Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Get Started Free →

/swarm - Collaborative Swarm Mode · vibeeval | DecimalAI

Swarm

SkillScore48Maintained

swarm

Tum ekibi ayni anda devreye sok. Agent'lar paralel calisir, birbirine soru sorar, birlikte gelistirir. Kullanim: /swarm <gorev>

agent-toolingby vibeeval

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

AI Engineer pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/swarm/SKILL.md

decimalai skills pull swarm --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/swarm/SKILL.md https://app.decimal.ai/s/swarm/SKILL.md

Raw file: /s/swarm/SKILL.md ↗ · machine-readable summary at /s/swarm.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

23 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 56d ago

—+39% tokens0% turns48% of whole cases pass with skill (11/23)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-15	✗→✓	▲ Improved	—	—
case-12	✗→✓	▲ Improved	—	—
case-17	✗→✓	▲ Improved	—	—
case-05	✗→✓	▲ Improved	—	—
case-11	✗→✓	▲ Improved	—	—

Showing 5 of 23 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 7/30/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

vibeeval / vibecosystem

skills/swarm/SKILL.md

View on GitHub ↗

License MIT LicenseImported 8/1/2026Version v1

Skill contents

SKILL.md

/swarm - Collaborative Swarm Mode

Tum ana ekibi ayni anda devreye sokar. Agent'lar paralel calisir, shared dosyalar uzerinden haberlesir.

Kullanim

/swarm Kullanici authentication sistemi ekle
/swarm Bu projeyi bastan incele ve iyilestir
/swarm Yeni e-commerce modulu gelistir

Adim 1: Proje Tespiti

Once mevcut dizini tara:

Tech stack ne?
CLAUDE.md var mi?
Mevcut kod ne durumda?

Yoksa /project-detect calistir.

Adim 2: Shared Klasor Hazirla

bash
mkdir -p .claude/shared
touch .claude/shared/swarm-questions.md
touch .claude/shared/swarm-answers.md
touch .claude/shared/swarm-review.md

Her dosyanin basina header ekle:

markdown
# Swarm Communication Log
Gorev: <gorev aciklamasi>
Baslangic: <tarih>

Adim 3: Phase 1 - Paralel Kesif

3 agent'i PARALEL calistir (Task tool ile):

Agent 1: scout

Gorevi: Codebase'i tara, mevcut pattern'leri bul, dosya yapisini cikar.
Cikti: .claude/shared/swarm-phase1-scout.md

Agent 2: project-manager (Sofia)

Gorevi: Gorevi alt gorevlere parcala, oncelik sir, bagimlilik belirle.
Sprint backlog olustur.
Cikti: .claude/shared/swarm-phase1-pm.md

Agent 3: architect

Gorevi: Mimari plan ciz, tech decisions yap, ADR yaz.
Cikti: .claude/shared/swarm-phase1-architect.md

Phase 1 tamamlaninca: 3 ciktiyi oku, sentezle, Phase 2'ye gec.

Adim 4: Phase 2 - Paralel Gelistirme

Phase 1 ciktilarini her agent'a ver, PARALEL calistir:

Agent 4: backend-dev (Dmitri)

Gorevi: API endpoints, database schema, business logic implement et.
Input: Phase 1 sentezi + PM sprint backlog'u
Cikti: .claude/shared/swarm-phase2-backend.md
Sorulari: .claude/shared/swarm-questions.md'ye yaz

Agent 5: frontend-dev (Aria)

Gorevi: UI components, pages, state management implement et.
Input: Phase 1 sentezi + Designer ciktisi (varsa)
Cikti: .claude/shared/swarm-phase2-frontend.md
Sorulari: .claude/shared/swarm-questions.md'ye yaz

Agent 6: designer (Marcus)

Gorevi: Design system, component specs, color/typography kararlari.
Input: Phase 1 sentezi
Cikti: .claude/shared/swarm-phase2-designer.md

Agent 7: devops (Kai)

