| name | scout |
| description | Discover great-but-low-visibility ideas — the fringe excellence mainstream search buries under popularity. Given a seed (a name, repo, post, or topic), traverse its citation/collaborator graph + the periphery sources, score each candidate on depth × novelty × practitioner-grounding MINUS a popularity penalty, dedup against the idea-board + installed plugins, and emit a ranked shortlist that routes the best finds into /forge. Use when you spot a promising fringe idea (often on LinkedIn / X / GitHub) and want the cluster around it, or to periodically sweep a niche for under-the-radar excellence. |
Skill: scout — the fringe-idea scout
Why this exists
Mainstream search ranks on popularity (SEO, backlinks, stars, virality) — which is orthogonal to quality. Great-but-low-visibility ideas are invisible to it by construction: a single-maintainer "WIP" GitHub repo, a niche practitioner's blog, a LinkedIn post from someone who actually ships. The canonical example this skill was extracted from: Kurt Buhler's data-goblin/power-bi-agentic-development — genuinely excellent, low stars, surfaced on LinkedIn, missed by every "best Power BI tools" listicle. Scout inverts the ranking: seek HIGH depth + LOW reach.
The score (decoupled from popularity)
great-but-obscure ≈ depth × novelty × practitioner-grounding − popularity_penalty
| Factor | Up-rank |
|---|
| depth | technical specificity, a real mechanism, production detail — not a listicle |
| novelty | diverges from the mainstream consensus; a non-obvious angle |
| practitioner-grounding | born of real shipped work, not theory or marketing |
| popularity_penalty | DOWN-rank the already-famous — a 10k-star repo, an official vendor doc, a conference-keynote name is not fringe. The whole point is the under-the-radar do-er. |
Rank high-depth + low-reach to the top. A high-depth + high-reach item is good but not this skill's job; a low-depth + low-reach item is just obscure, not a gem.
Method (5 steps, in order)
- Seed. Take the input — a name / repo / URL / LinkedIn post / topic. If it's a bare topic, find 1–2 known-great seeds first (a respected practitioner or a deep repo), then proceed.
- Traverse the graph — highest yield. From the seed, follow the edges: who it cites, links to, is built on, collaborates with, and who cites it. Fringe excellence clusters — the neighbours of a great fringe source are disproportionately great-fringe. This beats keyword search by a wide margin; do it first.
- Sweep the periphery (source map below) — GitHub by recently-updated, not stars; niche blogs/newsletters; LinkedIn/X practitioner posts; Hacker News "new"; niche subs/Discords; conference abstracts; arXiv.
- Score + dedup. Apply the rubric. Drop anything RavenClaude already has (check
../../../../docs/idea-board.md + the installed plugins) and anything plainly mainstream.
- Emit + persist + route. A ranked shortlist — each find: what · why-great · why-invisible · source URL · RavenClaude-fit. Then, in order:
- Persist the full run — don't let it die in the chat. Write the complete report to
docs/research/<YYYY-MM-DD>-scout-<slug>/report.md (<slug> = a short kebab of the seed; the same committed research-persistence home rc-deep-research uses, e.g. the DAX report — docs/ commits straight to main, no PR). The report captures what the distilled idea-board row can't: the full ranked shortlist with per-find reasoning, the dropped-and-why and ToS-flagged items, the per-lane / per-source detail, the load-bearing finding(s), and the Structured Output block. This is the durable, fuller record a later dedup or rc-deep-research pass actually reads — the idea-board points at it.
- Append the keepers to the committed curated index. Add each keeper as a row to
docs/idea-board.md (find · what · why-fringe · RC-fit · status), and have the run-section header link to the report from the bullet above. Two tiers, both committed: the idea-board is the distilled, shared index; the report is the full detail.
