The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill identifies files for deletion by scanning the repository (including plans, specs, and reports) using grep. Because the agent interprets content from these external files to determine if they are 'orphaned' or 'zombie' items, malicious instructions hidden within these files could potentially influence the agent's cleanup actions.
Ingestion points: Repository files (plans/specs/reports), .harness/ state files, and git status output.
Boundary markers: There are no specified delimiters or instructions to treat file content as untrusted data during the audit process.
Capability inventory: The skill is authorized to perform file deletions and execute system commands like git worktree remove.
Sanitization: The skill lacks explicit sanitization or verification steps for the data retrieved from the audited files.
[COMMAND_EXECUTION]: The skill uses a variety of git commands to inspect and manage the repository state.
Evidence: The workflow includes execution of git worktree list, git status --porcelain, git check-ignore, git log --follow, and git worktree remove.

sweep