convez.ai
MCP ServerDevOps & InfraestruturaAvançado
sergeeey

Claude Code Top 2026 — MCP Server

sergeeey/Claude-cod-top-2026

Elimina validação falsa em IA com Evidence Policy e 60 hooks Python

Ver no GitHub
BomPluginScore · de 0 a 100

MCP server com 60 hooks, 109 skills e 15 agents para eliminar "validation theater" em Claude Code. Enforça Evidence Policy como regras Python determinísticas, não instruções soltas. Inclui proteção contra hallucinations, injeção de prompt, vazamento de PII e falhas de MCP com auto-recovery em 60s.

Instalação

$npx convez-ai sergeeey/Claude-cod-top-2026

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
9
Forks
0
Acompanhando
0
Saúde
100%
Último commit
há 3 dias
Versão
v3.10.0
Issues abertas
6
Contribuidores
2

Casos de uso

  • Validar claims de modelos contra dados reais vs sintéticos
  • Proteger produção contra hallucinations com CircuitBreaker
  • Auditar segurança de código com review-squad paralelo

Veredito

Abordagem sistemática para production-grade Claude Code com foco em integridade de dados e auto-recovery. Excelente para equipes que precisam eliminar validação falsa em workflows críticos, embora requeira overhead de setup e compreensão profunda de hooks.

README

A tradução para o português deste README ainda está na fila.



Why This Config?

Claude Code без этого конфига — как Ferrari на ручнике: мощный, но 60% потенциала потеряно. With this config — каждый коммит верифицирован, каждый агент помнит контекст, каждая ошибка записана и больше не повторяется.

Most configs are a single CLAUDE.md bloated to 3000+ tokens. This is different:

Typical configThis config
Tokens/msg3 000 – 5 000~500 (core only)
Hallucinations"trust me"Evidence Policy + Confidence Scoring
MCP failuressession hangsCircuitBreaker — auto-recovery in 60s
Prompt injectionno protectionInputGuard — 8 categories, auto-block
PII leakagehope for the best17 patterns: secrets + PII (email, phone, card, passport, СНИЛС)
Code reviewoptionalreview-squad — parallel reviewer + sec-auditor
Permissionsask for everythingPermissionRequest hook — 75% auto-approved
Agent memorystateless4 agents with persistent memory across sessions
Tests"I'll write them later"1367 tests, TDD-first, Test Protection hard rule

Where This Fits in the Ecosystem

Three repos solve three different layers. They compose, not compete:

pm-skills (phuryn)        → WHAT to build: discovery, strategy, PRD, assumptions
agent-skills (addyosmani) → HOW to build: spec → plan → build → test → review → ship
Claude-cod-top-2026 (you) → GUARDRAILS: evidence policy, anti-hallucination, hooks, security
pm-skillsagent-skillsThis config
FocusProduct managementEngineering lifecycleRuntime guardrails
Primary userPMs, foundersEngineersEngineers on sensitive/research projects
Stars~15k~53kearly
Anti-hallucinationskill-level guidanceskill-level guidanceenforced in hooks — runs on every action
Evidence policynonenone[VERIFIED-REAL] vs [VERIFIED-SYNTHETIC] hard rule
Security hooksnonenonePII redact, secrets guard, injection block
Multi-platformClaude, Cursor, GeminiClaude, Cursor, Gemini, WindsuffClaude Code only

Use all three together for the full stack. Use only this one if you need runtime enforcement without PM tooling.


When to Use This vs everything-claude-code

everything-claude-code is a great alternative — bigger, multi-platform, Anthropic Hackathon Winner. Both are MIT, pick what fits.

Use everything-claude-code if:

  • You want multi-language coverage (TS, Go, Java, Kotlin, Rust, C++, PHP — 12+ ecosystems)
  • You work across multiple harnesses (Codex, Cursor, OpenCode, Gemini — not just Claude Code)
  • You want a GUI dashboard for browsing components
  • You like the paid tier path (ECC Tools GitHub App, free / pro / enterprise)

Use this config if:

  • "Validation Theater" is a $$$ risk for you, not abstract — Evidence Policy is enforced as hard rule, not just a skill
  • You work with sensitive data (PII, finance, healthcare) — built-in redaction hook scrubs sensitive strings before any external MCP call
  • You need to read every hook before installing — only ~10 MB, plain Python, no JS dependencies, every file readable in 10 minutes
  • You prefer Claude Code only with deep specialization over multi-platform breadth
  • You speak Russian — README and rules have RU-first sections, useful for CIS dev teams

