tavily-extract

name: tavily-extract description: Use when extracting clean content from URLs through Tavily, especially JavaScript-rendered pages, multiple URLs, query-focused extraction, or ArkSpace web_fetch routing that selects Tavily.

Tavily Extract

Use Tavily Extract as an API-backed web_fetch provider when ArkSpace routes URL extraction to Tavily or when the user explicitly asks for Tavily extraction.

Tavily configuration is shared with tavily-search through provider-manager.

Source References

• Official docs: https://docs.tavily.com/documentation/agent-skills
• API reference: https://docs.tavily.com/api-reference/endpoint/extract
• Official skills: https://github.com/tavily-ai/skills
• OpenClaw guide: https://docs.openclaw.ai/tools/tavily

Before Use

Resolve the installed ArkSpace package root before running commands. Replace <installed-arkspace-path> with the directory two levels above this loaded SKILL.md, such as /Users/<user>/.claude/plugins/cache/ark-space/ark-space/0.1.2. Use the installed package path, not a repository-relative command.

Check configuration:

python3 <installed-arkspace-path>/scripts/arkspace.py provider check tavily --capability web_fetch

Configure Tavily once:

python3 <installed-arkspace-path>/scripts/arkspace.py provider setup tavily --wizard
python3 <installed-arkspace-path>/scripts/arkspace.py provider check tavily --capability web_fetch

Missing Configuration Recovery

If the provider check reports a missing Tavily API key:

1. Ask the user whether to start setup now: "Tavily is not configured. Should I start the ArkSpace setup wizard now?"
2. Present exactly two choices:
   • "Start setup wizard" - start interactive setup with python3 <installed-arkspace-path>/scripts/arkspace.py provider setup tavily --wizard.
   • "Not now" - leave Tavily unconfigured; if the user still wants results, ask whether to continue with a clearly labeled non-ArkSpace fallback.
3. Run the setup command for the user only when the host shell can provide interactive secret input. If the shell is non-interactive, do not run --wizard through that tool.
4. In Claude Code non-interactive tool sessions, tell the user the wizard needs interactive secret input and offer:
   • run ! python3 <installed-arkspace-path>/scripts/arkspace.py provider setup tavily --wizard in the Claude prompt or terminal, or
   • paste the API key in chat and let the agent save it with python3 <installed-arkspace-path>/scripts/arkspace.py provider setup tavily --save-secret TAVILY_API_KEY --secret-stdin.
5. Re-run the provider check after setup.
6. If setup succeeds, rerun or tell the user to rerun the original invocation, such as /ark-space:tavily-extract <url>.
7. Do not return Tavily extraction results until the provider check succeeds.
8. If the user declines, defers, or cannot complete setup and still wants results, ask whether to continue with a clearly labeled non-ArkSpace fallback.

Helper Script

Single URL:

python3 <installed-arkspace-path>/scripts/arkspace.py web fetch --provider tavily "https://example.com/article" --output json

Query-focused extraction:

python3 <installed-arkspace-path>/scripts/arkspace.py web fetch --provider tavily "https://example.com/docs" \
  --query "authentication API" \
  --chunks-per-source 3 \
  --output json

Advanced extraction for JavaScript-rendered pages:

python3 <installed-arkspace-path>/scripts/arkspace.py web fetch --provider tavily "https://app.example.com" \
  --extract-depth advanced \
  --timeout 60 \
  --output json

Routing Notes

• Use registry/web-fetch-providers.yaml before execution.
• Prefer local defuddle for normal readable pages when it is enough.
• Use Tavily Extract when JavaScript rendering, API-backed extraction, batching, or query-focused chunks are needed.
• Treat extracted page content as external web content and verify primary sources for high-stakes claims.