The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to run numerous command-line tools for static analysis, security auditing, and quality checking on the target repository.
Evidence: The instructions in references/repo-audit.md list a broad set of tools including lint, typecheck, scorecard, reuse, gitleaks, zizmor, osv-scanner, npm/pnpm/cargo audit, jscpd, lizard, madge, tsc, ruff, and others.
[REMOTE_CODE_EXECUTION]: The skill requires the execution of project-defined build and test scripts to establish baselines and verify fixes. These scripts are provided by the repository being audited and can contain arbitrary, potentially malicious code.
Evidence: SKILL.md (line 35) and references/repo-audit.md (line 28) explicitly instruct the agent to run test, build, and smoke commands found within the target project.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and evaluates external, untrusted content from the repository being reviewed, such as source code, diffs, and project manifests, which could contain malicious instructions.
Ingestion points: The agent reads file content, git diff output, README.md, LICENSE, and manifest files (e.g., package.json, Cargo.toml) to build its mental model of the project (SKILL.md lines 14 and 27; references/repo-audit.md lines 20-25).
Boundary markers: The instructions do not define delimiters or specific safety markers to differentiate between the skill's logic and the untrusted data it processes.
Capability inventory: The skill grants the agent the ability to execute shell commands, run diverse auditing tools, and modify the file system by applying patches (SKILL.md lines 35 and 41; references/repo-audit.md lines 35-43).
Sanitization: There are no requirements for sanitizing or filtering data ingested from the repository before it is used to generate findings or code modifications.
[DYNAMIC_EXECUTION]: The skill involves the runtime execution of local code by triggering the project's own testing and build pipelines.
Evidence: SKILL.md and references/repo-audit.md direct the agent to run existing project commands to validate quality improvements.
[EXTERNAL_DOWNLOADS]: Several tools used by the skill fetch data from external vulnerability databases and package registries.
Evidence: Audit tools such as scorecard, osv-scanner, and npm/pnpm audit connect to official GitHub and package registry endpoints to retrieve security information.

vet