings to be evidence-backed and cited. Capability inventory: Local repository auditing via git and manual file modification handoff to duck-patch (SKILL.md). Sanitization: None explicitly defined for analyzed text. This risk is mitigated by a robust phase-gate approval process and the requirement for explicit user confirmation on all semantic changes.
[COMMAND_EXECUTION]: The skill utilizes standard git commands for repository analysis, such as identifying the default branch and calculating diffs. These operations are performed locally and are restricted to auditing the worktree state, posing no significant security risk.
[DYNAMIC_EXECUTION]: The skill identifies and flags outdated documentation and comments but does not execute them as code. The actual modification of files is handed off to a separate patching mechanism (duck-patch) under strict user supervision.

duck-tidy