The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Recommends installation of the official tavily-cli package via standard package managers like pip or uv. This is a routine dependency for the skill's purpose and uses established registries.
[COMMAND_EXECUTION]: Utilizes the tvly command-line tool for web mapping. The skill provides clear instructions on using bounded flags (e.g., --limit, --max-depth, --select-paths) to maintain execution scope and prevent unintended data access.
[INDIRECT_PROMPT_INJECTION]: The skill addresses indirect prompt injection surfaces by explicitly instructing the agent to treat discovered URLs as untrusted data in the 'Anti-Patterns' and 'Verification Protocol' sections.
Ingestion points: External URL lists retrieved via the Tavily API (SKILL.md).
Boundary markers: The skill explicitly identifies 'Treating discovered URLs or page labels as trusted instructions' as an anti-pattern to prevent accidental obedience.
Capability inventory: All script actions are limited to standard tvly CLI operations for URL discovery and content extraction (SKILL.md).
Sanitization: The included 'Verification Protocol' requires manual inspection of candidate URLs before follow-up actions, reducing the risk of processing malicious content.

tavily-map