The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes uncommitted diffs from the local repository to determine file groupings and generate commit messages or PR descriptions. Adversarial content within the diff (e.g., code comments or text in documentation) could potentially influence the agent's behavior.
Ingestion points: git status and the content of the diff itself are ingested as described in SKILL.md (Metadata section) and references/block-cycle.md (Baseline Step).
Boundary markers: No explicit delimiters or instructions to ignore instructions embedded in the diff are specified in the provided files.
Capability inventory: The skill performs shell commands (git, gh) and writes to the filesystem (state file, project files) as outlined in references/block-cycle.md.
Sanitization: No filtering or sanitization of the processed diff content is documented in the instruction set.
[COMMAND_EXECUTION]: The skill executes shell commands for version control and pull request management.
Evidence: references/block-cycle.md includes bash scripts for backing up branches, staging files, committing changes, and pushing to remotes.
[EXTERNAL_DOWNLOADS]: The skill interacts with GitHub for pull request creation.
Evidence: references/block-cycle.md (Step 6) uses the gh tool to interact with github.com, which is a well-known service.

splitting-pr