Kartik-Doye/Ravengard-AI-Recruiter / .agents/skills/scout
scout
VerifiedResearches version-sensitive technical questions using repository truth, installed sources, official documentation, standards, and upstream evidence. Use when an API, library, platform, security rule, or current implementation detail must be verified rather than recalled.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 10.7 KB |
| references/research-protocol.md | markdown | 2.4 KB |
| references/research.md | markdown | 740 B |
| references/source-ladder.md | markdown | 4.5 KB |
| scripts/check-note | other | 1.4 KB |
references/source-ladder.md
View on GitHub# Source ladder: trust tiers, citation format, note template
## Trust tiers
| Tier | Source | Use for | Notes |
| --- | --- | --- | --- |
| 0 | This repo (lockfile, vendored code) | Installed versions, actual behavior | Always check first. Prevents API version mismatch. |
| 1 | Model Context Protocol (MCP) servers | Live API specs, enterprise knowledge | Top-tier structured grounding when available. |
| 2 | Official docs for the exact version | API shape, config, defaults | Pin the doc URL to the version path when the docs offer it. Fetch the exact sub-page, not the monolithic index. |
| 2 | Official changelog / migration guide | Breaking changes, upgrade order | Authoritative for "what changed". |
| 3 | Package source on GitHub (tags/releases) | Real signatures, real defaults | When docs are vague, code isn't. |
| 4 | RFC / spec text | Protocol behavior | For HTTP, OAuth, JWT, CORS-level questions. |
| 5 | Vendor engineering blog | Rationale, roadmap history | Good context, weak for exact API. |
| 6 | Upstream issue tracker | Known bugs, feature gaps, timelines | Note issue state (open/closed) and date. |
| 7 | Reputable community (SO, established blogs) | Practical gotchas | Only when tiers 0–6 don't answer; always labeled. |
Anti-patterns (never load-bearing): AI-generated content farms, un-dated
tutorials, "top 10 libraries" listicles, answers for a different major
version without saying so. General web knowledge is the lowest tier and
must not override official documentation or vendored code.
## Citation format
Inline links with version + date context:
```markdown
- `useActionState` requires React 19. [React docs: 19.x](https://…)
- Rate limit is 100 req/min on the free tier per 2026-06 pricing page. [source](https://…) (verify before launch)
```
Offline mode cites file paths instead of URLs:
`node_modules/foo/CHANGELOG.md @ 4.2.0`.
## The note template
```markdown
<!-- generated by scout on YYYY-MM-DD -->
# <topic-slug>
**Question:** one sentence.
**Decision it informs:** what the user will do differently based on this.
**Versions:** repo=X@1.2.3 · stable=X@2.0.0 · researched=YYYY-MM-DD
**Mode:** web | offline (sources in this repo only)
## Recommendation
The answer, in 1–3 sentences, decidable.
## Load-bearing facts
- fact [source]: each one would change the recommendation if wrong.
## Incompatible / outdated patterns
- pattern that looks right but is deprecated/renamed/unsafe in the
pinned version.
## Uncertainty
- open questions, contradictions between sources, freshness risks.
- next step that would resolve each (experiment, issue to watch, doc to
wait for).
## Confidence: high | medium | low
One line why.
```
## Conflict resolution (when sources disagree)
1. **Recency × authority:** official docs for the pinned version beat
any blog, however good. A 2024 blog vs 2026 official docs → docs win;
a 2026 blog vs 2024 docs → check the docs' version selector, then
treat the blog as a signal to verify, not as truth.
2. **Date-check advice before using it:** StackOverflow answers and
tutorials carry their era in their code style (old APIs, removed
flags). If an answer predates the current major version, verify the
API still exists in tier 0/2 before repeating it.
3. **Unresolvable conflicts get recorded, not averaged:** state both
claims, their sources and dates, mark confidence low, name the
experiment that would resolve it.
## API stability check
Before recommending any API, classify it: **stable** (documented in the
current major's docs), **experimental** (flagged, behind a flag, or
major-version-0), or **deprecated** (marked in changelog/docs, or
absent from current docs while present in older ones). Experimental
APIs get recommended only with that label and an escape hatch note;
deprecated APIs are named as "works today, removal pending": never as
the recommended path.
## Alternatives comparison (when the question is "which X")
Compare on explicit criteria, not vibes: fit to the actual requirement,
maintenance health (recent releases, issue responsiveness), ecosystem
fit with what the repo already uses, migration cost from the current
choice, and exit cost. Compare only the credible candidates that could win;
an exhaustive list is often a decision dodge. State the decision rule you applied
("picked the one with the lowest exit cost because the requirement is
likely to change").
## Freshness rules
- Reverify when the source version, target version, scope, or ecosystem churn
makes a load-bearing claim plausibly stale. Record when you checked.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |