convez.ai
Skill
upflame

Synarc

upflame/Synarc

Ver no GitHub
BomPluginScore · de 0 a 100

A production-grade always-on engineering cognition runtime that provides architecture awareness, impact tracing, deployment risk analysis, and adaptive code intelligence across Claude Code, Claude Web, Codex CLI, Cursor, Windsurf, and SKILL.md runtimes.

Instalação

$npx convez-ai upflame/Synarc

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
2
Forks
0
Acompanhando
0
Saúde
85%
Último commit
há 4 meses
Versão
v6.0.0
Issues abertas
0
Contribuidores
2

README

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

Synarc — Engineering Intelligence Runtime

Synarc Banner

Version Stage License Runtime Build Vendor Lock-In Reference Integrity Package Size

Synarc is an engineering intelligence runtime for AI-assisted software development. It provides structured change classification, risk assessment, intent verification, audit trails, and session continuity across all major AI coding agents. The runtime operates as a skill pack — no executables, no network calls, no external dependencies.


Overview

AI coding agents execute changes rapidly, but without structured governance, each session risks context fragmentation, unplanned scope expansion, and undetected contract violations. Synarc addresses this by interposing a deterministic governance layer between developer intent and agent execution.

The runtime is organized as a multi-skill architecture:

  • Intent Contracts — formal commitments before execution: scope boundary, explicit promises, risk cap, and post-execution verification
  • Change Classification — 12 WorkTypes across 7 dimensions with deterministic risk floors
  • Verification Engine — post-execution diff analysis against contract promises
  • Audit & Compliance — immutable audit trail with rollback-to-intent and regulatory export (EU AI Act, SOC2, HIPAA)
  • Session Continuity — persistent session ledger across context resets
  • Cognition Mesh — multi-role team collaboration with shared working memory

Features

CapabilityDescription
Intent ContractsPropose → Accept → Execute → Verify → Fulfill lifecycle for every change
Intent Templates11 per-WorkType contract templates with default promises and scope rules
Verification EngineScope checking, promise verification, risk delta, composite verdict matrix
Audit TrailImmutable records per contract lifecycle event, rollback-to-intent protocol
Change Classification12 WorkTypes, 7 dimensions, deterministic risk floors per domain
Risk Assessment6-level risk ladder with hard floors for auth, payments, schema changes
Context InjectionCOMPACT (50 tokens), STANDARD (200 tokens), FULL (500 tokens) levels
Session TrackingImmutable ledger persisting across sessions and context resets
Quality GatesZero-tolerance enforcement per WorkType
Error Intelligence6-step protocol with persistent error memory
Cognition MeshMulti-role collaboration — up to 9 specialized roles per task

Architecture

The runtime implements a 7-step pipeline executed on every interaction:

Classify → Inject → Execute → Log → Aggregate → Checkpoint → Emit

Each step is deterministic and runtime-agnostic — the same pipeline operates identically across Claude Code, Codex CLI, Cursor, Windsurf, OpenCode, Gemini CLI, Copilot, Cline, and any AGENTS.md-compatible runtime.

Cache Architecture

TierContentsScope
0Pack header (AGENTS.md, manifest.yaml)Session
1Core reasoning (synarc-core, cognition-layer, schemas)Session (~60 KB)
2Active domain skill (one of 56)Task (~10 KB each)
3Skill referencesLazy-loaded
4Dynamic context (project files, tool outputs)Never cached

Scale Adaptation

ScaleThresholdTracking Depth
NANOSingle file, single purposeWorkType + risk only
MICRO2-10 filesCURRENT_STATE.md
SMALL<5k LOC, 1-5 modulesFull brain directory
MEDIUM5k-50k LOC, teamFull ledger
LARGE50k-500k LOC, multi-serviceService-boundary tracking
ENTERPRISE>500k LOC, regulatedCompliance audit trail

Auto-detected. Zero configuration.


Quick Start

Prerequisites

Any AI coding runtime: Claude Code, Codex CLI, Cursor, Windsurf, OpenCode, Gemini CLI, Copilot, or Cline.

