full-t3-clean
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill performs destructive local operations using Git and filesystem tools. It is instructed to use
git worktree remove --force --force,git reset --hard, andgit clean -fdto discard uncommitted work and reset repository states. - [PROMPT_INJECTION]: The instructions explicitly direct the agent to skip user confirmation for destructive actions, stating that invocation 'authorizes all of the above without asking for confirmation, without backups'. This circumvents typical safety guardrails for irreversible file operations.
- [INDIRECT_PROMPT_INJECTION]: The skill represents an attack surface for indirect injection by processing untrusted data from the local environment.
- Ingestion points: Reads state from
git worktree list,git branch, andt3_thread_listas described in SKILL.md. - Boundary markers: Absent. No delimiters or safety instructions are used to treat these strings as data.
- Capability inventory: Capabilities to delete files, remove worktrees, and modify branch state (
git clean,git branch -D,git worktree remove) across the local filesystem. - Sanitization: Absent. There is no mention of sanitizing or validating identifiers retrieved from the local environment before they are passed to execution tools.
Audit Metadata