The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze content from arbitrary web URLs, creating an attack surface for indirect prompt injection.
Ingestion points: Untrusted data enters the agent context through WebFetch, the Tavily API, and agent-browser output as described in SKILL.md and rules/tool-selection.md.
Boundary markers: The instructions lack explicit boundary markers or "ignore embedded instructions" warnings when the agent is directed to process or summarize the retrieved content.
Capability inventory: The skill utilizes Bash (permitting curl, jq, and CLI tool execution) and Write (permitting snapshot and diff storage to the filesystem).
Sanitization: There is no evidence of sanitization, filtering, or validation of the external web content before it is processed by the agent.
[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute shell commands via the Bash tool to process data and perform network operations.
The skill uses curl to interact with the Tavily API and jq to parse JSON data as shown in references/tavily-api.md and rules/monitoring-change-detection.md.
It utilizes an agent-browser CLI tool to perform browser automation, including sensitive operations like session state management (agent-browser state save/load) in rules/browser-patterns.md.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to external services to retrieve information.
It makes multiple curl calls to api.tavily.com for search, extraction, and crawling. Tavily is a well-known service for AI research and data extraction, and these operations are core to the skill's purpose.

web-research-workflow