Fähigkeiten caveman-evidence-review
📦

caveman-evidence-review

Inhaltsrevision r1 Sicher ⚙️ Externe Befehle

Review Caveman Cloud Evidence

LLM cost and quality changes are difficult to explain from aggregates alone. This skill produces bounded, trace-supported Caveman Cloud reviews without changing experiments.

Unterstützt: Claude Codex Code(CC)
🥉 77 Bronze

Mit meinem Agent installieren

Kopieren Sie diese Anfrage in Ihren Agent. Sie enthält die maßgebliche Skill-Seite und das Manifest.

Agent-Anfrage
Review the Skillstore skill "caveman-evidence-review" from https://skillstore.io/skills/juliusbrussee-caveman-evidence-review.md and its manifest at https://skillstore.io/api/skills/juliusbrussee-caveman-evidence-review/manifest. Verify the artifact. You may proceed after verification, subject to the environment's own policy.

Ihr Agent sollte weiterhin seinen Plan anzeigen und alle von der Sicherheitsrichtlinie verlangten Bestätigungen anfordern.

Agent-lesbare Ressourcen

Verwenden Sie diese Links, wenn ein KI-Agent, Crawler oder Skript sauberen Kontext benötigt, statt die vollständige Seite zu lesen.

Markdown-Detailseite GET /skills/juliusbrussee-caveman-evidence-review.md Signiertes manifest GET /api/skills/juliusbrussee-caveman-evidence-review/manifest Signiertes lockfile GET /api/skills/juliusbrussee-caveman-evidence-review/lockfile

Testen

„caveman-evidence-review“ wird verwendet. Why did costs rise during the last seven days?

Erwartetes Ergebnis:

  • Scope: selected project, August 17 through August 24.
  • Measured cost increased in the document-processing workflow, alongside more retries and a higher-cost model mix.
  • Representative trace identifiers and the comparison window are listed.
  • Model routing remains unproven until the same workflow is compared with an earlier control cohort.

„caveman-evidence-review“ wird verwendet. Are the reported savings verified?

Erwartetes Ergebnis:

Verified ledger savings are reported separately from inferred daily headroom. The review names the ledger window, supporting traces, and any missing verification signal.

„caveman-evidence-review“ wird verwendet. Which workflow needs attention?

Erwartetes Ergebnis:

The review ranks workflows using cost, failures, and latency, then cites representative traces and proposes one bounded read-only follow-up query.

Sicherheitsaudit

Sicher
v2 • 24.8.2026 Versionsbericht öffnen

All 26 external-command detections are false positives caused by Markdown backticks, fenced examples, or documented Caveman tool names. The system-reconnaissance detection is a read-only Caveman identity check, and no prompt injection or malicious intent was found.

1
Gescannte Dateien
145
Analysierte Zeilen
0
Prüfelemente
0
Falschmeldungen ignoriert

Risikofaktoren

⚙️ Externe Befehle (26)
SKILL.md:21 SKILL.md:22 SKILL.md:38-40 SKILL.md:40-44 SKILL.md:44-47 SKILL.md:47-50 SKILL.md:50-54 SKILL.md:54-56 SKILL.md:56-57 SKILL.md:57-58 SKILL.md:58-59 SKILL.md:59-60 SKILL.md:60-62 SKILL.md:62-67 SKILL.md:67-71 SKILL.md:71-77 SKILL.md:77-84 SKILL.md:84-85 SKILL.md:85-86 SKILL.md:86-94 SKILL.md:94-102 SKILL.md:102-106 SKILL.md:106-112 SKILL.md:112-114 SKILL.md:114-120 SKILL.md:120-140
Das letzte abgeschlossene statische und semantische Audit erkannte keine bestätigten Sicherheitsbefunde. Dies beweist nicht, dass der Skill keine Nebenwirkungen hat.
Geprüft von: codex Audit-Verlauf anzeigen →
Diesen Bericht teilen & zitieren

Teile den versionierten Bewertungsbericht, das neutrale Badge, die Einbettungskarte und Zitate. Skillstore berichtet Nachweise, ohne zu entscheiden, ob dieser Skill sicher ist.