Comparison at a glance:

everything-claude-codeThis config
Surface48 agents · 182 skills · 68 commands · ~31 MB15 agents + 3 squads · 114 skills · 60 hooks · ~10 MB
LanguagesTS, Py, Go, Java, Kotlin, Rust, C++, PHP, PerlPython primarily
HarnessesClaude Code, Codex, Cursor, OpenCode, Gemini, AntigravityClaude Code only
Anti-hallucinationcontinuous-learning v2 with confidence scoringEvidence Policy + Validation Theater Guard + Audit Verification Gate (synthetic ≠ real, enforced)
PII / sensitive datagenericdedicated redaction hook + local-first (Ollama)
Audit Verification Gatenot in corerules/audit-verification-gate.md — agent's [VERIFIED] = your [INFERRED]
Recurring mistake trackinginstinct-based[×N] counter — after 3 occurrences a mistake becomes a hard rule
LicenseMIT (open core, paid GitHub App)MIT (no paid tier)

If multi-language / cross-harness matters more than anti-hallucination focus — pick ECC. If anti-hallucination on sensitive data is your job-critical risk — pick this one.


What This Config Does NOT Do

Honest scope fence — to prevent misuse and save your time:

  • ❌ Not multi-language. Python primarily. TS/JS supported via hooks but agents/skills are Python-tuned.
  • ❌ Not multi-harness. Claude Code only. No Codex / Cursor / Gemini support — see everything-claude-code for cross-harness.
  • ❌ Not a GUI / dashboard. Pure CLI + file-based config. scripts/hook_metrics.py is the closest thing (terminal table).
  • ❌ Not a SaaS / managed service. Self-host only. No paid tier, no telemetry to author.
  • ❌ Not for >50% of generic projects. This is optimized for anti-hallucination on sensitive data (PII, finance, healthcare, research). If your stack doesn't have validation theater risk, the overhead may exceed the value.
  • ❌ Not a replacement for human review. Hooks catch deterministic mistakes (commits to main, leaked secrets, debug prints). Logic bugs still need Agent(reviewer) or human eyes.
  • ❌ Not a methodology textbook. The rules (rules/*.md) document the patterns we use, but they're not a tutorial. Read docs/methodology.md for the explained version.

🚀 Start Here (pick your path)

New to this? Don't install everything at once. Pick the path that matches your goal:

PathWhat you getTimeCommand
Evidence Only[VERIFIED] markers + anti-hallucination2 min--profile=minimal
Daily Driver+ 60 hooks + 15 agents + 114 skills5 min--profile=standard
Full Setup+ MCP profiles + PII redaction + memory10 min--profile=full

Minimal path (recommended to start): installs just 3 files — CLAUDE.md, integrity.md, security.md. No hooks, no agents, no complexity. Add more when you need it.


Quick Start

# One-liner — Mac / Linux / WSL
git clone https://github.com/sergeeey/Claude-cod-top-2026.git && cd Claude-cod-top-2026 && bash install.sh --profile=standard --non-interactive

Windows (PowerShell): git clone https://github.com/sergeeey/Claude-cod-top-2026.git; cd Claude-cod-top-2026; bash install.sh --profile=standard --non-interactive

After install: restart Claude Code (/clear or new session) — hooks activate automatically.

Verify the stack

pytest tests/ -q --tb=short          # 1367 tests, 0 failures
bash tests/test_all.sh               # 296/296 smoke tests
ruff check hooks/ scripts/ tests/    # 0 errors

Plugin Install (recommended — Claude Code v2.1.80+)

# Register this repo as a marketplace source (once per machine)
/plugin marketplace add sergeeey/Claude-cod-top-2026

# Install the plugin
/plugin install claude-cod-top-2026

Windows note: Claude Code doesn't pre-register third-party marketplaces on Windows. Add to your ~/.claude/settings.json manually if /plugin marketplace add fails:

"extraKnownMarketplaces": {
  "claude-cod-top-2026": {
    "source": { "source": "github", "repo": "sergeeey/Claude-cod-top-2026" }
  }
}

Classic Install (all platforms)

git clone https://github.com/sergeeey/Claude-cod-top-2026.git
cd Claude-cod-top-2026

bash install.sh                                    # interactive
bash install.sh --link full                        # symlink + auto-update
bash install.sh --profile=full --non-interactive   # CI / headless
ProfileInstallsFor whom
minimalCLAUDE.md + integrity + securityTry Evidence Policy
standard+ all rules + hooks + skills + agentsDaily work
full+ MCP profiles + PII redaction + memoryFull control

60 Hooks — 25 Events

Hooks run 100% of the time — deterministic Python guards, not probabilistic instructions.