Installation by Runtime

RuntimeMethod
Claude Codeclaude plugin marketplace add upflame/Synarc then claude plugin install synarc
Codex CLICopy synarc-universal/AGENTS.md to repo root
OpenCodeCopy synarc-universal/AGENTS.md to repo root or ~/.config/opencode/AGENTS.md
CursorCopy .cursor/rules/synarc-core.mdc to .cursor/rules/
WindsurfCopy shared/runtime-adapters/windsurf.md to .windsurfrules
Gemini CLICopy synarc-universal/AGENTS.md to GEMINI.md in repo root
CopilotAppend shared/runtime-adapters/copilot.md to .github/copilot-instructions.md
ClineCopy synarc-universal/skills/* to .cline/skills/

Verify Installation

> what did we change?
── Session Ledger ──
[14:00] FEATURE | auth/router.ts (+12, -3) | MEDIUM | IN_SCOPE
[14:05] FIX    | auth/middleware.ts (+5, -0) | LOW | IN_SCOPE
Aggregate: MEDIUM (stable)
─────────────────────

Available Commands

CommandResponse
what did we change?Full session ledger
summarize this sessionCognitive summary
is this safe to deploy?Risk delta + explicit YES/NO
what tests are missing?All unfilled test gaps
generate a snapshotBrain snapshot entry
full handoffAgent handoff block + brain updates
run quality gatesAll gates PASS/FAIL report

Project Structure

synarc-universal/
├── AGENTS.md                    # Activation entry point
├── package.json                 # npm package manifest
├── manifest.yaml                # Universal manifest
├── scripts/
│   ├── install.js               # CLI installer
│   └── install.ps1              # PowerShell installer
├── skills/                      # 56 domain skills
│   ├── synarc-core/             # Core runtime (always active)
│   ├── architect/               # System design & trade-off analysis
│   ├── backend-engineer/        # Service architecture & API design
│   ├── security-engineer/       # Threat modeling & defense
│   ├── debug-engineer/          # Systematic debugging
│   └── ...                      # 51 additional domain skills
├── shared/
│   ├── schemas/                 # JSON Schema definitions
│   ├── workflows/               # Canonical workflow definitions
│   ├── guardrails/              # Constitutional safety rules
│   ├── standards/               # Naming conventions, frontmatter spec
│   ├── runtime-adapters/        # Per-runtime compilation rules
│   ├── prompts/                 # Fallback prompt tiers
│   └── checklists/              # Review checklists
├── docs/                        # Documentation
└── security/                    # OWASP mapping, adversarial scenarios

Documentation

CategoryLinks
Getting StartedInstallation · Architecture · Usage · Deployment · Migration
WorkflowsChange Classification · Risk Assessment · Context Injection · Session Tracking · Quality Gates · Error Intelligence
Intent ContractsContract Lifecycle · Templates · Verification · Audit & Compliance
SchemasIntent Contract · Intent Template · Verification Result · Audit Record · Ledger Entry · Risk Assessment · Guardrails

Security & Compliance

GuardStatus
Sandboxed executionEnabled
No network accessVerified
No filesystem write outside /brain/Enforced
Deterministic activationValidated
Safe fallbacks on protocol errorConfigured
Hash-verified integrity (SHA-256)Active
Tamper protectionEnabled

Risk hard floors:

DomainMinimum RiskReasoning
Auth, billing, payments, securityHIGHRevenue, access, or trust impact
Schema change (remove/rename)CRITICALData integrity + migration complexity
Environment variable renameCRITICALAll deployments affected
Public API response changeHIGHAll consumers must adapt
Network / IAM configCRITICALSecurity boundary change
INCIDENT responseCRITICALProduction emergency

Regulatory mappings: EU AI Act, SOC2, HIPAA — export templates included.


Contributing

See CONTRIBUTING.md for guidelines on submitting issues, feature requests, and pull requests.

This project follows Semantic Versioning and maintains a CHANGELOG.


License

MIT — see LICENSE.

Built by UpFlame Labs.

Ver o original no GitHub

Bom para

Relacionados