The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests content from arbitrary external URLs and processes this data through multiple analysis agents without employing boundary markers or sanitization. This establishes an attack surface where malicious content on a target website could attempt to influence the agent's behavior, file writes, or generated reports.
Ingestion points: The steps/01-flag-detection.md file implements a browser cascade to fetch content from user-specified URLs.
Boundary markers: No specific delimiters or safety instructions are defined to separate the untrusted content fetched from the URL from the agent's internal instructions.
Capability inventory: The skill is capable of spawning sub-agents via the Task tool, writing multiple files to the local file system (e.g., reports in Step 2, Step 4, and Step 6), and executing shell commands (Step 7).
Sanitization: There is no evidence of filtering, escaping, or validation logic for the content retrieved from external URLs before it is passed to sub-agents.
[COMMAND_EXECUTION]: The skill instructs the agent to execute a shell command (aqe memory store) in steps/07-learning-persistence.md that incorporates dynamic variables, creating a potential command injection vector.
Evidence: The command aqe memory store --key "qcsd-ideation-${epicId}-${Date.now()}" performs shell interpolation. If the epicId variable is populated with unsanitized data extracted from the user-provided epic or URL content, it could lead to unauthorized command execution.
[EXTERNAL_DOWNLOADS]: The skill implements a sophisticated "5-tier browser cascade" (Vibium, agent-browser, Playwright+Stealth, WebFetch, and WebSearch) to retrieve live content from external websites. While this is the intended functionality, it facilitates the ingestion of untrusted data.
[DYNAMIC_EXECUTION]: The skill dynamically generates and executes a shell command for state persistence in steps/07-learning-persistence.md. It also relies on local scripts such as scripts/fetch-content.js to handle the complexity of the browser-based content retrieval.

qcsd-ideation-swarm