gaia-research/gaia-skill-tree / .agents/skills/ev-scout
ev-scout
VerifiedBroad, pre-curation evidence reconnaissance for a declared Gaia skill, intake batch, upstream repository, or proposed generic capability. Run before choosing a discovery strategy or mapping topology, to give the curator independent research leads and counterevidence. Context only: never scores, grades, ratifies, ingests, or mutates the registry. Later ev-discovery and ev-pipeline phases validate/promote any selected leads.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 5.0 KB |
SKILL.md
View on GitHub---
name: ev-scout
description: >-
Broad, pre-curation evidence reconnaissance for a declared Gaia skill, intake
batch, upstream repository, or proposed generic capability. Run before choosing
a discovery strategy or mapping topology, to give the curator independent
research leads and counterevidence. Context only: never scores, grades,
ratifies, ingests, or mutates the registry. Later ev-discovery and ev-pipeline
phases validate/promote any selected leads.
version: 1.0.0
---
# ev-scout — breadth-first pre-curation context
**Run before Phase 1 of /gaia-full-pipeline**, including an already-submitted
`gaia push` batch (Strategy F). May also run standalone, before L4 mapping.
This is a lightweight read/search pass, **not** the later /ev-discovery Phase 0
nor a substitute for /ev-pipeline. Scouting is deliberately permissive;
downstream verification and /ev-adversarial-audit are deliberately skeptical.
## Inputs
Use the declared skill URL(s), upstream `SKILL.md` or repository, intake issue,
and proposed generic concepts (if available). Keep multiple skills in a batch
distinct, especially for named-only evidence. Do not require an L4-approved
generic mapping: report tentative alternatives when topology is unresolved.
## Search widely, without grading
Explore several search angles and independent source families:
1. **Generic capability:** peer-reviewed papers, preprints, evaluations,
general benchmark methodologies, foundational approaches, rival systems,
and criticism or failure modes.
2. **Named implementation:** source repo, specific `SKILL.md` blobs,
releases, tests, CI/reproduction claims, independently authored reviews,
eligible first-party or third-party `social-signal` demonstrations (videos,
walkthroughs, posts directly covering the skill; no 30-day publication hold),
external non-first-party `proxy-containment` repositories (>= 10,000 stars),
reported use, security findings, adoption or reproducible demonstrations
(see [`docs/agents/curation-authority-map.md`](../../../docs/agents/curation-authority-map.md) and #2473).
3. **Comparison:** alternative architectures and negative findings that could
falsify claimed benefits. Prefer disconfirming evidence when the first
results are uniformly promotional.
4. **Evaluation opportunities:** candidate benchmark IDs, public methodology,
datasets, harnesses and missing reproducibility details. A vendor claim is
**never** a verified `benchmark-result`.
Try broad and narrow queries; sample independent sources rather than repeatedly
searching the same host. Favor diversity of *useful leads* over an illusion of
completeness. Do not discard a relevant lead just because it may later fail.
Do not turn mere topic relevance into implementation-specific proof.
Use available search and repository-read tools. If Firecrawl is available,
follow /ev-discovery's authentication/preflight and quoting conventions.
If not, use other available read/search tools, or report search limitations.
Absence of search is not a curation blocker: return a clearly incomplete packet.
## Output: scout packet
Produce a Markdown packet in the handoff or at
`generated-output/ev-scout/<intake-or-slug>.md` when a persistent file is
needed. Include:
- target skill(s), upstream URL(s), date, searched queries/source families,
tools used or unavailable, and mapping status (tentative or ratified);
- a table of **leads**, with URL, title/publisher, publication date if known,
possible canonical Gaia evidence type, generic/named/benchmark-catalog scope,
what it might substantiate, and what a verifier must still confirm;
- a separate **counterevidence and competing approaches** section;
- likely evaluation benchmarks as **benchmark-source candidates only**,
including proposed generic applicability and missing reproducibility fields;
- a short mapping-relevant synthesis that distinguishes architecture from
demonstrated performance and marks uncertainty explicitly;
- suggested verification priorities and handoff to /ev-discovery, /ev-pipeline,
and /gaia-ingest-batch *after their normal human gates*.
Use "candidate", "possible", or "unverified" until evidence is checked.
Never fabricate publication dates, citations, percentile values, peer-review
status, scores, or external adoption. Mark unavailable facts unknown.
## Non-authority and handoff
No registry mutations, evidence-lake edits, GitHub approval labels, L4
ratification, star grades, Trust Magnitude, or automatic evidence ingestion.
This packet **informs** /gaia-draft-curate, /gaia-curate[-chain/-dynamic],
and the human L4 review; it does not override their workflows.
Later /ev-discovery can investigate selected Stage-2 leads; /ev-collection
normalizes typed candidate rows; /ev-benchmark-verification and
/ev-adversarial-audit challenge them; /ev-link-validation checks links;
human review and canonical Gaia CLI ingestion remain authoritative.
If a recent scout packet for this exact source exists, reuse and refresh it
rather than rescouting blindly. Record any new material or limitations.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deployment | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy. | Stars: 119k | Updated | Claude CodeCodex | Verified | Install |
| childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |