The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted "Generated Source Code Files" and "Sprint Tickets", creating a surface for indirect prompt injection.
Ingestion points: Audits external source code and acceptance criteria.
Boundary markers: Absent; no delimiters are used to separate user data from instructions.
Capability inventory: High-power capabilities including browser automation (computer-use, chrome-devtools-mcp) and CLI tool execution (graphify).
Sanitization: Absent; no content validation is performed.
[COMMAND_EXECUTION]: The skill is instructed to run command-line tools for auditing.
Evidence: Executes graphify query and graphify path.
[EXTERNAL_DOWNLOADS]: The skill references external resources for automation fallbacks.
Evidence: Mentions stablyai/orca@computer-use, midscene-skills@computer-automation, and codex-skills@gemini-computer-use as external skills for automation.

developer-qa-reviewer