Add evidence-aware cross-thread protocol and seven-agent council

[safal207]

What changed

• adds a vendor-neutral CrossThreadEvent v0.1 contract on top of the existing LS Agent Trust Runtime;
• separates message delivery, evidence verification, state acceptance, and action authorization;
• adds capability grants, consent, expiry, revocation, archive/resume behavior, stale-state rejection, idempotency, and hash-chained audit receipts;
• implements a seven-agent council: Idea, Customer, Consumer, Designer, Executor, Stabilizer, and Innovator;
• adds JSON Schema, examples, a validator CLI, a deterministic 10-case conformance suite, documentation, and GitHub Actions CI;
• includes a reproducible update packet for Follow-up: typed, evidence-aware CrossThreadEvent contract for safe peer-thread orchestration openai/codex#36843.

Why

Cross-thread messaging has shipped in Codex, but transport success must not be confused with verified state or execution authority. This PR provides a narrow, advisory-only reference implementation and conformance fixture for that missing trust boundary.

Local validation

• 27 passed
• reference conformance suite: 10/10
• seven-agent council: 7/7 accepted stages
• trust ledger and cross-thread audit chains verified

Safety boundary

The implementation does not execute merges, deployments, payments, messages, permission changes, or other external effects. Action requests remain advisory and continue through the existing human approval and sandbox path.

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 1b715338-1d80-418f-a058-40e3609d3863

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

LS multi-model PR review

• Exact head: bfc8188de0902e3d2b7dee8901d541f54b3240fc
• Base: 0090938dedd5c477cd95ddeafcdbaa1f387b87de
• Provider: openrouter
• Status: PARTIAL
• Aggregate verdict: COMMENT
• Mode: advisory
• High-risk route: true
• Diff truncated: true
• Files represented in bounded evidence: 15/24
• Policy would block: true

Model executions

Role	Model	Status	Verdict
architecture_and_governance_reviewer	nvidia/nemotron-3-ultra-550b-a55b:free	INVALID	-
fast_diff_reviewer	cohere/north-mini-code:free	INVALID	-
deep_implementation_reviewer	poolside/laguna-xs-2.1:free	INVALID	-

Confirmed findings

No finding reached independent two-model confirmation.

Candidate findings

No structured candidate finding was produced.

Incomplete lanes

• diff\_coverage: {'key': 'diff_coverage', 'reason': 'bounded diff was truncated; the review cannot claim complete PR coverage', 'reviewed_files': ['.github/workflows/cross-thread-protocol.yml', 'prototypes/openai-agent-trust-runtime/AGENTS.md', 'prototypes/openai-agent-trust-runtime/CONFORMANCE.md', 'prototypes/openai-agent-trust-runtime/OPENAI_ISSUE_36843_UPDATE.md', 'prototypes/openai-agent-trust-runtime/docs/cross-thread-protocol.md', 'prototypes/openai-agent-trust-runtime/docs/seven-agent-council.md', 'prototypes/openai-agent-trust-runtime/examples/action-request.json', 'prototypes/openai-agent-trust-runtime/examples/stale-event.json', 'prototypes/openai-agent-trust-runtime/examples/unverified-state.json', 'prototypes/openai-agent-trust-runtime/examples/verified-result.json', 'prototypes/openai-agent-trust-runtime/pyproject.toml', 'prototypes/openai-agent-trust-runtime/schema/cross-thread-event-v0.1.schema.json', 'prototypes/openai-agent-trust-runtime/src/ls_agent_trust/__init__.py', 'prototypes/openai-agent-trust-runtime/src/ls_agent_trust/conformance.py', 'prototypes/openai-agent-trust-runtime/src/ls_agent_trust/council.py'], 'omitted_files': ['prototypes/openai-agent-trust-runtime/src/ls_agent_trust/cross_thread.py', 'prototypes/openai-agent-trust-runtime/src/ls_agent_trust/cross_thread_demo.py', 'prototypes/openai-agent-trust-runtime/src/ls_agent_trust/schemas/__init__.py', 'prototypes/openai-agent-trust-runtime/src/ls_agent_trust/schemas/cross-thread-event-v0.1.schema.json', 'prototypes/openai-agent-trust-runtime/src/ls_agent_trust/validator.py', 'prototypes/openai-agent-trust-runt
• independent\_challenger: {'key': 'independent_challenger', 'requested_model': 'hy3', 'candidates': ['tencent/hy3:free', 'poolside/laguna-xs-2.1:free', 'cohere/north-mini-code:free'], 'reserved_candidates': []}

Authority boundary

This output is evidence for human review. It cannot approve or merge the PR, and a single-model finding remains a candidate rather than a gate decision.