The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill audits external websites by fetching and parsing their HTML, sitemaps, and robots.txt files. This creates an attack surface where a malicious site could provide data containing instructions for the AI agent. \n
Ingestion points: The loadSite function in scripts/lib/pages.mjs fetches content from external URLs provided as arguments or discovered via crawling. \n
Boundary markers: The skill instructions do not specify the use of delimiters or 'ignore' instructions when processing fetched external data. \n
Capability inventory: The skill is authorized to modify specific project files (such as metadata, robots.txt, and sitemaps) during its 'Repair' phase and writes establishing facts to .readyvibe/context.md. \n
Sanitization: The skill uses a custom HTML parser in scripts/lib/html.mjs to extract data from the ingested content. \n- [DYNAMIC_EXECUTION]: The skill's browser automation library (scripts/lib/browser.mjs) uses dynamic import() on computed paths to load Playwright. This mechanism is intended to locate the library within the project being audited or in global node modules. \n- [COMMAND_EXECUTION]: The secret scanning utility (scripts/scan-secrets.mjs) executes the git command via execFileSync to determine which files are tracked in the repository. The target directory is determined by the --root argument provided to the script.

launch-all