The Agent Skills Directory

[Indirect Prompt Injection]: The skill is designed to ingest and analyze external solution files (.zip) and browser captures (.har). While this is the intended functionality, it creates a surface where malicious instructions embedded in the analyzed data could potentially influence the agent's reasoning.
Ingestion points: Browser HAR captures and Power Platform solution ZIP files (SKILL.md, Step 1).
Boundary markers: The skill specifies an authoritative list of sources for best practices (SKILL.md, Step 0), but does not define explicit delimiters to separate user data from instructions during the reasoning phase.
Capability inventory: The skill utilizes playwright-browser_navigate, playwright-browser_file_upload, and local file system writes (SKILL.md, Step 8).
Sanitization: Output is structured and validated against a specific JSON schema (findings.schema.json), which helps maintain integrity.
[External Downloads]: The skill fetches configuration data and citation guidelines from external sources.
Context: It retrieves best-practice guidelines from an official GitHub repository (microsoft/power-cat-skills). This is used for providing grounded citations in the generated audit reports.
[Command Execution]: The skill leverages browser automation tools to assist the user in reviewing findings.
Context: Tools like playwright-browser_navigate and playwright-browser_file_upload are used to open a viewer interface and load the results of the analysis. A hard rule is included to ensure that sensitive artifacts remain local and are not uploaded to external servers.

powercat-overpage