The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources which could potentially contain malicious instructions.
Ingestion points: Branch names via git branch, PR titles and metadata via gh pr list, and Linear issue summaries/team names via MCP (Step 3, 4, and 6 in SKILL.md).
Boundary markers: Absent. The skill instructions do not specify using delimiters or "ignore instructions" warnings when presenting these external strings (like PR titles) to the agent context.
Capability inventory: The agent has the ability to delete local and remote branches, remove worktrees, delete filesystem directories using the bundled script, and update Linear issue states.
Sanitization: The skill uses a strict, deterministic regex pattern for extracting issue IDs from branch names, which reduces the risk of misinterpreting arbitrary branch tokens as commands.
[COMMAND_EXECUTION]: The skill executes destructive commands to clean the repository and filesystem.
Evidence: Instructions include git branch -D, git push origin --delete, and the execution of a bundled script scripts/filesystem-hygiene.mjs which uses rmSync for recursive deletion.
Context: These operations are consistent with the skill's primary maintenance purpose and are protected by mandatory user confirmation gates for each pass and a comprehensive dry-run mode.

cleanup-repo