The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute a bundled Node.js script (scripts/scan.mjs) to identify code patterns. It also directs the agent to run existing repository linter tools (e.g., ruff, ESLint, Go errcheck) using local binaries at the versions declared in the repository config as described in SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The scan.mjs script extracts snippets of code ('evidence') from the target repository and presents them to the AI agent for classification. This creates a surface where maliciously crafted comments or code strings within the repository could be interpreted as instructions by the agent.
Ingestion points: The scripts/scan.mjs script reads file content from the user-provided directory scope.
Boundary markers: The SKILL.md instructions do not specify the use of delimiters or 'ignore embedded instructions' warnings when the script output is processed by the agent.
Capability inventory: The agent has the capability to write files (implied in SKILL.md workflow) and execute local commands and scripts (SKILL.md workflow steps 2 and 3).
Sanitization: The 'evidence' text extracted from source files is passed to the agent's context without escaping or sanitization.

demock