open-knowledge-pack-knowledge-base-research

What it does

Conduct evidence-driven research on a topic and produce a provisional research article under research/ in a Knowledge Base project. The article is provisional, not canonical, and should capture findings, trade-offs, and open questions at a point in time. It preserves sources verbatim via an ingest workflow and writes the article incrementally so crashes do not lose work. It always cites every claim and links findings back into the graph. It does not promote findings to canonical knowledge on its own.

How it works

• Step 0: Create workflow checkpoint tasks via TaskCreate for each stage (e.g., scan coverage, collaborative scoping, ingest, read/analyze, write article, link Q&A, validate, recap).
• Step 1: Scan existing coverage in the content directory using commands like grep and ls to determine prior work and route. Classify coverage as Fully covered, Partially covered, or Not covered and present routing options accordingly.
• Step 2: Collaborative scoping (Supervised). Propose a rubric structure including the narrowed question, dimensions, candidate sources, success criteria, and output format. In Supervised mode, wait for user confirmation before external fetch; Headless mode auto-confirms and proceeds.
• Step 3: Capture raw sources via ingest for each relevant URL/document, ensuring sources are preserved before analysis.
• Step 4: Read + analyze ingested sources plus existing canonical articles, prior research, relevant code if requested, and project context. Take structured notes including key claims with evidence, trade-offs, contradictions, unknowns, and relevance. Write notes into the article as you go and create the article skeleton before deep reading.
• Step 5: Write the research article (Path A). If the article already exists from Step 4, finalize and tighten it; if starting anew after an interruption, load partial content and fill missing sections. Save the markdown document under the project content directory following the repository’s conventions.
• Step 6–8: Link aggressively, file Q&A back, validate frontmatter and dead links, and provide a recap with follow-up directions.

When to use it

Use Path A (default) to create a persistent provisional article. Use Path B only when the user explicitly requests a direct answer (no doc). Path C is for updating an existing article.

What it can touch

This skill relies on external sources and the knowledge-base project structure. It uses commands and ingest procedures defined by the /open-knowledge platform rules and the project’s ingest workflow. It requires the OpenKnowledge MCP server and skills/tools: Claude Code, Claude Desktop, Claude Cowork, Claude.ai web.

Caveats

Provisional articles are not canonical and require the /consolidate-notes skill to promote findings. All claims must be cited with sources captured via ingest, and the process enforces crash-safety by persisting work incrementally. If a fetch fails, the agent should pause and ask the user to supply the missing source instead of fabricating evidence.