Gorevi: CI/CD, Docker, deployment config, monitoring.
Input: Phase 1 sentezi
Cikti: .claude/shared/swarm-phase2-devops.md

Phase 2 tamamlaninca:

Sorulari oku (.claude/shared/swarm-questions.md)
Cevaplanmamis sorulari ilgili agent'a yonlendir
Cevaplari .claude/shared/swarm-answers.md'ye yaz
Catismalari coz

Adim 5: Phase 3 - Paralel Review

Tum Phase 2 ciktilarini 4 review agent'a ver:

Agent 8: code-reviewer

Gorevi: Kod kalitesi, best practices, maintainability.
Cikti: .claude/shared/swarm-review-code.md

Agent 9: security-analyst (Zara)

Gorevi: OWASP top 10, secrets, input validation, auth.
Cikti: .claude/shared/swarm-review-security.md

Agent 10: qa-engineer (Priya)

Gorevi: Test stratejisi, edge cases, test plani yaz.
Cikti: .claude/shared/swarm-review-qa.md

Agent 11: data-analyst (Yuna)

Gorevi: Analytics events, metrik tanimlama, A/B test plani.
Cikti: .claude/shared/swarm-review-data.md

Phase 3 tamamlaninca: Review bulgularini topla, CRITICAL olanlari listele.

Adim 6: Phase 4 - Duzeltme + Test

Review'dan gelen CRITICAL ve HIGH bulgulari duzelt:

Agent 12-13: backend-dev + frontend-dev

Gorevi: Review fix'lerini implement et.
Input: Phase 3 review bulgulari

Agent 14: tdd-guide

Gorevi: QA engineer'in test planina gore testleri yaz.
Input: Phase 3 QA ciktisi + Phase 2 implementasyon

Agent 15: verifier

Gorevi: Son quality gate - build, test, lint, security.
Input: Tum Phase 4 ciktilari
Cikti: PASS / FAIL raporu

Adim 7: Phase 5 - Finalizasyon

Agent 16: self-learner

Gorevi: Tum swarm'dan ogrenimleri cikar, CLAUDE.md'ye kaydet.
Input: Tum phase ciktilari + review bulgulari

Agent 17: technical-writer (Noah)

Gorevi: README, API docs, changelog guncelle.
Input: Phase 2 implementasyon + Phase 1 mimari

Agent 18: growth (Camille) opsiyonel]

Gorevi: Launch notu, feature announcement, GTM.
Input: Phase 1 + Phase 5 docs

Swarm Raporu

Her swarm sonunda su raporu uret:

markdown
# SWARM RAPORU
Gorev: <gorev>
Tarih: <tarih>
Sure: <toplam sure>

## Katilimcilar
| Agent | Phase | Durum |
|-------|-------|-------|
| scout | 1 | Tamamlandi |
| ... | ... | ... |

## Iletisim Ozeti
- Toplam soru: X
- Cevaplanan: Y
- Catisma cozulen: Z

## Review Bulgulari
- CRITICAL: X (hepsi cozuldu)
- HIGH: Y (Z tanesi cozuldu)
- MEDIUM: W

## Verifier Sonucu
PASS / FAIL

## Ogrenimler
- <self-learner ciktisi>

## Sonraki Adimlar
- <kalan isler>

Kurallar

Her phase bitmeden sonrakine gecme
CRITICAL review bulgulari MUTLAKA duzeltilmeli
Verifier FAIL verirse Phase 4'e don
Agent'lar arasi catismada maestro karar verir
Her agent kendi alaninin disina karismasin
Paralel calistirilabilecek agent'lari MUTLAKA paralel calistir

Illustrative Want to see what swarm would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Latest benchmark · publisher runVerified

7/29/2026, 01:43 UTC · 23 cases

Measured on 21 comparable cases — below the 22 needed to publish a lift, so this result is shown here but not on the registry card or leaderboard.

