The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the repository's file system, including code diffs, commit logs, and a local configuration file (.gitkkal/config.json), which could contain malicious instructions.\n
Ingestion points: SKILL.md workflow steps 3 and 5 involve reading git diff, git log, and git status output. Step 1 loads a local configuration file.\n
Boundary markers: The skill explicitly requires the agent to 'Present the commit plan before execution' and 'Ask for user confirmation before staging and committing'.\n
Capability inventory: The skill uses git status, git diff, git log, git add, and git commit commands.\n
Sanitization: The mandatory user confirmation step serves as a primary defense against unauthorized or malicious actions derived from ingested data.\n- [COMMAND_EXECUTION]: The skill instructs the agent to execute Git commands on the host system.\n
Evidence: The workflow details execution of git add <path> and git commit in steps 9 and 10.\n
Mitigation: The skill specifically forbids bulk staging (e.g., git add .) and requires sequential execution to prevent index corruption. It also explicitly excludes known sensitive files, such as .env files and private keys, from being staged.

gitkkal-commit