The Agent Skills Directory

[DATA_EXPOSURE]: The skill implements strict exclusion rules for credentials and secrets. SKILL.md (Section 4) and references/file-selection.md explicitly direct the agent to leave secret-related paths unstaged and verify their absence before any commit.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from git diffs and file contents.
Ingestion points: The skill reads git diff, git status, and repository files in SKILL.md (Steps 1, 2, 3, 6).
Boundary markers: Mandatory verification gates are established using git diff --cached --check and hunk-level inspection (Step 6) to ensure only intended changes are committed.
Capability inventory: The skill utilizes git state-mutating commands, shell redirections, and local utility commands (printf, grep, cat).
Sanitization: Employs single-quoted heredocs (<<'EOF') in references/history-and-trailers.md to prevent shell variable expansion or command substitution when transporting commit messages.
[COMMAND_EXECUTION]: The skill utilizes standard Git porcelain commands for repository management. The implementation of GIT_SEQUENCE_EDITOR=: in references/history-and-trailers.md for non-interactive history folding is a standard Git practice and does not introduce arbitrary execution risks. All shell operations are focused on local repository maintenance.

vibe-commit