Lift

+43%

48% vs 4%

Agent turns

0%

1.0 vs 1.0

Tokens

+39%

2,347.67 vs 1,691.38

fail → pass · 10pass → pass · 1fail → fail · 12

Per-case results (23)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-21 Generate a standard conventional git commit message for my currently staged changes.	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-20 I have a syntax error on line 42 of utils/helpers.py. Can you fix the missing closing parenthesis in that file?	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-15 When two agents hold opposing technical opinions on an API contract during parallel development, who holds decision authority?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-01 I need to add a comprehensive user authentication and authorization system to our project. Please initiate a full multi-agent collaborative swarm workflow spanning initial codebase exploration, architectural planning, parallel development, multi-perspective reviews, test verification, and documentation updates, ending with the standard execution report.	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-16 If devops agent Kai notices an incorrect CSS selector during Phase 2, should Kai modify the React components directly?	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-18 Is the growth agent (Camille) mandatory for every swarm execution?	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-12 Where are the output artifacts stored for backend-dev, frontend-dev, designer, and devops during Phase 2?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-06 In Phase 4, the verifier agent runs automated quality checks and reports a FAIL status due to lint and test failures. Should the workflow immediately proceed to Phase 5 documentation?	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-14 What specific sections are required in the final markdown summary report generated at the end of a swarm?	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-17 In Phase 4, which agent writes automated tests and what inputs does it depend on?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-19 What preliminary step must be executed prior to starting agent discovery if technology stack context or CLAUDE.md is missing?	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-05 How should the shared communication workspace be initialized on disk before launching swarm agents?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-22 Please review this single line for security bugs: db.execute('SELECT * FROM users WHERE id = ' + user_input)	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-02 We are building a new e-commerce checkout module from scratch. Execute a complete multi-specialist team swarm process—from initial code scan and sprint backlog creation through parallel frontend, backend, design, and DevOps work, down to code/security reviews, quality verification, and final reporting.	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-11 Which specialized roles participate in the Phase 3 parallel review stage?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-04 We are starting Phase 1 of a major codebase refactor. Should scout, project-manager (Sofia), and architect be executed sequentially one after another?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-10 Which agent updates the project README, API documentation, and changelog during finalization, and what inputs does it consume?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-13 If project-manager finishes sprint backlog creation early during Phase 1, can backend-dev start Phase 2 immediately while architect is still working?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-03 Can you run a full multi-stage agent swarm process to perform a comprehensive audit and refactoring of this repository? I want the team to handle discovery, architecture planning, parallel refactoring, security and QA reviews, critical issue fixes, and final documentation, concluding with a detailed swarm summary.	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—
case-09 After Phase 4 verification succeeds, which agent extracts learnings and where are those insights recorded?	pass→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-07 Phase 3 security review identified CRITICAL security vulnerabilities alongside MEDIUM styling suggestions. Can final documentation proceed while CRITICAL findings are deferred to a later sprint?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-08 During Phase 2 development, backend-dev Dmitri encounters a question regarding database schema constraints. Which file should receive this question and where should answers be written?	fail→pass	—	—	—	—	—	—	—	—	—	—	—	—
case-23 Write a standard Dockerfile for a Python FastAPI application using uvicorn.	fail→fail	—	—	—	—	—	—	—	—	—	—	—	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 21 counted toward the lift figure. The other 2 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +43 percentage points is the difference between those two pass rates over the 21 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 4.4%
Pass rate with the skill: 47.8%
Cases attempted: 23
Cases counted toward lift: 21
Cases that regressed: 1
Skill version measured: v1
Run completed: 7/29/2026, 01:43 UTC

The per-case answers from this run were removed by the retention sweep, so the case table below shows the verdicts without the text either arm produced. The counts above were recorded at the time and are unaffected. Answers are now kept for 180 days.

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run48

Live evalNeeds ≥10 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns+21% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 55 days ago

Other measured skills in the registry, with their headline benchmark lift.