The Agent Skills Directory

Indirect Prompt Injection (HIGH): The skill processes untrusted data from the local repository (commit messages, diffs, and history) and possesses high-privilege write capabilities.
Ingestion points: Data enters the agent context through git log, git reflog, git show, and git diff outputs.
Capability inventory: The skill uses the Bash tool to execute commands like git reset --hard, git restore, and git checkout, which modify the filesystem and repository state.
Boundary markers: No explicit delimiters are used to isolate git output from agent instructions.
Sanitization: No sanitization or filtering of commit messages or file content is performed before processing.
Command Execution (MEDIUM): The skill executes powerful shell commands via Bash.
Destructive Operations: Scenarios 2, 4, 5, and 6 involve git reset --hard or git restore, which can permanently discard uncommitted changes or modify history.
Risk: While the skill includes safety warnings and verification steps, an attacker-controlled commit message could theoretically attempt to escape the git command context or manipulate the agent into bypassing these safeguards.

git-recovery