claude-org-runtime 0.1.18

Python runtime for claude-org-ja: dispatcher, state schema, and reference role prompts.

claude-org-runtime

Python runtime extracted from claude-org-ja for Claude Code orchestrator setups using renga panes. Sibling of core-harness; provides a dispatcher runner, state schema, and reference role prompts (secretary / dispatcher / curator).

Not an AI agent framework — this is plumbing for human-driven Claude Code orchestrator workflows, not an autonomous-agent toolkit.

Status

0.1.0: dispatcher runner + settings generator + reference role prompts + state schema + v1->v2 polymorphic migrate. This is the first release that consumers (notably claude-org-ja) can pip install to replace their in-tree tools/dispatcher_runner.py / tools/generate_worker_settings.py / tools/role_configs_schema.json.

Install

pip install claude-org-runtime==0.1.0

For local development:

git clone https://github.com/suisya-systems/claude-org-runtime
cd claude-org-runtime
python -m pip install -e .[dev]

Quick start

# Render a per-role settings.local.json:
claude-org-runtime settings generate \
    --role default \
    --worker-dir /path/to/worker \
    --claude-org-path /path/to/claude-org \
    --out /path/to/worker/.claude/settings.local.json

# Compute a Dispatcher delegation action plan:
claude-org-runtime dispatcher delegate-plan \
    --task-json .state/dispatcher/inbox/<task_id>.json \
    --panes-json panes.json \
    --state-dir .state

See docs/cli.md for the full CLI reference and the migration recipe for claude-org-ja consumers.

Reference role prompts

The runtime ships English reference prompts for the three roles used in the claude-org-ja reference organization (secretary, dispatcher, curator):

from claude_org_runtime.prompts import load, load_meta

prompt = load("dispatcher")          # raw markdown, frontmatter included
meta = load_meta("dispatcher")       # {'role': 'dispatcher', 'source': ..., 'status': ...}

These are reference, not prescriptive. They capture one working configuration from claude-org-ja, not an "agent framework" opinion. Consumers are expected to load them as a starting point and then override or rewrite sections from their own project-root CLAUDE.md (or skill files) to match their organization's conventions, terminology, and slash commands.

Related

• core-harness — sibling repo with reusable safety primitives for Claude Code harnesses.
• claude-org-ja#129 — tracking issue for the extraction effort that produced this runtime.
• Issues for this repo.

License

MIT — see LICENSE.