The Agent Skills Directory

[COMMAND_EXECUTION]: The skill automates several git operations and uses specialized tools to manage the repository.
The scripts/git-commit.sh bash script validates and runs git commit to finalize changes, using robust validation against control characters and invalid formats.
Step 7 in SKILL.md invokes skill-changelog with parameters to update the CHANGELOG.md file, which involves writing to the local file system.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) by ingesting untrusted commit subjects from the repository history.
Ingestion points: Commit messages are ingested from the git repository during the version decision process (Step 6) and changelog generation.
Boundary markers: The skill does not define specific boundary markers or instruct the agent to ignore instructions found within commit messages, creating a surface where malicious commit text could be interpreted as instructions.
Capability inventory: The agent can write to project files (CHANGELOG.md) and execute git commands such as push and commit.
Sanitization: While the git-commit.sh script validates outgoing messages effectively, historical commits retrieved from the repository are not sanitized or validated before being evaluated by the agent's decision logic.

git-guardrails