The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill has a large attack surface for indirect prompt injection due to its core function of analyzing external, potentially untrusted data sources.
Ingestion points: The agent is instructed to ingest and analyze web pages (accessibility auditing), API responses (API testing), screenshots, and external JSON test results (evidence collection and analysis).
Boundary markers: There are no instructions or delimiters (like XML tags or triple quotes) provided to the agent to distinguish between its instructions and the data being analyzed.
Capability inventory: The skill has significant capabilities, including executing shell commands via npx, performing network requests via fetch and Python's requests, and writing detailed reports to the file system.
Sanitization: The instructions lack any guidance on sanitizing or escaping content from external sources before processing it or interpolating it into summaries.
[COMMAND_EXECUTION]: The skill instructs the agent to execute a local script named ./qa-playwright-capture.sh in the testing-evidence-collector.md and testing-reality-checker.md files. This script is not provided within the skill package, creating a dependency on an external, unverified local file.
[EXTERNAL_DOWNLOADS]: The specialist references include instructions to use npx for executing tools like @axe-core/cli and lighthouse. These are well-known, reputable tools from established organizations, but they involve runtime execution of external code.

testing