The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill interacts with external APIs to fetch web data.
Evidence: Requests are sent to https://ollama.com/api/web_search and https://ollama.com/api/web_fetch using curl in scripts/ollama-search.sh and scripts/ollama-fetch.sh.
[COMMAND_EXECUTION]: The skill requires the execution of system binaries to function.
Evidence: Uses bash, curl, and jq as specified in the SKILL.md metadata and implemented in the scripts/ directory.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it processes untrusted content from the internet.
Ingestion points: Web search snippets and full page content are ingested via scripts/ollama-search.sh and scripts/ollama-fetch.sh.
Boundary markers: Absent; the skill instructions in SKILL.md advise the agent to summarize results without providing specific delimiters or warnings to ignore embedded instructions.
Capability inventory: The agent can execute shell commands (bash, curl, jq) via the OpenClaw environment.
Sanitization: The scripts use jq to parse structured JSON, but the raw text content from external websites is passed to the model without further sanitization or filtering.

ollama-search