菜单
Syntropic137 infrastructure operations; service map, Docker stack, workspace lifecycle, Envoy token injection security, QA/testing commands, and troubleshooting recipes
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/syntropic137/syntropic137-claude-plugin --skill platform-ops命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Search, install, and manage Syntropic137 workflow plugins from marketplace registries; browse available workflows, review before installing, manage registry sources
Query Syntropic137 agent sessions, tool timelines, token metrics, cost breakdowns, and interpret observability data; why was a session expensive, why did it fail
Control running Syntropic137 executions; list, pause, resume, cancel, and check status of workflow executions
基于 SOC 职业分类
正在显示 SKILL.md
| name | platform-ops |
| description | Syntropic137 infrastructure operations; service map, Docker stack, workspace lifecycle, Envoy token injection security, QA/testing commands, and troubleshooting recipes |
When something is broken at the infrastructure level (a service is down, a workspace won't build, API calls are failing), this is the reference for diagnosing and fixing it. Start with just health-check before reading logs. It pinpoints the failing service in seconds and saves you from grepping through megabytes of log output.
Use this when you are: diagnosing a service that won't start, understanding the security model for agent token injection, running QA, operating the selfhost stack, or navigating the project structure.
Not needed for first-time setup; use the setup skill for that. Not needed for session costs or observability data; use the observability skill.
Syntropic137 has two core capabilities:
The platform is split into two lanes: Domain State (event-sourced aggregates, where aggregates decide, handlers execute) and Telemetry (append-only time-series data). Both feed the same API but are queried differently.
| Service | Port | Role |
|---|---|---|
syn-api | 8137 | FastAPI HTTP: all routes and application services |
syn-collector | 8080 | Event ingestion from agents |
| Event Store | 50051 | Rust gRPC event store |
| TimescaleDB | 5432 | PostgreSQL + time-series (projections + telemetry) |
| Redis | 6379 | Cache and pub/sub |
| MinIO | 9000/9001 | S3-compatible artifact storage |
| Envoy Proxy | 8081/9901 | API key injection sidecar |
| Dashboard | 5173 (dev) | Vite + React real-time UI |
TimescaleDB is the most critical dependency. If it's unhealthy, the Event Store can't connect, the API can't start, and nothing works. Always check TimescaleDB first when diagnosing cascading failures.
| Task | Command |
|---|---|
| Start dev stack | just dev |
| Stop (preserve data) | just dev-stop |
| Remove containers (preserve volumes) | just dev-down |
| View logs | just dev-logs |
| Diagnose env issues | just dev-doctor |
| Check health | just health-check |
| Wait for health | just health-wait 180 |
| Build workspace image | just workspace-build |
| Run full QA | just qa |
| Seed sample data | just seed-all |
Agents never see real API keys. The Envoy proxy sidecar (port 8081) handles credential injection:
InjectTokensCommand sends real keys to Envoyproxy-managed)This means agents can't exfiltrate credentials even if the prompt is hijacked. Network isolation: agents on agent-net can only reach Envoy, not the internet directly.
PENDING → CREATING → READY → RUNNING → DESTROYED
↘ ERROR
Each workflow phase gets its own isolated workspace. Rebuild the image when tools need updating: just workspace-build. Signs a rebuild is needed: agents failing with "command not found", missing CLIs, or security CVEs in the base image.
API not responding:
just health-check to identify which service is downdocker compose ps to check if containers are actually runningjust dev-logs to look for startup errorsEvent Store not connecting:
Event Store (port 50051) is a Rust gRPC service that depends on TimescaleDB. If TimescaleDB starts slow, Event Store will fail to connect on boot. Fix: just dev-stop && just dev (restart after DB is healthy).
Stale workspace image:
Agents fail with missing tools or old Claude CLI version. Fix: just workspace-build. On Apple Silicon, allow 5-10 min for the Rust components.
Database issues:
just dev-stop && just dev for soft restart. If tables are corrupted: just dev-down destroys volumes and you'll lose local data.
Dashboard not loading:
just dashboard-install then just dashboard-frontend. Dashboard requires the API to be healthy on port 8137.
just qa # full suite: lint + typecheck + test + coverage + vsa-validate
just test-unit # fast, parallel
just test-integration # needs test stack running
just typecheck # Pyright strict mode
just lint # Ruff
just vsa-validate # Vertical Slice Architecture validation
Use setup skill for first-time configuration. After diagnosis, use execution-control to check execution state or observability to check session health. Run just dev-doctor for automated env diagnostics before filing a bug.