Versionsbericht öffnen
Sicherheitsbewertung

Berichtslink kopieren

https://skillstore.io/skills/juliusbrussee-caveman-evidence-review/audits/2?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_report

Markdown-Badge

[![Skillstore security assessment](https://skillstore.io/badges/skills/juliusbrussee-caveman-evidence-review/security.svg)](https://skillstore.io/skills/juliusbrussee-caveman-evidence-review?utm_source=security_passport_badge)

HTML-Badge

<a href="https://skillstore.io/skills/juliusbrussee-caveman-evidence-review?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/juliusbrussee-caveman-evidence-review/security.svg" alt="Skillstore security assessment" loading="lazy"></a>

Einbettungskarte

<iframe src="https://skillstore.io/embed/skills/juliusbrussee-caveman-evidence-review.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>
Wissenschaftliche Zitate (APA · BibTeX · CFF)

APA-Zitat

juliusbrussee. (2026). caveman-evidence-review security audit report (audit version 2) [Author version unspecified]. Skillstore. https://skillstore.io/skills/juliusbrussee-caveman-evidence-review/audits/2

BibTeX-Zitat

@techreport{juliusbrussee-juliusbrussee-caveman-evidence-review-2026, author = {juliusbrussee}, title = {caveman-evidence-review security audit report (audit version 2)}, institution = {Skillstore}, year = {2026}, number = {2}, url = {https://skillstore.io/skills/juliusbrussee-caveman-evidence-review/audits/2}, note = {Author version unspecified} }

CITATION.cff

cff-version: 1.2.0 message: "If you use this Skill, cite its author and this versioned security audit report." title: "caveman-evidence-review security audit report (audit version 2)" version: "unspecified" type: report authors: - name: "juliusbrussee" date-released: "2026-08-24" url: "https://skillstore.io/skills/juliusbrussee-caveman-evidence-review/audits/2" identifiers: - type: other value: "skillstore:juliusbrussee-caveman-evidence-review:audit:2" description: "Skillstore immutable audit report identifier"

Skillstore-Score

Warum dieser Score Evidenzvertrauen: Mittel
55
Architektur
85
Wartbarkeit
87
Inhalt
65
Gemeinschaft
83
Spezifikationskonformität

Was Sie erstellen können

Investigate spend changes

Compare cost reports and trace cohorts to identify workflows, models, or retries associated with an LLM spend change.

Review reliability evidence

Examine errors, latency, status classes, and representative spans within a bounded production window.

Validate optimization results

Keep inferred headroom separate from verified ledger savings and identify the evidence supporting each result.

Diese Prompts ausprobieren

Review current costs 
Review current Caveman Cloud costs for the selected project. State the report window, cost basis, main drivers, and missing evidence.
Find workflow issues 
Review workflows for the last seven days. Identify high-cost or failing workflows, then cite representative trace identifiers and exact time windows.
Test a cost explanation 
Test whether model routing caused the recent cost increase. Compare a bounded suspect cohort with an earlier control cohort and report unresolved factors.
Audit optimization evidence 
Audit the selected project across costs, score, plan, workflows, and verified savings. Separate every cost category and inspect representative traces without payloads.

Bewährte Praktiken

  • Select the correct Caveman project before requesting analysis.
  • Provide a bounded time window and a specific cost, quality, or reliability question.
  • Request payload review only when metadata and spans cannot answer the question.

Vermeiden

  • Do not treat an empty result as proof of zero cost or zero risk.
  • Do not combine measured costs, inferred headroom, verified savings, and evidence costs.
  • Do not claim causality from one aggregate or one expensive trace.

Häufig gestellte Fragen

Does this skill change Caveman experiments?
No. It performs read-only reviews and only proposes possible actions.
What access does it require?
It requires Caveman authentication and a selected project context.
Does it read prompt or completion content?
Not by default. Payload retrieval requires an explicit user request.
Can it explain a cost increase?
It can test explanations using bounded cohorts and traces, but it labels unsupported causes as unproven.
How does it report savings?
It keeps verified ledger savings separate from inferred daily headroom and measured costs.
Which tools does it support?
It prefers Caveman MCP tools and uses documented Caveman CLI commands as a fallback.

Entwicklerdetails

Lizenz

MIT

Skillstore-Revision

r1

Versionshinweis

Der Autor hat keine Version angegeben.

Repository

https://github.com/juliusbrussee/caveman/tree/7bb71309e8749a4f112aacd3a54b3941d8689905/skills/caveman-evidence-review

Ref.

18c566f98cd9d675bf70dca057f35ef24e82ac57

Aktualität der Wartung

25.8.2026

Nutzung

1 Downloads · 0 Aufrufe

Dateistruktur

📄 SKILL.md