The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill manages sensitive authentication data including bearer tokens and basic auth credentials (username/password).
Evidence: The scripts/search.py script and references/configuration.md documentation describe reading auth.token, auth.user, and auth.pass from a local TOML/JSON configuration file.
Evidence: The script resolves these values either directly from the file or via environment variables, then includes them in the Authorization header of outgoing HTTP requests.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes web search results, which are externally controlled and potentially adversarial.
Ingestion points: The scripts/search.py script fetches data from a SearXNG API and parses the JSON response, specifically extracting title, url, content, and infobox data in the format_results function.
Boundary markers: There are no explicit delimiters or boundary markers used in the prompt workflow to separate the untrusted search results from the agent's core instructions.
Capability inventory: The agent has the capability to execute shell commands (Bash(python3 scripts/*)), make network requests to a user-defined base_url, and read local configuration files.
Sanitization: The skill performs no sanitization or filtering of the search result content before presenting it to the agent, creating a risk that malicious instructions embedded in web content could influence the agent's subsequent actions.

searxng-search