The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data to generate commit plans. 1. Ingestion points: Repository history (git log) and working tree changes (git diff). 2. Boundary markers: The skill presents a full plan for human confirmation before execution, which acts as a manual review boundary. 3. Capability inventory: The skill can write files, stage changes, and perform commits via the Bash tool. 4. Sanitization: It includes a guardrail to scan staged content for secrets, keys, and tokens before committing.
[DYNAMIC_EXECUTION]: The skill generates partial patches based on file diffs and applies them to the git index via the git apply command. It also executes the resolve-config.sh script included in the skill templates to handle JSON configuration resolution.
[COMMAND_EXECUTION]: Extensive use of git and shell utilities (jq, cksum, printf) to inspect repository state and enforce commit conventions.

atomic-commit