The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from the repository to generate commit messages and pull request descriptions, which could be exploited if malicious content is present in the codebase.
Ingestion points: Test titles matched via regex in scripts/select-task.js and the rationale field returned by the task selection script.
Capability inventory: The skill can modify source code, execute local project scripts (pnpm), perform Git operations, and use the GitHub CLI (gh) to open pull requests.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions when the agent interpolates ingested data into its final outputs.
Sanitization: No sanitization or validation is performed on the extracted test titles or rationales before they are included in generated text.
[COMMAND_EXECUTION]: The skill relies on executing various CLI tools to fulfill its maintenance objectives.
Evidence: The skill uses mise exec to run pnpm scripts, node for task selection, and git and gh for version control management.
Context: These commands are necessary for the intended function of an automated maintenance agent and are limited by strict line-count and file-count constraints.
[DYNAMIC_EXECUTION]: The skill involves the generation and execution of code modifications to improve the repository.
Evidence: The agent identifies bugs or coverage gaps, writes code fixes, and immediately executes test suites to verify the changes.
Context: This behavior is the primary intended function of the agent, and risks are mitigated by hard constraints on the scope of changes.

agent-upkeep