amirkiarafiei/open-scholar-peer/1-osp-summary

/1-osp-summary — Internal Compression

Extracts the paper's claims, method, and evidence into a structured summary Ŝ that downstream personas rely on.

Activation

Invoke the osp-summary-agent skill. The agent assumes the Summary Agent persona and follows that skill's protocol exactly.

Inputs — none of these is a gate

• phases.onboarding.status == "completed" in session.json. If onboarding was skipped you have no

venue and no review guidelines; extract the summary anyway, and record under NOTE and in Provenance that the venue is unknown. Mention /0-osp-onboarding once. Do not insist, and do not refuse.

The one thing that cannot be worked around

Every other input in this protocol is optional. This one is not: with no readable text of the paper there is nothing to review, so this guard stops — and it is the only place in Open ScholarPeer that does.

PDF/DOCX/TEX in .brain/input/ is not enough on its own; a binary path handed to the agent is the single biggest cause of silent downstream failure.

Before activating the skill, verify in this order:

1. `.brain/input/paper.md` exists and is readable — ideal case, proceed.
2. `.brain/input/paper.md` is missing but a PDF/DOCX exists — try to convert it now using the markitdown MCP tool (convert_to_markdown). Save the result to .brain/input/paper.md. Update session.json.paper.parsed_path.
3. `markitdown` MCP is unavailable AND only a PDF/DOCX is present — stop here, and say why in one

line: there is no readable text to summarise. Print: > Paper at .brain/input/<file> is in a binary format and markitdown MCP is not available. Either: > (a) install markitdown — uvx markitdown-mcp --help confirms it is fetchable; plain > uvx markitdown-mcp starts the server and waits, which looks like a hang — or > (b) convert the paper manually and place the markdown at .brain/input/paper.md, or > (c) if your host tool reads this format natively (Claude Code reads PDFs, for one), read it > yourself, write the extracted text to .brain/input/paper.md, and re-run.

1. Nothing in `.brain/input/` — stop here too; there is no paper. Point at /0-osp-onboarding,

which will help locate it.

Do NOT silently proceed with a PDF path on the assumption that the host tool will handle it. Even if it can, the conversion result becomes part of the audit trail and must be in .brain/input/paper.md.

Opening block (print before step 1)

Render the opening block exactly as .agents/defaults/phase_block_template.md defines it — that file holds the rail, the rules and the widths, and it is the only place they are written down. This is phase 2 of 7 (summary); read the rail's state from session.json. Values:

DOING compress the paper into claims and evidence READS .brain/input/paper.md WRITES .brain/raw/01_structured_summary.md COST ~2 min, no external calls

Steps

1. Run the hard input guard above.
2. Read .brain/session.json and .brain/input/paper.md (the validated markdown produced or verified by the guard).
3. Activate the osp-summary-agent skill — it owns the extraction protocol (claims, method, evidence) and the output format.
4. The skill writes .brain/raw/01_structured_summary.md.
5. The skill updates session.json:

• phases.summary.status = "completed"
• phases.summary.completed_at = <now>
• phases.summary.notes = "<N> claims, <M> evidence items, method identified"
• resume_from = "literature"

Closing block (print when the phase ends)

Render the closing block from .agents/defaults/phase_block_template.md. Build the rail from `session.json` — ● only where status == "completed", ○ for pending and skipped. A phase the user skipped must not show as done. Drop BLOCKED and NOTE when there is nothing to put on them. Values:

DONE <N> claims, <M> evidence items, method identified .brain/raw/01_structured_summary.md NEXT /2-osp-literature retrieve prior work

Re-run behavior

If phases.summary.status == "completed", warn the user once that re-running will overwrite, then proceed.