The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use local CLI tools, specifically git and forest, for managing worktrees and repository state. It includes safety guidelines that forbid the agent from inventing commands or installing missing tools without authorization.
[INDIRECT_PROMPT_INJECTION]: The skill has a broad ingestion surface as it reads repository content, including code, tests, and documentation, to reconcile task status.
Ingestion points: Processes README.md, NOW.md, project specifications, and source code files to build and update the delivery board.
Boundary markers: The skill uses structured markdown templates (EPIC.md, NOW.md) to manage state, though it lacks explicit instructions for sandboxing untrusted input from general project files.
Capability inventory: Possesses file-writing capabilities within the .agents/roadmap/ directory and can execute version control commands via git and forest.
Sanitization: No explicit sanitization or filtering of external file content is described in the instructions.
[SAFE]: The skill is well-documented and focused on coordination. It explicitly cautions against unauthorized execution and requires verifiable evidence (tests, demos) before advancing task status, which helps prevent accidental or malicious state changes.

task-master