- Hand the top find(s) to to deepen each keeper — fan-out across the source's wider footprint, fetch + adversarially verify the "why-great" claim (a fringe gem's depth must survive scrutiny, not just look deep), and surface the maturity / adoption / risk signal a one-pass scout can't. The research report : decides whether (and how) to build, now grounded in a verified brief rather than a single-source impression (the same seed→plan path Buhler took, with the verification step added).
Periphery source map (look here, not Google-top)
| Source | How to query for fringe |
|---|
| GitHub | search_repositories / search_code sorted by updated, filtered by topic; favour single-maintainer / "WIP" / low-star repos — the star-bias bypass |
| Practitioner blogs | the author's own site + their blogroll / links section |
| LinkedIn / X | the practitioner's posts + who they engage with |
| Hacker News | "new" + a niche keyword (never "top") |
| Newsletters / Substack | niche-specific, practitioner-authored |
| Conference abstracts | niche conferences' schedules (not the keynotes) |
| arXiv / preprints | recent + niche |
| Reddit / Discord | niche practitioner communities |
Tools
- WebSearch with
allowed_domains to target GitHub + niche sources; recency-tuned queries.
- The GitHub MCP (
search_repositories / search_code) sorted by updated — the deterministic star-bias bypass (load it via tool-search first; it's deferred).
- The
rc-deep-research fan-out harness, RE-TUNED — from verify-a-claim-across-sources to discover-the-cluster-around-a-seed (breadth over a niche, scored for great-but-obscure). Reuse its parallel-fetch shape; swap the verification grader for the rubric above.
Anti-patterns
- Ranking by stars / views / followers — that's the popularity bias scout exists to defeat.
- Surfacing the already-famous — down-rank it; if it's on the front page it isn't fringe.
- Inventing obscure-sounding names to look thorough — every find MUST carry a real, fetched URL. An unverifiable "fringe gem" is a hallucination — the exact failure this skill must avoid (claim-grounding applies: cite the fetched source or drop the find).
- Skipping the dedup — re-surfacing what RavenClaude already has wastes a slot.
- Stopping at keyword search — if you didn't traverse the seed's graph, you skipped the highest-yield move.
- Letting the run die in the chat transcript — a scout whose detail was never written to
docs/research/<date>-scout-<slug>/report.md (and whose keepers never reached docs/idea-board.md) can't be deduped against, handed to rc-deep-research, or audited later. The distilled idea-board row is not the full record. Persist both before you route.
Composition
Scout is the front door of the idea pipeline; rc-deep-research is the deepening pass; FORGE is the gate: scout finds great-but-invisible ideas, rc-deep-research verifies and enriches the keepers, and /forge decides whether (and how) to build them. The flow is /scout <seed> → ranked shortlist on the idea-board → rc-deep-research on the top find(s) → verified brief → /forge <top find> → routed plan. Buhler is the worked example: spotted on LinkedIn → scout-traversed → researched → /forge'd into the pixel-perfect-reporting plan. Research sits between scout and forge so the build decision is made on a verified brief, never on a single fetched source — and so a fringe find whose depth doesn't survive scrutiny is dropped before it consumes a forge slot.
Output Contract
Three persisted artifacts, then the handoff:
- The full run report at
docs/research/<YYYY-MM-DD>-scout-<slug>/report.md — the ranked shortlist with per-find reasoning, the dropped-and-why + ToS-flagged items, the per-lane/per-source detail, and the load-bearing finding(s). Committed to main (docs, no PR).
- The distilled keepers appended as rows to
docs/idea-board.md (find · what · why-fringe · RC-fit · status), the run-section header linking to artifact 1. Committed to main.
- The top find(s) handed to
rc-deep-research, whose verified brief then routes to /forge.
End the handoff with the cross-plugin Structured Output JSON block per ../structured-output/SKILL.md ({status, summary, deliverables, handoff_recommendation, confidence, next_actions}) — list the report path and the idea-board update in deliverables, and set handoff_recommendation.to_specialist to rc-deep-research (the next stage), not /forge.