The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool for git operations and task execution. It implements a PreToolUse hook that executes a local script, guard-git-hygiene.sh, which in turn uses python3 to analyze and potentially block shell commands. Evidence: hooks section in SKILL.md and scripts/guard-git-hygiene.sh.\n- [PROMPT_INJECTION]: The skill processes task plans as input, which serves as an indirect prompt injection surface. These plans guide autonomous tool usage and code modification. \n
Ingestion points: The plan argument and JSON files located in the .ulpi/plans/ directory.\n
Boundary markers: The skill relies on a manual human approval step; technical delimiters for plan content are not explicitly defined in the prompt.\n
Capability inventory: The skill possesses extensive capabilities including Bash execution, file system modifications (Read, Write, Edit), and subagent spawning.\n
Sanitization: Structural validation of the plan is performed by validate-plan.mjs, although this does not filter the semantic content of the task instructions.

auto-build