SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI codingagentic-os

agentic-os

Read-only access to the agentic-os governance, SDLC and QE methodology, with install planning and verification.

Open website

About this tool

Purpose

agentic-os is a governed, portable coding-agent engineering platform designed to address several recurring problems in multi-agent coding: agent write scope quietly exceeding authorized files, agents rationalizing defects during self-review that they should have caught, autonomous mode being a crude on/off switch rather than a disposition strategy graded by decision risk, and stack facts being re-guessed every session with rules potentially leaking across stacks. The platform closes each of these gaps with mechanisms that are enforced rather than merely requested in prompts.

Main Components

The repository is a marketplace hosting three plugins:

  • agentic-os: the product itself, containing the three skills /agentic-init, /agentic-doctor, and /agentic-upgrade, along with the template library, generators, and role presets from which they generate scaffolding.
  • agentic-sdlc: the SDLC pipeline and gate arbiter, which handles each judgment gate according to decision risk (HITL mode asks a human; autonomous mode follows deterministic → fast path → surrogate subagent → escalation), and retains a complete decisions.jsonl audit trail.
  • agentic-qe: a tool-agnostic catalog of quality engineering AI blueprints (28, organized by STLC phase), containing two skills: qe-blueprints generates a fillable agent framework from a selected blueprint, and eval-harness generates a two-layer evaluation framework for skills and agents. There is also mcp/: a read-only MCP server that provides the same governance/SDLC/QE methodology to MCP clients that do not support the Claude Code or Cursor plugin formats, packaged independently and published on npm and the MCP Registry.

Main Capabilities

  • After a single interview (role, autonomy level, stack), it scaffolds a consistent system into the project: canonical agent contracts, enforcement hooks (pre-commit blind review, write-scope guard, instruction quality generation gate), a human-in-the-loop escalation ladder, and generation of stack-specific agents when adapting to the stack.
  • Stack-agnostic: instantly recognizes six curated profiles (Next.js, Django, Spring, Rails, Go, Playwright); other stacks (such as a NestJS backend, a schemaless MongoDB service, or no code stack at all) generate evidence-backed persistence and API agents through live repository discovery. Frontend/UI generation uses the same discovery mechanism but is at an earlier validation cycle.
  • Role-agnostic: supports roles such as developer, QA, BA/PO, architect, and delivery; non-coding roles work entirely through ordinary chat phrases.
  • The contracts produced by the scaffolding are by design host-agnostic (thin per-host pointer files overlay a normalized, host-agnostic body).

How to use

Install two plugins in an AI coding editor (Claude Code or Cursor). Quick flow: install the plugin for your editor; open the git repository you want to improve (not this marketplace repository); run /agentic-init --presets <role[,role...]> (or /agentic-init for a full role-selection interview); run /agentic-doctor and expect to see passed: true in .agentic/agentic-os/doctor.json; review the scaffolded files with git status; nothing is committed automatically; if using the SDLC pipeline, run sdlc-preflight and expect to see passed: true in .agentic/agentic-sdlc/doctor.json. It is recommended to first walk through the flow in a throwaway repository before touching a real project.

Dependencies: Claude Code or Cursor; python3 on PATH (the enforcement hooks are Python scripts); git (the target repository must be a git repository; /agentic-init can run git init); node on PATH when using agentic-sdlc; superpowers ≥ 6.1.0 and agentic-sdlc ≥ 0.4.4 (required by /agentic-init); optional gh (GitHub CLI) for the GitHub issue/MR adapter.

Role selection is explicit; for example, when the work leans toward requirements, customers, and delivery, choose ba-po, which installs only the orchestrator and requirements workflows, not the developer/code-writing agents. Later you can rerun /agentic-init --presets ba-po,developer to add roles; the installer merges the existing architecture and preserves all user files.

MCP is optional; when setting it up you can choose: connect now, configure later, or continue without MCP (paste tables, attach CSVs, share screenshots, or paste Power BI findings). Cursor project MCP configuration is .cursor/mcp.json, and the global one is ~/.cursor/mcp.json; the shared project configuration for Claude Code is .mcp.json. Secrets should be placed in OAuth or environment variables and never written into committed JSON.

When it applies

  • Teams that want to set clear boundaries for AI coding agents, enforce human-in-the-loop and independent review.
  • Teams that want only guardrails and not the pipeline, or (on Codex) only the pipeline.
  • Requirements, customer, and delivery workflows for non-programming roles.
  • Query this methodology read-only through any MCP-capable assistant, list role presets and QE blueprints, turn roles into an installation plan, and audit an existing installation.

Limitations

  • The plugin is delivered as Claude Code and Cursor plugins; agentic-sdlc additionally provides a Codex package.
  • This plugin is a custom marketplace and is not in Cursor's Browse Marketplace → All; you need to add a custom marketplace and install from the User tab.
  • Frontend/UI generation on non-curated stacks is in an earlier validation cycle.
  • No information on fees, pricing, etc. is provided in the text.

Related tools

Browse category