The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes git and gh commands to interact with the local repository and GitHub API. Evidence: SKILL.md and context files contain numerous calls to git worktree, git rev-parse, git log, and gh pr list.
[COMMAND_EXECUTION]: The skill uses dynamic context placeholders to retrieve repository status when loaded. Evidence: SKILL.md uses !git branch --show-current and `!`git worktree list to provide the agent with current context. These commands are benign and standard for development workflows.
[DATA_EXFILTRATION]: The skill facilitates the propagation of local project configuration and settings (e.g., .claude/settings.local.json) to new worktrees. Evidence: context/create.md explicitly mentions checking for and copying these files to ensure the environment is functional.
[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests data from external sources like branch names and PR metadata. 1. Ingestion points: SKILL.md and context/status.md read output from git worktree list and gh pr list. 2. Boundary markers: The skill instructions emphasize structured parsing of porcelain output and batched API responses. 3. Capability inventory: The skill can execute shell commands (git, gh), write to the filesystem (creating worktrees), and delete directories (rm -rf in context/cleanup.md). 4. Sanitization: Branch names are strictly validated against a regex allowing only alphanumeric characters, dots, underscores, and dashes, with a 64-character limit as described in context/create.md.

worktree