Skip to content

feat: add opt-in public web sources to operator research - #96

Open
georgeatparallel wants to merge 1 commit into
darkzOGx:masterfrom
georgeatparallel:george/parallel-mcp-youtube-automation-agent-0bf5d832
Open

georgeatparallel wants to merge 1 commit into
darkzOGx:masterfrom
georgeatparallel:george/parallel-mcp-youtube-automation-agent-0bf5d832

Conversation

@georgeatparallel

Copy link
Copy Markdown

What & why

Autonomous research builds its source catalog from YouTube trends and competitor videos. For an explicit public topic, the native planning pass currently has no way to include an official documentation page or article in that catalog. This adds a separate Parallel MCP source feed enabled only by PARALLEL_RESEARCH_TOPIC. It supplies up to five public sources, preserves their exact URLs in the research catalog and plan citations, and lets the template fallback cite the configured topic. Public sources stay separate from popularity signals. The feed is off by default, sends only the configured topic as research input, uses anonymous access, and falls back to existing research on errors or a 20-second deadline. Setup and opt-out are documented with a link to Parallel MCP setup.

I work at Parallel. This patch was prepared with AI assistance and checked by an independent automated reviewer.

Checklist

  • One focused concern per PR (split unrelated changes into separate PRs)
  • No package-lock.json regeneration unless the PR is specifically about dependencies
  • npm run lint passes
  • npm test passes (add a test if you fixed a bug)
  • Rebased on current master

How I tested it

  • npm run lint passed; npm test passed all 50 system tests.
  • The new Optional Public Source Research system test runs native researchAndPlanChannel against a local MCP fixture. It checks the original URL, including escaped query and fragment, in research.publicSources, the deduplicated sourceCatalog, and the resulting plan. It also checks retained YouTube evidence, citation allowlisting, default-off and opt-out behavior, JSON/SSE responses, MCP session/protocol headers, the product User-Agent on HTTP traffic, absence of provider credentials, and empty/error/rate-limit/deadline fallbacks.
  • Separately exercised the adapter against live anonymous Parallel MCP, then ran native researchAndPlanChannel with live search and deterministic doubles for unrelated YouTube, database history, and AI text services. The fallback plan preserved the returned URLs, including https://nodejs.org/api/test.html, exactly from the research sources. All Parallel requests carried youtube-automation-agent/2.11.0 (public-source-feed) and no Parallel credential. This validates native research and planning; no videos were generated or published.
  • The existing Slideshow Renderer test used its browser-safe image embedding assertion because Chromium is unavailable in this environment.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant