中文文档

aiops is a set of AI engineering best practices — lean-gated, resumable workflow delivered as skills and agents to 5 IDEs (Cursor, Claude Code, Codex CLI, GitHub Copilot, OpenCode) plus a generic AGENTS.md harness. Type /aiops in your AI IDE and describe the work you want done; aiops guides the task from clarification to implementation, review, and final approval.

Entry: **/aiops** (guided workflow, resumable via flow.state.yaml).

• Start from the task — describe a feature, bug, or refactor in natural language
• Follow a guided flow — aiops asks for missing decisions, shows the current step, and saves progress
• Keep AI changes small — lean discipline, TDD, prune, and review run before delivery
• Resume later — continue from .scratch/<feature>/flow.state.yaml with /aiops continue
• Stay in control — commits only happen after you explicitly approve them
• Go deeper when needed — optional agents, skills, and code graph tooling support larger work

npx -y github:yugasun/aiops

In your project chat:

/aiops Add a health check endpoint

Resume later:

/aiops continue

For a small feature, aiops usually runs:

1. Clarify scope and acceptance criteria
2. Agree the design before coding
3. Write tests first
4. Implement the smallest working change
5. Prune excess code and review the diff
6. Wait for your approval before commit

Bug reports skip the alignment ceremony and go straight to diagnosis. Larger features can be turned into a PRD and vertical-slice issues before implementation.

Architecture scans can use /code-graph for structured code understanding with graphify: Tree-sitter AST parsing plus Louvain community detection. This is optional; the core aiops flow works without extra Python tooling.

Install it only when you want stronger architecture and impact analysis:

uv tool install graphifyy
graphify --version

If you do not use uv, pip install graphifyy or pipx install graphifyy also works. The PyPI package name is graphifyy (double-y) — the CLI command is graphify.

Default install is interactive (arrow keys + space to pick IDEs, ctrl+a for all, then project vs global and hooks). Use --yes in CI.

npx -y github:yugasun/aiops                        # interactive install
npx -y github:yugasun/aiops --yes                  # non-interactive: all detected IDEs, project agents/hooks, skills in ~/
npx -y github:yugasun/aiops --all                  # same as --yes
npx -y github:yugasun/aiops --ide claude           # Claude Code only (still prompts scope/hooks unless --yes)
npx -y github:yugasun/aiops --ide cursor           # Cursor only
npx -y github:yugasun/aiops --ide codex            # Codex CLI only
npx -y github:yugasun/aiops --ide copilot          # GitHub Copilot only
npx -y github:yugasun/aiops --ide opencode         # OpenCode only
npx -y github:yugasun/aiops -g                     # global agents/hooks/skills under ~/
npx -y github:yugasun/aiops --skills-only          # slash-command skills only (no hooks/agents/always-on lean)
npx -y github:yugasun/aiops --commands-only        # alias for --skills-only
npx -y github:yugasun/aiops --agents-only          # only install agents
npx -y github:yugasun/aiops --agents-md            # append aiops block to AGENTS.md (never overwrites)
npx -y github:yugasun/aiops --no-hooks             # skills + agents, skip SessionStart hooks
npx -y github:yugasun/aiops --list                 # show detected IDEs and install targets
npx -y github:yugasun/aiops uninstall              # remove installed files
npx -y github:yugasun/aiops uninstall --ide codex  # uninstall from Codex only
npx -y github:yugasun/aiops --uninstall            # alias for uninstall

/plugin marketplace add yugasun/aiops
/plugin install aiops@aiops

5 IDEs plus a generic AGENTS.md harness for any tool that reads it.

Skills always install to user-global directories (never into the project tree). Project scope only places agents / hooks / always-on rules in the repo. Global scope puts those under ~/ as well.

AGENTS.md is off by default. Pass --agents-md (or say yes in the interactive prompt) to append a marked aiops block — never overwrite your existing file.

Full install (interactive default, or --yes) writes skills, agents, always-on lean, and hooks. Existing hooks.json entries are merged, not replaced.

To opt out of lean after a full install: run npx -y github:yugasun/aiops uninstall, or remove aiops entries from ~/.codex/hooks.json / ~/.claude/hooks.json manually.

┌─────────────────────────────────────────────────────────┐
│  /aiops (Flow Conductor)                                │
│  ├── journey state  →  .scratch/<slug>/flow.state.yaml  │
│  └── phase dispatch →  agents/*.md + skills/*/SKILL.md  │
├─────────────────────────────────────────────────────────┤
│  Adapter seam (scripts/adapters/*)                      │
│  ├── skills   → ~/…/skills/ (always global, never in repo) │
│  ├── cursor   → .cursor/rules/*.mdc + .cursor/agents/   │
│  ├── claude   → .claude/agents/ (+ ~/.claude/skills)    │
│  ├── codex    → .codex/agents/*.toml + optional AGENTS.md │
│  ├── copilot  → .github/copilot-instructions.md         │
│  └── opencode → .opencode/agents/                       │
└─────────────────────────────────────────────────────────┘

You do not need to memorize these to use aiops. They are documented for teams that want to inspect or customize the workflow.

Full list: skills/manifest.json

Delivery sequence: lean → TDD → prune → review → commit. The final commit runs only when you explicitly approve it.

Before writing any code, stop at the first rung that holds:

1. Does this need to exist? (YAGNI)
2. Stdlib does it?
3. Native platform feature?
4. Already-installed dependency?
5. One line?
6. Minimum code that works

Rules: No unrequested abstractions. Deletion over addition; shortest working diff. Mark deliberate shortcuts with // lean: <ceiling and upgrade path>.

Never cut: Trust-boundary validation, data-loss prevention, security, accessibility, explicitly requested behavior.

1. Open the project in your AI IDE and run /aiops <task>
2. Follow the prompted steps; aiops saves progress automatically
3. Resume with /aiops continue whenever you come back
4. Add aiops.yaml only if your team wants GitHub/GitLab issue tracking

Details: docs/getting-started.md

• Getting started
• Agent registry
• Skill registry
• Website

Real run logs based on aiops-demo. Full index →

• Health check walkthrough — a small API feature through clarification, TDD, review, and approval
• Architecture scan + code graph — evidence-backed architecture scan, then one chosen refactor
• Effect analysis — direct AI coding compared with the guided aiops flow
• Automated benchmark — bash docs/demos/benchmark.sh runs the comparison

Apache 2.0 — see LICENSE. Contributing: CONTRIBUTING.md.