| name | haipipe-discovery-review |
| description | Review type specialist for the discovery layer: analyze across sources — judge a claim (prior_art_check / counterevidence -> verdict.md) or map a field (landscape_review / benchmark_landscape -> landscape.md). Dispatches research-lit / comm-lit-review / academic-researcher; owns the Review Output Contract. Trigger: judge claim, prior art, counterevidence, landscape, lit review for a discovery, /haipipe-discovery-review. |
| argument-hint | [<discovery-folder> | "<claim-or-topic>"] |
| allowed-tools | Bash, Read, Write, Edit, Grep, Glob, Skill |
| metadata | {"version":"0.1.0","last_updated":"2026-07-03","summary":"Type specialist owning Review: judge -> verdict.md, synthesize -> landscape.md. Home of the Review Output Contract."} |
Skill: haipipe-discovery-review (type specialist)
Owns the Review type: the Execute stage of a Review discovery-folder, or a one-off analysis. role: picks the flavor — judge (prior_art_check, counterevidence -> verdict.md) or synthesize (landscape_review, benchmark_landscape -> landscape.md).
Workers: research-lit (default, multi-source), comm-lit-review (communications domain), academic-researcher (cross-discipline template).
Procedure (Execute of a Review folder)
- Read
discovery.yaml: question, role, and sources. If sources.from_source_folder names a Search folder, read ITS sources.md/notes.md instead of searching; otherwise search + read inline (writing sources.md + notes.md as work products).
- Dispatch a review worker with the Review Output Contract (below) appended to its args.
- Judge roles: weigh evidence for/against, write
verdict.md (status + confidence + Answer + Evidence + Caveats). Synthesize roles: cluster the field, write landscape.md (taxonomy + gaps + references). Templates: haipipe-discovery/ref/discovery-yaml-schema.md.
- Counter-evidence is reported, never smoothed over; a verdict never overstates what the sources say.
- Return to the caller: the terminal path, the verdict status (or cluster/gap counts for a landscape), and the NEEDS-VERIFICATION count. discovery.yaml itself is the orchestrator's to write, not this skill's.
One-off calls (no folder): same analysis, returned INLINE in the terminal's format — write no files.
Review Output Contract (canonical home)
Append this to the args of any review-worker dispatch. Short author-year tags alone are NOT matchable by the reader. Every analysis MUST satisfy all five rules.
1. FULL NAMES, never bare tags. Each paper gets a FULL citation on first mention AND a line in
the reference list: full title (verbatim), full author list (first three + et al. only when
>= 6 authors), venue with vol(issue):pages or "arXiv preprint"/status, year + a LOCATOR
(arXiv ID / DOI / URL). A short tag in prose is OK only if it maps to exactly one full entry.
2. NUMBERED REFERENCE LIST at the end ("References (full, verified)"): one self-contained,
deduped line per paper; any in-text mention matches exactly one numbered entry.
3. DISAMBIGUATE COLLISIONS. Same author+year papers each get a distinguishing nickname in EVERY
mention (Lu et al. 2025a vs Lu et al. 2025b). Never leave two papers both as "Lu 2025".
4. ONE-LINE PLAIN FINDING per paper (jargon-free).
5. VERIFICATION FLAG per paper: VERIFIED (id/DOI/venue confirmed via search) vs NEEDS-VERIFICATION.
Never assert an unchecked citation; fabrication is the worst failure.
Any per-paper listing follows the one-paper-one-subsection rule (full title in the heading, never a table; canonical format: haipipe-discovery/ref/source-format.md). For deeper rigor (systematic search + adversarial citation verification), escalate to the deep-research lit-review pipeline (Tools/plugin-workflows/academic-research-skills/deep-research, mode lit-review).
This skill only executes; the folder lifecycle (Plan/Report) belongs to /haipipe-discovery.
