The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local repository (via git diff) to generate commit messages and determine commit types. An attacker could potentially embed malicious instructions within code comments or documentation that the agent might misinterpret as instructions.
Ingestion points: SKILL.md (Step 1 in flows) reads data using git diff --staged and git diff.
Boundary markers: The skill explicitly instructs the agent to "not invent" information and mandates a manual confirmation step ("Propuesta de commit") for every operation.
Capability inventory: The skill can execute local git commands (git add, git commit, git reset, git restore) as defined in the "Flujo" sections of SKILL.md.
Sanitization: Automated secret scanning is performed using grep (defined in references/secret-detection.md) to detect and block the inclusion of credentials or sensitive file types before committing.
[COMMAND_EXECUTION]: The skill utilizes several local shell commands to interact with the Git version control system.
Evidence: SKILL.md and references/secret-detection.md include commands such as git status, git diff, git add, git commit, and a complex grep pipeline for secret detection. These are restricted to the local environment and are standard for the skill's purpose.

git-commit