The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes extensive shell logic using bash heredocs for version control management, file system organization, and tool invocation.\n
Evidence: references/phase2-scripts.md and references/phase3-gate-logic.md include scripts for git push, git checkout, mkdir, and prettier.\n- [PROMPT_INJECTION]: The skill uses extremely forceful language to override agent behavior and processes external data from local plan files, creating an attack surface for both direct and indirect injection.\n
Evidence: SKILL.md employs mandatory constraints like 'NON-NEGOTIABLE' and 'STOP AND READ', while references/todo-merge-strategy.md instructs the agent to integrate data from ~/.claude/plans/*.md.\n- [DATA_EXFILTRATION]: The workflow is designed to automatically sync local project artifacts and architectural records to remote git servers.\n
Evidence: references/phase2-scripts.md contains automated git push commands that transmit local documentation to the repository's origin.\n- [REMOTE_CODE_EXECUTION]: The skill executes local Python scripts via the uv tool with paths constructed from environment variables.\n
Evidence: references/preflight-checkpoint.md invokes uv run "$PLUGIN_DIR/.../preflight_validator.py" using a dynamically calculated plugin directory path.

go