HookProtects Against
input_guardPrompt injection via MCP (8 categories)
mcp_circuit_breakerSession hang on MCP failure (auto-recovery 60s)
mcp_locality_guardMCP call without local search first
pre_commit_guardCommits to main · rm -rf · DROP TABLE
read_before_editEdit without prior Read
security_verifySensitive file edits without review
plan_mode_guard3+ files edited without a plan
permission_policy75% fewer permission prompts
checkpoint_guardRisky ops without checkpoint
HookProtects Against
mcp_circuit_breaker_postRecords MCP failures for recovery
post_formatUnformatted code (ruff / prettier)
memory_guardForgotten memory update after commit
post_commit_memoryContext loss after commits
pattern_extractorLost lessons from fix: commits
drift_guardScope creep (NOT NOW keywords)
post_tool_failureRepeated failures without strategy change
config_auditUnauthorized settings changes
elicitation_guardElicitation events logging
moc_autolinkNotes written without Obsidian MOC links
observation_captureObservations lost after file edits
HookEventRole
session_startSessionStartLoad context from memory
session_endSessionEndTrim + save state
session_saveStop (async)State persistence on exit
post_compactPostCompactContext reminder after compaction
pre_compactPreCompactData preservation
keyword_routerUserPromptSubmitAuto-trigger skills + power modes
thinking_levelUserPromptSubmitBoost thinking for complex tasks
statuslinePostToolUseLive bar: model · context% · cost
worktree_lifecycleWorktreeCreate/RemoveTrack experiment branches
agent_lifecycleSubagentStart/StopContext in agent handoffs
subagent_verifySubagentStopVerify agent output quality
team_rebalanceTeammateIdleRebalance idle agents
stop_failureStopFailureSilent API error handling
task_auditTaskCreated/CompletedTask event logging
instructions_auditInstructionsLoadedTrack loaded rules
env_reloadFileChangedStale env after .env change
direnv_loaderCwdChangedWrong env after cd
async_wrapper—Non-blocking wrapper for bg hooks
webhook_notifyStop (async)Slack/Telegram on commit + session end
thematic_index_routerStopRoute wiki entries to thematic indices

⚡ Power Modes — Magic Keywords

Type anywhere in your prompt:

KeywordModeEffect
ralphPersistentDon't stop until done. Auto-retry. No confirmations.
autopilotFull AutonomyPlan + execute all steps. Only stop if truly blocked.
ultrawork / ulwMax ParallelismLaunch agents concurrently. Batch ops. Speed > caution.
deepDeep AnalysisRead everything. Evidence-mark all claims.
quick / быстроSpeedMinimal output. No explanations. Just do it.

Modes are additive — ralph security audit = Persistent mode + security-audit skill.


13 Agents + 3 Teams

╔══════════════════════════════════════════════════════════╗
║  STRATEGIC — Opus          ·  20% of tasks               ║
║  navigator(memory:user)  architect  sec-auditor  teacher  ║
╠══════════════════════════════════════════════════════════╣
║  WORKHORSE — Sonnet        ·  80% of tasks               ║
║  builder(worktree)  tester(worktree)  explorer  reviewer  ║
╠══════════════════════════════════════════════════════════╣
║  TEAMS — parallel execution                              ║
║  review-squad  →  reviewer + sec-auditor (parallel)      ║
║  build-squad   →  builder  + tester     (isolated wt)    ║
║  research-squad →  explorer + verifier  (search+verify)  ║
╚══════════════════════════════════════════════════════════╝

4 agents with persistent memory · 2 agents with worktree isolation · Sonnet-first, Opus escalation only


Evidence Policy

Every factual claim is tagged:

[VERIFIED-HIGH]    ≥2 sources confirmed       → can be used as fact
[VERIFIED-MEDIUM]  1 source + inference        → careful wording
[VERIFIED-LOW]     indirect evidence           → "there are signs, but..."
[UNKNOWN]          no confirmation             → do not guess

Confidence Scoring 0.0–1.0 based on source count. Rationalization Prevention — 10 common AI excuses with countermeasures baked into rules.


