The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes a vulnerability surface by ingesting untrusted source code, directory structures, and pull request diffs, as specified in the 'Workflow' and 'Scope' sections of SKILL.md. This data is processed as part of the agent's context, potentially allowing an attacker to embed instructions within a codebase to influence the agent's behavior. The skill instructions do not mandate the use of explicit boundary markers (e.g., XML tags or unique delimiters) or input sanitization for the analyzed content.
[COMMAND_EXECUTION]: The skill is configured to execute external linting and validation tools, including shellcheck, ruff, mypy, eslint, and infrastructure validators for Terraform, Ansible, and Helm, as detailed in Workflow Step 3 and the maintainability-patterns.md reference. Although the skill employs a security-conscious approach by restricting these to 'non-mutating check commands' and forbidding '--fix' or '--write' flags, the capability to spawn subprocesses based on untrusted project data represents a potential execution surface.

code-simplification