About setup
The setup skill guides configuration of Ouroboros from a Claude Code session, including runtime selection, optional CLAUDE.md integration, and repository registration. It is for users installing or configuring the Ouroboros plugin while preserving host-owned MCP configuration.
Key features
- 01Guided Claude runtime setup
- 02Python and package-runner detection
- 03Optional CLAUDE.md reference integration
- 04Brownfield repository scan and defaults
Use cases
- 01Configure Ouroboros for Claude Code
- 02Check Python and uv readiness
- 03Add or remove the project reference block
- 04Register existing repositories for interview context
MCP support
The skill distinguishes the Claude SDK profile on MCP 1.x from a plugin-owned MCP 2 process using a CLI worker. It explicitly avoids modifying ~/.claude/mcp.json.
Install
Copy the command and run it in a terminal.
npx skills add https:/ / github.com/ q00/ ouroboros --skill setupAfter installing, type/setupin Claude Code to confirm it loaded.
Details
From the project README/ouroboros:setup
Guided Ouroboros setup inside a Claude Code session.
Other runtimes (Codex, Copilot, OpenCode, and others) configure through the CLI instead, for example
ouroboros setup --runtime codexorouroboros setup --runtime copilot. See Getting Started.
Usage
ooo setup
/ouroboros:setup
/ouroboros:setup --uninstall
Claude setup confirms the runtime profile (the Claude Agent SDK on MCP 1.x)
and optionally adds an Ouroboros reference block to the project's CLAUDE.md.
It deliberately leaves ~/.claude/mcp.json untouched because marketplace
plugin wiring owns that file. [claude] and its explicit [claude-sdk] alias
use MCP 1.x. The plugin launches [mcp] in a separate MCP 2 process with the
dependency-free [claude-cli] worker.
Setup Wizard Flow
Ask questions with the active runtime's ask_user capability unless a step
says otherwise.
Python Runtime (Required)
Define this resolver in the same shell before any snippet below. It accepts
only Python 3.12 or newer (python3, then python, then uv) and preserves
arguments and stdin. Quote every argument passed to ouroboros_python.
ouroboros_python() {
if command -v python3 >/dev/null 2>&1 &&
(unset PYTHONHOME PYTHONPATH PYTHONPLATLIBDIR PYTHONEXECUTABLE __PYVENV_LAUNCHER__; command python3 -c 'import sys; raise SystemExit(sys.version_info < (3, 12))') >/dev/null 2>&1
then
(unset PYTHONHOME PYTHONPATH PYTHONPLATLIBDIR PYTHONEXECUTABLE __PYVENV_LAUNCHER__; command python3 "$@")
return
fi
if command -v python >/dev/null 2>&1 &&
(unset PYTHONHOME PYTHONPATH PYTHONPLATLIBDIR PYTHONEXECUTABLE __PYVENV_LAUNCHER__; command python -c 'import sys; raise SystemExit(sys.version_info < (3, 12))') >/dev/null 2>&1
then
(unset PYTHONHOME PYTHONPATH PYTHONPLATLIBDIR PYTHONEXECUTABLE __PYVENV_LAUNCHER__; command python "$@")
return
fi
if command -v uv >/dev/null 2>&1; then
(unset PYTHONHOME PYTHONPATH PYTHONPLATLIBDIR PYTHONEXECUTABLE __PYVENV_LAUNCHER__; command uv run --no-project --quiet --python '>=3.12' python "$@")
return
fi
printf '%s\n' 'Ouroboros skills require Python >= 3.12 or uv on PATH.' >&2
return 127
}
Step 0.5: Community Support
If star_asked is not true in ~/.ouroboros/prefs.json, ask "Ouroboros is
free and open-source. A GitHub star helps other developers discover it. Star
the repo?" with Star on GitHub (run gh api -X PUT /user/starred/Q00/ouroboros)
and Skip for now. After any answer, merge star_asked: true into the
prefs file, keeping existing keys:
ouroboros_python - <<'PY'
import json, os
path = os.path.expanduser('~/.ouroboros/prefs.json')
os.makedirs(os.path.dirname(path), exist_ok=True)
try:
with open(path, encoding='utf-8') as f:
prefs = json.load(f)
if not isinstance(prefs, dict):
prefs = {}
except Exception:
prefs = {}
prefs['star_asked'] = True
with open(path, 'w', encoding='utf-8') as f:
json.dump(prefs, f, indent=2)
f.write('\n')
PY
Step 1: Environment Detection
ouroboros_python --version
which uvx 2>/dev/null && uvx --version 2>/dev/null
which claude 2>/dev/null
uv python list 2>/dev/null | grep "cpython-3.1[2-9]"
The resolver already rejects Python below 3.12 and falls back to a uv-managed
interpreter; that does not make the [claude-sdk] and MCP 2 profiles
import-compatible. Report what was found, then:
| Environment | Action |
|---|---|
| Python >= 3.12 + Claude CLI | Ready: configure [claude] SDK/MCP 1 and skills |
| uvx + Python >= 3.12 | MCP-capable elsewhere: use a supported CLI-backed runtime setup for isolated ouroboros-ai[mcp] |
| Python < 3.12 only | Run uv python install 3.12, then re-run ooo setup |
| No package runner or Ouroboros package | Install uv (below), then re-run ooo setup |
Prefer package managers for uv: pipx install uv, pip install --user uv, or
brew install uv; the vendor one-liner
curl -LsSf https://astral.sh/uv/install.sh | sh is the last resort.
