The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill crawls untrusted web content which could contain malicious instructions designed to influence the agent during the feature extraction or document generation phase.
Ingestion points: The skill utilizes mcp__playwright__browser_snapshot and navigation tools to read HTML and UI elements from live web applications (SKILL.md).
Boundary markers: No explicit delimiters or instructions are provided to the agent to disregard potential commands found within the extracted web text.
Capability inventory: The skill possesses access to Bash, Write, and browser_run_code tools, which could be misused if an injection is successful (SKILL.md).
Sanitization: There are no defined processes to sanitize or filter the extracted content before it is used for PRD generation.
[COMMAND_EXECUTION]: The skill uses shell commands to perform environment maintenance.
Evidence: Automatically removes browser lock files using rm -f in the $HOME/.playwright-profile/ directory during Step 0 (SKILL.md).
[EXTERNAL_DOWNLOADS]: Recommends the installation of external packages from well-known registries.
Evidence: Suggests the user install the @playwright/mcp and firecrawl-mcp packages via npx (SKILL.md).

web-to-prd