OSOP

April 5, 2026 · View on GitHub

Your AI coding assistant does complex multi-step work. But what did it actually do? OSOP gives you structured, portable execution logs — not chat transcripts.

Install (One Line)

Option A — Copy CLAUDE.md into any project:

curl -sL https://raw.githubusercontent.com/Archie0125/osop-openclaw-skill/main/CLAUDE.md >> CLAUDE.md

Option B — Claude Code plugin:

claude /install-plugin https://github.com/Archie0125/osop-openclaw-skill

Option C — OpenClaw:

clawhub install osop

That's it. Claude Code will now generate .osop + .osoplog.yaml after multi-step tasks.

What You Get

Before: 200-line chat transcript, hard to skim
After:   Structured workflow + execution record → visual HTML report

Two Files

File	What	Example
`.osop.yaml`	What was planned — DAG of steps	"Explore → Implement → Test → Review"
`.osoplog.yaml`	What actually happened — timestamps, tools, outputs	"Read 4 files, edited 2, ran tests in 3.2s"

View Reports

Drag both files into https://osop-editor.vercel.app for:

Visual DAG diagram
Step-by-step execution timeline
Tool usage breakdown
Risk analysis overlay

OSOP Core — 4 Types, 4 Modes

OSOP uses a minimal schema for AI agent workflows:

Node types: agent (AI/LLM), api (HTTP), cli (shell), human (manual) Edge modes: sequential, conditional, parallel, fallback

A complete workflow fits in 15-25 lines of YAML.

Skills

Skill	What it does
`/osop`	Validate + render any .osop file
`/osop:auto-log`	Auto-generate .osoplog after task completion
`/osop:sop-report`	Generate standalone HTML report
`/osop:sop-security`	Security risk scan before execution
`/osop:sop-diff`	Compare two executions

Works With

Claude Code, Cursor, Windsurf, Codex, Copilot, Cline, Aider, Continue.dev, Devin, Zed, and more.

License

Apache 2.0