ORP supports both:
- docs-first template adoption (
PROTOCOL.md, templates), and - optional runtime CLI usage (
orp) for gates/packets/packs.
The default runtime story is now:
orp auth ...,orp ideas ...,orp feature ...,orp world ...,orp checkpoint ..., andorp agent ...for hosted workspace workorp discover ...for profile-based GitHub scanning and opportunity selectionorp collaborate ...for repository collaborationorp erdos ...for Erdos-specific workflowsorp pack ...only when you need advanced/internal template install behavior
Optional global CLI install:
npm i -g open-research-protocol
orp
orp -h
orp about --jsonCLI prerequisites:
- Python 3 on
PATH PyYAML(python3 -m pip install pyyaml)
Agent-friendly discovery surfaces:
- bare
orpfor the CLI home screen with packs, repo status, and quick actions orp home --jsonfor machine-readable landing contextorp auth loginfor hosted workspace loginorp whoami --jsonfor the current hosted identityorp ideas list --jsonfor hosted idea listingorp world bind --idea-id <idea-id> --project-root /abs/path --codex-session-id <session-id> --jsonfor hosted world bindingorp checkpoint queue --idea-id <idea-id> --jsonfor queueing a hosted checkpointorp agent work --once --jsonfor processing one hosted checkpoint joborp discover profile init --jsonfor a portable discovery profile scaffoldorp discover github scan --profile orp.profile.default.json --jsonfor ranked GitHub repo/issue/person recommendationsorp collaborate initfor immediate collaboration scaffoldingorp collaborate workflows --jsonfor built-in collaboration workflow discoveryorp collaborate gates --workflow full_flow --jsonfor the exact gate chainllms.txtfor quick repo/package discoveryorp about --jsonfor machine-readable capabilities, schemas, artifacts, and bundled packsdocs/DISCOVER.mdfor the discovery profile model and scan artifactsdocs/AGENT_LOOP.mdfor the intended agent operating rhythmorp pack list --jsonfor machine-readable bundled pack inventory
Fastest collaboration setup in a fresh repo:
orp collaborate init
orp collaborate workflows --json
orp collaborate run --workflow full_flow --jsonThis uses ORP's built-in collaboration ability. You do not need to think in terms of separate governance packs for the default collaboration path.
Fastest hosted workspace loop from the published ORP binary:
orp auth login
orp whoami --json
orp ideas list --json
orp world bind --idea-id <idea-id> --project-root /abs/path --codex-session-id <session-id> --json
orp checkpoint queue --idea-id <idea-id> --json
orp agent work --once --jsonTo choose where collaboration should start, scaffold a profile first:
orp discover profile init --owner SproutSeeds --owner-type org
orp discover github scan --profile orp.profile.default.json --jsonThis is the fastest way to verify the published CLI before integrating ORP into a real repo.
mkdir test-orp && cd test-orp
npm i -g open-research-protocol
orp init
orp gate run --profile default
orp packet emit --profile default
orp report summary
find orp -maxdepth 3 -type f | sortExpected outcomes:
orp.ymlis created in the working directoryorp/artifacts/<run_id>/RUN.jsonis written aftergate runorp/packets/*.jsonandorp/packets/*.mdare written afterpacket emitorp/artifacts/<run_id>/RUN_SUMMARY.mdis written afterreport summary
Then validate a real public pack flow:
orp pack list
orp pack install --pack-id erdos-open-problems --include catalog
orp --config orp.erdos-catalog-sync.yml gate run --profile erdos_catalog_sync_active
orp report summaryThis exercises the published pack install path plus a real pack-backed gate run.
- Copy the folder into your repo (recommended:
orp/):
mkdir -p /path/to/your/repo/orp
cp -R /path/to/orp/* /path/to/your/repo/orp/- Link it from your repo
README.md(example):
## Protocol
This project follows ORP: `orp/PROTOCOL.md`.-
Edit
orp/PROTOCOL.mdand fill in the Canonical Paths section for your repo. This is required for correctness. -
Start using templates for all claims/verifications:
orp/templates/CLAIM.mdorp/templates/VERIFICATION_RECORD.mdorp/templates/FAILED_TOPIC.md
- Optional (agent users): integrate ORP into your agent’s primary instruction file:
- Read
orp/AGENT_INTEGRATION.md - Or run:
orp/scripts/orp-agent-integrate.sh --sync /path/to/your/agent/instructions.md- Optional checkpoint tool (writes a process-only handoff/compaction log):
orp/scripts/orp-checkpoint.sh --sync --agent-file /path/to/your/agent/instructions.md "checkpoint note"
- Optional checkpoint tool (writes a process-only handoff/compaction log):
- Create a new project directory and copy ORP in:
mkdir -p /path/to/new-project
cp -R /path/to/orp/* /path/to/new-project/-
Rename/edit
README.mdandPROTOCOL.mdfor your project. -
Define canonical paths (paper/code/data/etc) in
PROTOCOL.md. -
Optional (agent users): integrate ORP into your agent’s primary instruction file:
- Read
AGENT_INTEGRATION.md - Or run:
scripts/orp-agent-integrate.sh --sync /path/to/your/agent/instructions.md- Optional checkpoint tool (writes a process-only handoff/compaction log):
scripts/orp-checkpoint.sh --sync --agent-file /path/to/your/agent/instructions.md "checkpoint note"
- Optional checkpoint tool (writes a process-only handoff/compaction log):
If you want a guided copy:
./scripts/orp-init.sh /path/to/your/repo/orpIf you are evaluating ORP for the first time, prefer Option 0 before copying files into a larger repo.
ORP becomes real only when your team adopts the procedure:
- claims must be labeled,
- Exact/Verified claims must have verification hooks,
- disagreements are resolved by verification or downgrade,
- and failures are recorded as first-class artifacts.
There is no automated enforcement unless you add it (CI hooks, PR checks, etc.).