The Agent Skills Directory

[COMMAND_EXECUTION]: Uses the Bash tool to run non-destructive git commands for state analysis. It requires explicit user confirmation before performing any commits and forbids the use of dangerous flags like --no-verify or force-pushing.
[EXTERNAL_DOWNLOADS]: Installation instructions point to the developer's GitHub repository. The skill does not perform any remote downloads or network requests at runtime.
[DATA_EXFILTRATION]: All file writes are confined to the project root, specifically within a tasks/ directory. No sensitive data is transmitted externally.
[PROMPT_INJECTION]: The skill processes project-specific data to maintain continuity between sessions. * Ingestion points: Reads tasks/SESSION.md and tasks/ROADMAP.md to establish context. * Boundary markers: Uses structured Markdown templates to organize data, though it lacks explicit delimiters to ignore embedded instructions. * Capability inventory: Includes access to Bash, Read, Write, and Edit tools for local project management. * Sanitization: No explicit sanitization of previously written log content before it is re-interpolated into new files.

wrapup