Status Line

Zero token cost — always visible at the bottom of the terminal:

[claude-sonnet] ▓▓▓▓▓▓▓░░░░░░░░░░░░░ 35% | main | $0.42 | 3m5s
Context %ColourSignal
< 50%🟢 GreenKeep working
50–70%🟡 YellowPlan a /clear soon
> 70%🔴 Red/clear now

Security

InputGuard — 7 injection categories:

CategoryExampleAction
encoding_attacknull bytes, zero-width charsAUTO-BLOCK
command_injection; rm -rf · `$(curl)`AUTO-BLOCK
system_override"ignore previous instructions"Block + warn
jailbreak"DAN mode", "bypass safety"Block + warn
data_exfil"send to http", "curl | bash"Block + warn
role_injection[SYSTEM], <system>Warn
credential_harvest"show me your api key"Warn

PII Redaction — 12 patterns stripped before external MCP calls: National IDs · Bank cards · IBAN · API keys · GitHub tokens · AWS keys · JWT · Email · Phone · IPs


Testing

pip install pytest pytest-cov ruff mypy

pytest tests/ -v --cov=hooks --cov-report=term-missing   # 1367 tests
ruff check hooks/ scripts/ tests/
mypy hooks/utils.py hooks/input_guard.py
bash tests/test_all.sh   # 82 smoke tests
1367 passing · 0 failing · 75% coverage · 296/296 smoke tests

Obsidian Integration

Two automation hooks keep your Obsidian vault in sync with Claude Code activity:

HookTriggerWhat it does
moc_autolinkPostToolUse Write/EditTags new notes → auto-links to relevant MOC (Claude-cod, GeoMiro, Research…)
thematic_index_routerStopRoutes fresh wiki entries to Claude-Code / Lessons / Projects indices

Vault layout (~/.claude/memory/):

wiki/          ← processed knowledge (auto-generated)
raw/           ← quick drop → auto-converted at session end
mocs/          ← Maps of Content (6 MOCs)
_auto/wiki/    ← thematic indices (Claude-Code / Lessons / Projects)
daily/         ← session reports

graph.json colorGroups must be set while Obsidian is closed — the app overwrites on launch.


MCP Profiles

CORE (default)    SCIENCE              DEPLOY
context7          + ncbi-datasets      + vercel
basic-memory      + uniprot            + netlify
playwright        + pubmed-mcp         + supabase
ollama                                 + sentry
~/.claude/mcp-profiles/switch-profile.ps1 science

CircuitBreaker auto-fallback: context7 → WebSearch · playwright → WebFetch · ollama → cloud


Claude-cod-top-2026/
├── CLAUDE.md                      Core config (66 lines, ~500 tokens)
│
├── rules/                         14 modular rules (9 core + 5 extended, loaded on demand)
│   ├── coding-style.md
│   ├── security.md
│   ├── testing.md
│   ├── integrity.md
│   ├── memory-protocol.md
│   ├── context-loading.md
│   ├── permissions.md
│   ├── mentor-protocol.md
│   ├── rationalizations.md
│   ├── doubt-driven-development.md
│   ├── falsification-ladder.md
│   ├── estimand-ops.md
│   ├── audit-verification-gate.md
│   └── skeptic-triggers.md
│
├── hooks/                         60 Python guards (56 hooks + 4 support libs)
│   ├── utils.py                   21 shared functions (DRY)
│   ├── settings.json              Hook registry + 27 deny patterns
│   ├── input_guard.py             Prompt injection
│   ├── mcp_circuit_breaker.py     MCP resilience
│   ├── statusline.py              Terminal status bar
│   └── ...                        55 more hooks
│
├── agents/                        15 active + 3 teams
│   ├── navigator.md               Strategic (Opus, memory:user)
│   ├── builder.md                 Code (Sonnet, worktree)
│   ├── reviewer.md                Review (Sonnet, memory:project)
│   ├── sec-auditor.md             Security (Opus, memory:project)
│   └── teams/                     review-squad · build-squad · research-squad
│
├── skills/
│   ├── core/                      12 universal skills
│   └── extensions/                102 domain skills
│
├── assets/                        Visual assets
│   ├── banner.svg                 Hero banner (animated)
│   └── pipeline.svg               Hook execution pipeline diagram
│
├── tests/                         1367 tests · 39 files
├── docs/                          Architecture · guides · anti-patterns
├── mcp-profiles/                  3 profiles (core/science/deploy)
└── .github/workflows/ci.yml       pytest + ruff + mypy + secrets scan
DocumentDescription
Proof PackEvery README claim verified + reproduce commands
Architecture6-layer system design
Evidence PolicyAnti-hallucination + Confidence Scoring
Hooks GuideAll 60 hooks with examples
Skills GuideCreating and managing skills
Anti-Patterns9 critical mistakes to avoid
Troubleshooting10-point diagnostic checklist
CONTRIBUTINGContribution guidelines
CHANGELOGVersion history

Used in Production

This config runs on a live system (29K LOC, real users, real deploys):

  • pre_commit_guard blocked accidental push to main during a hotfix
  • pattern_extractor auto-logged debugging lessons from fix: commits
  • memory_guard kept activeContext.md current across 3 deploy cycles

Not a demo. Real system. Real incidents.


Ver o original no GitHub

Bom para

Relacionados