The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads untrusted data from the local repository to help the agent compose commit messages. This creates a surface where malicious instructions embedded in file contents or diffs could influence the agent's behavior during the commit message generation step.
Ingestion points: scripts/git-commit-context.sh reads and outputs the contents and diffs of changed files in the working tree.
Boundary markers: None present; file contents and diffs are printed directly to the agent's context without delimiters or instructions to ignore embedded commands.
Capability inventory: scripts/git-commit-push.sh can perform git add, git commit, and git push operations.
Sanitization: No sanitization or filtering of file content is performed before it is provided to the agent.
[COMMAND_EXECUTION]: The script scripts/git-commit-push.sh executes git add -A, which stages all changes in the working tree. As noted in the skill's 'Gotchas' section, this behavior may include files that the agent's generated message does not describe, potentially leading to the accidental commitment and pushing of sensitive data (e.g., local configuration files, logs, or secrets) if they are not explicitly listed in .gitignore.

pflow-commit