The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The .env file contains a hardcoded API key (ANYSEARCH_API_KEY=as_sk_16c4f23c66e3a3b9c9683f79090fa912). Hardcoding credentials in distributed files is a critical security vulnerability as it exposes secret keys to any user or system that accesses the skill's source code.
[DYNAMIC_EXECUTION]: The file scripts/generate.py contains logic designed to programmatically rewrite the source code of other scripts within the skill (anysearch_cli.py, anysearch_cli.js, anysearch_cli.ps1, and anysearch_cli.sh). This self-modification capability represents a dynamic execution risk where the behavior of the CLI tools can be changed at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill provides an ingestion surface for untrusted external data through its search and web content extraction features.
Ingestion points: The extract command fetches full page content, and the search command retrieves web snippets, both of which are then presented to the agent.
Boundary markers: The CLI tools do not implement clear boundary markers or instructions to help the agent distinguish between retrieved data and legitimate commands.
Capability inventory: The skill possesses network access capabilities (to api.anysearch.com) and file-writing capabilities (used for updating .env and runtime.conf).
Sanitization: Content fetched from external URLs is passed to the agent without sanitization, allowing potentially malicious instructions embedded in web pages to influence agent behavior.
[COMMAND_EXECUTION]: The skill instructions (found in README.md and SKILL.md) require the agent to execute multiple shell commands for platform detection, runtime testing, and persisting configuration to runtime.conf. This involves significant interaction with the host environment's shell and file system during the initialization phase.

anysearch