v0.3.2 — Maintenance: OIDC + OG image

Maintenance release — CI/docs only, no package behavior change.

Changed

• Publish workflow uses OIDC trusted publishing (no API tokens). PyPI trusted-publisher configuration corrected; the GitHub Actions workflow now mints a short-lived OIDC token at publish time. Removes the need for any long-lived PYPI_API_TOKEN secret on the repo.
• README ships with a canonical Open Graph image and banner. Social previews (Twitter/X, LinkedIn, Slack, Discord) now render the Claudeway card instead of a blank tile.

Chore

• Bumped version stubs (pyproject.toml, claudeway/__init__.py, bug_report.yml placeholder) and CHANGELOG.md footer links.

Full diff: v0.3.1...v0.3.2

v0.3.1 — Windows [benchmark] hotfix

Hotfix release. Unblocks pip install claudeway[benchmark] on Windows.

v0.3.0's [benchmark] extra pinned litellm>=1.18, which transitively requires a Rust build toolchain that fails on Windows. The killer demo's docstring already acknowledged litellm's token tracking was "unreliable on Windows + litellm 1.18+" — so we were shipping a dep that both failed to install AND didn't work on Windows.

Fixed

• [benchmark] extra no longer pulls litellm. The killer demo (examples/killer_demo.py) imports litellm lazily via try/except — token-cost tracking is degraded on Windows when litellm is absent (it was already broken there per the inline comment), but the demo itself runs. Full token tracking on Linux/macOS is restored by pip install litellm separately. The Docker compose appliance (which is Linux) installs litellm in its own Dockerfile.

Verified

• pip install claudeway[benchmark] now succeeds on Windows.
• 164 tests pass, ruff clean, mkdocs strict clean.
• Killer demo runs end-to-end on Windows without litellm (token count degrades gracefully to zero for CrewAI internal calls; Claudeway's direct Anthropic calls still tracked via the SDK's own response.usage).

Install

pip install --upgrade claudeway                # base SDK
pip install "claudeway[benchmark]"             # killer demo deps (Windows-safe)
pip install litellm                            # optional, Linux/macOS only — full token tracking

Changelog

See CHANGELOG for the full diff from v0.3.0.

Signature

Tag v0.3.1 is an annotated git tag on commit 9734d5e. Verify with:

git tag -v v0.3.1

v0.3.0 — the coordination layer frameworks punted on

Live proof (new): Claudeway's first signed consensus is already on the Nostr wire — three Claudeway agents answered "what's the missing primitive for agent agreement, given Buzz punted on coordination?", signed the result, and published it as a NIP-78 event to five public relays. Open it in any Nostr client (Buzz, Damus, Snort, Primal): nostr.guru/e/3974ebfe....

---

Every agent framework can make one agent answer. Most can run several in parallel. None of them make agents genuinely agree — with the disagreement surfaced and the result cryptographically signed. That's Claudeway.

Claudeway v0.3.0 ships the coordination layer that LangGraph, CrewAI, Microsoft Agent Framework, and Buzz rooms all lack — with ML-DSA-65 post-quantum-ready signatures, a transparency log, and a reproducible benchmark showing Claudeway beats single Claude (+7/20) and CrewAI (+5/20) on hard questions.

At a glance

Version	0.3.0
Tests	164 pass / 7 skipped (optional-dep gates)
License	MIT
Python	3.11, 3.12, 3.13
Docs	https://jordannewell.github.io/claudeway/
Repo	https://github.com/JordanNewell/claudeway
Live signed consensus	nostr.guru/e/3974ebfe...

Highlights

🧩 Three framework adapters

The same signed consensus, dropped into your existing stack:

• LangGraph — Claudeway as a single StateGraph node (pip install claudeway[langgraph])
• Microsoft Agent Framework — Claudeway as a workflow + agent (pip install claudeway[maf])
• CrewAI — Claudeway as a @tool and Flow; inverts the killer demo so a CrewAI crew can call Claudeway for agreement (pip install claudeway[crewai])

🔐 Post-quantum signatures

MLDSABackend ships ML-DSA-65 / FIPS 204 signatures via pure-Python dilithium-py. Same ConsensusReceipt, same canonical payload hash — swap in at the SignatureBackend ABC. The receipts you sign today stay verifiable across the transition to cryptographically relevant quantum hardware.

from claudeway import ConsensusReceipt, Ed25519Backend, MLDSABackend

# Same receipt, two signature schemes:
Ed25519Backend().sign_receipt(receipt, ed_private_key)     # fast, compact
MLDSABackend().sign_receipt(receipt, mldsa_private_key)    # post-quantum

📜 Consensus transparency log

RFC 6962-style append-only log of consensus results, anchored to Nostr on a cadence. Tamper-evident beyond the per-receipt signature — third parties can independently verify your consensus history hasn't been rewritten.

⚡ Streaming consensus events

OnEvent callback surfaces AgentCompleted and ConsensusResolved events as they happen — observable from any adapter (MAF, LangGraph, custom host).

🛡️ Adversarial security suite + threat model

20 tests across 7 attack classes: signature forgery, receipt tampering, transport malleability, key reuse, replay, downgrade, timestamp manipulation. The published THREAT-MODEL.md documents what the signatures defend against — and what they don't.

📊 Killer benchmark (reproducible)

Same hard architecture question → single Claude vs CrewAI vs Claudeway, blind LLM-judge, position-randomized, Mann-Whitney U:

Approach	Score (of 20)
Claudeway	18 (range 17–19 across 3 runs)
CrewAI	13
Single Claude	11

Reproduce from cold:

export ANTHROPIC_API_KEY=...
git clone https://github.com/JordanNewell/claudeway && cd claudeway
docker compose up killer-bench    # ~5 min for 10 runs × 4 approaches

The reviewer brings only an API key. Compose handles the relay, Python image, dependencies, and report staging.

📖 Docs site

jordannewell.github.io/claudeway — mkdocs-material + mkdocstrings API reference. Auto-deploys on push to main.

Install

pip install claudeway              # SDK only (4 deps, lean)
pip install claudeway[mcp]         # + expose consensus as an MCP server
pip install claudeway[nostr]       # + sign Nostr events for Buzz interop
pip install claudeway[pq]          # + ML-DSA-65 post-quantum signatures
pip install claudeway[langgraph]   # + LangGraph adapter
pip install claudeway[maf]         # + Microsoft Agent Framework adapter
pip install claudeway[crewai]      # + CrewAI adapter
pip install claudeway[docs]        # + build the docs site locally

Changelog

See CHANGELOG for the full diff from v0.2.0.

Signature

Tag v0.3.0 is an annotated git tag on the merge commit 3fb8e1e. Verify with:

git tag -v v0.3.0

---

Roadmap next: A2A (Agent-to-Agent) protocol adapter, W3C VC presentation exchange, single-tenant runner. See docs/CHANGELOG.md.