CLAUDE Guide Summary

This page distills the operational guidance from CLAUDE.md into a maintainer-facing checklist.

What CLAUDE.md Is Doing

CLAUDE.md is not just a project introduction. It acts as a maintainer/operator playbook for:

• fast repo orientation
• common runtime commands
• architectural expectations
• quality vocabulary
• collaboration patterns across codex and claude

Maintainer Priorities

1. Keep contract truth upstream

When behavior changes, maintainers should prefer fixing:

1. content/standards/
2. content/roles/ or content/skills/
3. content/templates/
4. tooling/pipelines/, content/workflow/workflows/, or content/distribution/plugins.yaml
5. packages/python-qiongli/src/qiongli/
6. generated qiongli-workflow/ only through materialization

2. Treat workflows as entry UX, not truth

Slash commands are convenient entrypoints. Artifact truth, routing truth, and task truth still live in content/standards/.

3. Use the repo through stable commands

Common commands pulled from CLAUDE.md:

python3 -m bridges.orchestrator doctor --cwd .
python3 -m bridges.orchestrator task-run --task-id F3 --paper-type empirical --topic my-topic --cwd .
python3 scripts/validate_research_standard.py --strict
python3 -m unittest tests.test_orchestrator_workflows -v

4. Think in collaboration modes

The maintainer mental model is:

• single-agent execution for narrow/debug flows
• draft/review fallback for standard task execution
• triad review when independent audit matters
• role split or parallel fanout when work can be decomposed

When To Open The Original CLAUDE.md

Use the original file when you need:

• raw command examples
• the long-form workflow descriptions
• the full project-specific terminology block
• the exact examples for collaboration and stage usage

For most day-to-day navigation, this summary plus Architecture and Conventions should be enough.