The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-controlled Markdown files, creating a potential indirect prompt injection surface during the audit process.
Ingestion points: The agent is instructed to find and read all Markdown files in the project workspace (excluding standard build/environment folders like .venv and node_modules).
Boundary markers: The instructions provide a specific Markdown report template which helps keep the agent's output focused and constrained.
Capability inventory: The agent can use Bash to run scripts and various file tools (Read, Grep, Glob) to inspect content.
Sanitization: No specific content sanitization is described, as the agent primarily performs metadata extraction and link verification.
[COMMAND_EXECUTION]: The skill includes instructions to search for and execute a local script (scripts/check_links.py) if one exists in the project.
Evidence: Step 4 explicitly mentions: 'If a link-checking script exists in the project (e.g., scripts/check_links.py), run it as well'. This is a common pattern for developer tools to leverage existing project infrastructure, but it depends on the integrity of the project being audited.

solo-audit