IMPORTANT: Never install [mcp,claude], [mcp,claude-sdk], or [all,mcp]
together and never write a direct
ouroboros or python -m ouroboros MCP fallback. MCP 2 launchers must use an
isolated uvx --isolated --python '>=3.12' --from 'ouroboros-ai[mcp]' ... or
pipx run --spec 'ouroboros-ai[mcp]' ... process. Only [mcp,claude-cli] is
supported because the CLI worker is out of process. Do not write
an Ouroboros entry to ~/.claude/mcp.json; the plugin owns that registration.
Step 1.5: Runtime Configuration
Write the Claude runtime selection to ~/.ouroboros/config.yaml. The welcome
gate and the CLI read it; nothing else in this wizard writes it. This step
never touches ~/.claude/mcp.json.
Define ouroboros_cli in the same shell as each command below (shell state
does not persist between commands):
ouroboros_cli() {
if command -v ouroboros >/dev/null 2>&1; then
command ouroboros "$@"
else
uvx --isolated --python '>=3.12' --from 'ouroboros-ai' ouroboros "$@"
fi
}
ouroboros_cli config show orchestrator
ouroboros_cli config show llm
The welcome gate is ready only when orchestrator.runtime_backend is claude
or claude_mcp AND llm.backend is claude. A missing file or section
counts as not set.
- Both conditions hold: already configured. Skip to Step 2.
runtime_backendisclaude_mcpbutllm.backendis notclaude: runouroboros_cli setup --runtime claude-cli --non-interactive(keeps the CLI transport).runtime_backendis another runtime (for examplecodex): ask the user whether to switch the default runtime to Claude. If they decline, keep the existing configuration and skip to Step 2.- Otherwise (
runtime_backendisclaudewith anotherllm.backend, no config, noorchestratorsection), or after the user agrees to switch, run:
ouroboros_cli setup --runtime claude --non-interactive
Both setup commands set llm.backend: claude; --runtime claude sets
orchestrator.runtime_backend: claude and --runtime claude-cli sets claude_mcp.
If it fails, show the error and continue the wizard; Step 4 reports the
runtime as not configured.
Step 2: MCP Profile Boundary
Tell the user: the default Claude SDK profile stays on MCP 1.x, and the
plugin-owned MCP server runs MCP 2 separately with the [claude-cli] worker.
Do not create, update, or remove ~/.claude/mcp.json. Existing entries may
be user-managed or belong to another compatible runtime. Advanced MCP
workflows require a host-managed isolated [mcp] launcher, which the Claude
marketplace plugin or another supported host setup owns.
Step 3: CLAUDE.md Integration (Optional)
Offer Integrate / Skip / Preview first for adding this ~40-line reference block (pipeline, command table, agent catalog) to the project's CLAUDE.md. On Preview first, show it:
<!-- ooo:START -->
<!-- ooo:VERSION:0.59.9 -->
# Ouroboros — Specification-First AI Development
> Before telling AI what to build, define what should be built.
> As Socrates asked 2,500 years ago — "What do you truly know?"
> Ouroboros turns that question into an evolutionary AI workflow engine.
Most AI coding fails at the input, not the output. Ouroboros fixes this by
**exposing hidden assumptions before any code is written**.
1. **Socratic Clarity** — Question until ambiguity ≤ 0.2
2. **Ontological Precision** — Solve the root problem, not symptoms
3. **Evolutionary Loops** — Each evaluation cycle feeds back into better specs
```
Interview → Seed → Execute → Evaluate
↑ ↓
└─── Evolutionary Loop ─────┘
```
## ooo Commands
Each command loads its agent/MCP on-demand. Details in each skill file.
| Command | Loads |
|---------|-------|
| `ooo` | — |
| `ooo interview` | `ouroboros:socratic-interviewer` |
| `ooo seed` | `ouroboros:seed-architect` |
| `ooo run` | MCP required |
| `ooo evolve` | MCP: `evolve_step` |
| `ooo evaluate` | `ouroboros:evaluator` |
| `ooo unstuck` | `ouroboros:{persona}` |
| `ooo status` | MCP: `session_status` |
| `ooo setup` | — |
| `ooo help` | — |
## Agents
Loaded on-demand — not preloaded.
**Core**: socratic-interviewer, ontologist, seed-architect, evaluator,
advocate, contrarian, judge
**Support**: hacker, simplifier, researcher, architect
<!-- ooo:END -->
On Integrate: back up CLAUDE.md to CLAUDE.md.bak, append the block, and confirm.
Step 4: Verification
Confirm (with the Step 1.5 config show commands) that the saved Ouroboros
config selects the default Claude Agent SDK runtime
on MCP 1.x while ~/.claude/mcp.json was not mutated by this setup. The
dependency-free Claude CLI worker remains a distinct, explicit [claude-cli]
selection for the isolated MCP 2 process.
