The Agent Skills Directory

[PROMPT_INJECTION]: The skill utilizes the $ARGUMENTS variable to allow user-supplied input to define the research target, which serves as a potential vector for prompt injection. Additionally, the skill processes untrusted content from the local codebase (e.g., README files, landing page copy) to identify features, creating an indirect prompt injection surface where embedded instructions in those files could influence the analysis. (Ingestion points: README, package metadata, and codebase files; Boundary markers: None; Capability inventory: Web search and file writing; Sanitization: None).
[EXTERNAL_DOWNLOADS]: To perform its research, the skill fetches competitor marketing pages and pricing details from external URLs. It also consults well-known technology profiling services like BuiltWith and Wappalyzer to analyze competitor technology stacks.
[COMMAND_EXECUTION]: The skill performs file system operations including reading project configuration files (e.g., package.json, Cargo.toml) and source code to build a product profile. It also writes its final analysis report to the local file system at docs/competitive-gap-analysis.md and appends telemetry data to a file in the ~/.claude/projects/ directory.
[DATA_EXFILTRATION]: The skill derives search queries (such as the product's domain and feature set) from the local codebase to identify competitors on the web. This behavior is inherent to the tool's intended competitive analysis purpose and does not represent an unauthorized transfer of source code.

compete