The Agent Skills Directory

[SAFE]: No malicious patterns or security vulnerabilities were detected during the analysis of the skill. The logic follows git best practices for history manipulation.- [COMMAND_EXECUTION]: The skill executes local git commands to manage branch history. It includes specific safety checks to prevent execution on protected branches (e.g., main, master), ensures the working tree is clean, and uses absolute paths for isolated worktrees to prevent path-related errors.- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from commit messages and code diffs which could potentially contain malicious instructions. The risk is effectively mitigated by the skill's architecture.
Ingestion points: Commit history and diff content read in SKILL.md.
Boundary markers: Explicit instructions to the agent to never skip the review gate, even if the user requests it.
Capability inventory: Local file system modification through git reset --hard and remote repository updates via git push --force-with-lease.
Sanitization: Content integrity is verified using a byte-for-byte diff comparison between the worktree and the original branch before applying changes.

recomposing-commits