Step 5: Success Summary
Ouroboros Setup Complete!
Mode: Claude Agent SDK (MCP 1.x)
MCP Server: Host-owned (config not mutated)
CLAUDE.md: <integrated | skipped>
Next: ooo interview "your project idea" | ooo tutorial | ooo help
Step 5.1: Model Choice (Claude Code)
Offer optional model control. Ask "설정이 완료됐어요. 기본 모델 설정으로 바로
시작할 수 있고, 모델은 언제든 나중에 바꿀 수 있어요." (header 모델 설정):
- 바로 시작하기 (권장): 기본 모델 설정으로 바로 작업을 시작해요. Continue to Step 5.5.
- 직접 모델 설정하기: 단계별로 모델을 바꾸거나 목록에 없는 모델 ID를 입력해
고정해요. Read and follow
../config/SKILL.md. In the local Claude Code harness it opens the settings UI in the user's browser at a temporarylocalhostaddress. They can reopen it any time withooo config; this choice never permanently locks a model.
Step 5.5: Brownfield Repository Scan
Register existing git repositories and linked worktrees in the Ouroboros DB so interviews can use them as brownfield context. Use the MCP tool only, not the CLI or Python scripts.
Loading MCP tools (deferred-schema guard)
Ouroboros tools may be deferred, and a schema loaded on one turn can unload by the next. Immediately before each ouroboros_* call, load it with the active runtime's tool-discovery capability: "+ouroboros brownfield" for ouroboros_brownfield (both the scan and the later set_defaults call). A call whose schema is not loaded fails with "Invalid tool parameters". If the tool is already callable, discovery is a no-op and an empty result is expected. If the tool is genuinely absent, skip this step and continue to Step 6.
-
Scan and register:
Tool: ouroboros_brownfield Arguments: { "action": "scan" }The walk starts at
scan_root(default: the user's home directory) and goes at most two directory levels deep, skipping dot-prefixed and noisy directories such asnode_modules. Normal repos (.gitdirectory) and linked worktrees (.gitfile) reached by the walk are registered self-only. Git worktree families are not expanded. Repos with no remote or any remote name are eligible. Existing defaults are preserved. -
The scan response
textalready holds a numbered list with default markers. Do NOT make any additional MCP calls to list or query repos. Show every repo in a plain-text two-column grid inside a code block (not a markdown table), keeping the*default markers exactly as returned:Scan complete. 8 repositories registered. 1. repo-alpha 5. repo-epsilon 2. repo-bravo * 6. repo-foxtrotIf no repos were found, skip to Step 6.
-
Do NOT use the active runtime's
ask_userdialog for this selection. Text emitted between tool calls is not guaranteed to render, so a dialog can appear without the list, and option previews truncate long lists. Instead, end the turn with the repo grid as the final message, followed by the prompt:- Defaults exist:
Current defaults: <names> (numbers <numbers>), thenReply with repo numbers to change defaults (e.g. "6, 18, 19"), "keep" to keep the current defaults, or "none" to clear them. - No defaults:
No defaults set. Reply with repo numbers to set defaults (e.g. "6, 18, 19"), or "none" to run interviews in greenfield mode.
- Defaults exist:
-
On the next turn: numbers (any separator) select those IDs; "keep" skips the MCP call and leaves defaults unchanged; "none" clears all; anything else gets the prompt again (never guess). Then make ONE call that replaces all defaults (
"indices": ""for none):Tool: ouroboros_brownfield Arguments: { "action": "set_defaults", "indices": "<comma-separated IDs>" }Confirm the resulting defaults, or that interviews will run in greenfield mode, and mention that
ooo setuporooo brownfieldchanges them later.
Step 6: First Project Nudge
Suggest ooo interview "Build a CLI tool for [something you need]" or
ooo tutorial.
Uninstall
On --uninstall, ask for confirmation, explaining that it removes only the
Ouroboros block from CLAUDE.md, not plugin files, seeds, or project data. If
confirmed:
- Leave
~/.claude/mcp.jsonuntouched - Remove
<!-- ooo:START -->to<!-- ooo:END -->block from CLAUDE.md (if exists) - Confirm: "Ouroboros plugin configuration removed. To remove plugin files, run: claude plugin uninstall ouroboros"
Setup Troubleshooting
- "No compatible Python found": plugin mode works without a global Python
when uv is on PATH. Otherwise install Python 3.12 (
brew install [email protected],sudo apt install python3.12, or python.org/downloads). - "uvx: command not found": uvx is recommended, not required. For the
default Claude SDK runtime,
pip install 'ouroboros-ai[claude]'([claude-sdk]is an explicit alias). Use[claude-cli]only for the isolated MCP 2 server worker. Never combine[claude],[claude-sdk], or[all]with[mcp].
RFC #1392 State Breadcrumb Footer
End the final response with exactly one line ◆ <current state> → next: <recommended action>. Derive the state from live session state via ouroboros_session_status when available, otherwise from this skill's actual outcome. Never use a linear Step N of M footer; when the next step is a real choice, list 2-3 options.