[v0.8] Add mode registry and release_mode plumbing

[Stahl-G]

Parent: #89

Summary

Add a first-class mode registry and make release_mode available to runtime state, policy packs, semantic review, evidence reporting, and release readiness.

This is the foundation for v0.8: the same brief can be acceptable as an internal_draft but blocked as a formal_release_candidate.

Motivation

Current gates mostly answer whether the run is complete and whether existing claims are traceable/fresh/target-relevant. v0.8 needs mode-aware behavior:

• demo should not imply reliance;
• internal draft may allow conditional evidence;
• research review requires stronger evidence and human-visible blockers;
• formal release candidate requires evidence readiness, release readiness, and approval state.

Proposed files

configs/mode_registry.yaml
src/multi_agent_brief/modes/__init__.py
src/multi_agent_brief/modes/registry.py
src/multi_agent_brief/cli/modes_commands.py
tests/test_mode_registry.py

Update as needed:

src/multi_agent_brief/core/schemas.py
src/multi_agent_brief/cli/main.py
src/multi_agent_brief/cli/run_commands.py
src/multi_agent_brief/orchestrator/runtime_state.py
configs/orchestrator_contract.yaml
configs/stage_specs.yaml

Required modes

modes:
  demo:
    semantic_review_required: false
    evidence_required: false
    release_required: false
    human_approval_required: false

  internal_draft:
    semantic_review_required: optional
    evidence_required: conditional
    release_required: false
    human_approval_required: false

  research_review:
    semantic_review_required: true
    evidence_required: true
    release_required: conditional
    human_approval_required: false

  formal_release_candidate:
    semantic_review_required: true
    evidence_required: true
    release_required: true
    human_approval_required: true

  audit_only:
    semantic_review_required: true
    evidence_required: true
    release_required: false

  final_render_only:
    semantic_review_required: false
    evidence_required: false
    release_required: conditional

CLI

Add:

multi-agent-brief modes list --json
multi-agent-brief modes show research_review --json
multi-agent-brief modes validate --json

Config behavior

Support in config.yaml:

release:
  mode: research_review

Default should be internal_draft unless an existing config already indicates demo/synthetic usage.

Runtime behavior

• release_mode should be exposed in runtime manifest / handoff metadata.
• Unknown mode should fail validation with a clear error.
• Mode registry must be read-only repo-scoped config, not a workspace mutation.

Acceptance criteria

• configs/mode_registry.yaml exists and validates.
• PipelineContext or equivalent runtime context carries release_mode.
• CLI can list/show/validate modes.
• config.yaml can set release.mode.
• Unknown mode fails deterministically.
• Runtime handoff / manifest exposes selected mode.
• Tests cover default mode, configured mode, unknown mode, and JSON output.

Non-goals

• Do not implement evidence scoring here.
• Do not implement release readiness here.
• Do not fork schemas per runtime.
• Do not change final delivery behavior yet.