chore(governance): add WG constitution, seats, and decision records (spec-guardian stage 1)

[bokelley]

Summary

Stage 1 of specs/spec-guardian.md: replace the "reviewer that pretends to be Brian" model with an institutional one — the AAO Secretariat, staff for the human AdCP Working Group. This PR creates the shared operating layer every automated desk consumes and re-grounds the existing desks on it. No new automation is introduced; the behavior changes are in what the existing desks cite and how they route decisions.

What's new

• .agents/wg/constitution.md — the operating rules for all review/triage/secretary agents: identity, the WG charter's decision classes made binding for machines (Editorial = autonomous, Normative = memo for human ratification, Breaking = humans always), an explicit escalation list, the spec-invariant review floor, design principles, precedent rules, and the amendment process. Extracted from the Argus prompt, the triage prompt, and the playbook — not invented.
• .agents/wg/seats.md — the expert panel formalized: seat-selection table (unifying Argus's delegation triggers and triage's expert matrix) plus chair rules (synthesize by severity, record dissent, MUST FIX verdicts flow through).
• governance/decisions/ — decision records (DR-NNNN), the WG's institutional memory. README defines the format/lifecycle; nine records backfilled from operative precedent, all refs verified:
  • DR-0001 conformance assertions need a normative basis (PR fix(compliance): assert governance_context echo + gate governance_approved/conditions (#5716) #5719, compliance: align governance_approved/conditions with the two-phase governance token flow (they under-assert today) #5716)
  • DR-0002 feed_format is a seller-side parsing label (Resolve openai_product_feed platform-agnosticism policy in brand.json feed_format enum #3456, Document catalog-driven single-item render pattern for DPA (Meta, Snap Collection, TikTok Shopping) #5277)
  • DR-0003 error codes ship on both enum surfaces (PR feat(schema): manifest.json + structured enumMetadata (closes #3725) #3738, Spec proposal: publish a machine-readable tool/error registry to stop SDK drift #3725)
  • DR-0004 BCP 47 language sets; don't enum an open corpus (media_buy capabilities: add supported_languages[] to language targeting object #5706, media_buy targeting: relax targeting-overlay language to BCP 47 (companion to #5706) #5720)
  • DR-0005 macro substitution is not wire-observable (feat(compliance): macro-identifier substitution storyboard (parked: Live Integration RFC) #5646)
  • DR-0006 v2 maintenance/sunset (3.0: Define v2 maintenance and sunset policy #2220)
  • DR-0007 release cadence + 12-month support window (Publish release cadence: 4.0 early 2027, major releases every 1–2 years #2312, PR docs: publish release cadence policy + tighten red-team FAQ titles (#2312, #2313) #2359)
  • DR-0008 SDK is canon for wire shape under spec ambiguity
  • DR-0009 additionalProperties:true default; tightening is policy-wide

Behavior changes

• .github/ai-review/expert-adcp-reviewer.md — Argus's persona clause ("in the voice of Brian O'Kelley") retired; Argus is now the Secretariat's review desk applying the constitution and citing DR-NNNN precedent. New rule: contradicting a decision record is spec drift; Breaking-class changes are never auto-approved even with a correct major changeset.
• .github/workflows/ai-review.yml — the build-prompt step appends the constitution from the base SHA (same prompt-injection posture as the reviewer prompt; fail-soft when absent at base, e.g. maintenance lines). Bash block extracted and bash -n validated; YAML parse validated.
• .agents/routines/triage-prompt.md — the Flag outcome now routes by decision class: Normative + expert consensus → decision memo in record format + needs-wg-review, no maintainer ping; Breaking/security-sensitive/governance-document/expert-disagreement → @bokelley, your call as before. Controlling decision records short-circuit re-litigation.
• .agents/playbook.md — pointer to .agents/wg/ and governance/decisions/ under Cross-Agent Integration.
• specs/spec-guardian.md — full architecture (committed for reference), including the personification decision: GitHub identity = "AAO Secretariat" bot (new App preferred over renaming aao-release-bot; App creation + ARGUS_BOT_LOGIN/secret swap is an org-admin follow-up).

Not in this PR (next stages)

Shadow-mode Normative ratification tracking, the 3.2 scope-gate pass (102 open issues), Addie's secretary tools (GitHub comment/label, info-request routing, weekly WG report), and the release-train jobs.

Testing

• Precommit suite (unit, dynamic-imports, callapi-state-change, server-unit, typecheck) passed on commit.
• node scripts/check-changeset-protocol-scope.cjs origin/main — passed (policy/governance-only change, no changeset by design).
• node scripts/check-pr-title.cjs — passed.
• ai-review.yml: YAML parse + extracted bash -n of the modified step.

Touches sensitive paths (.agents/*, .github/workflows/*) — Argus will abstain from approving; human review required, which is the intended posture for constitution changes.

🤖 Generated with Claude Code

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
adcp	🟢 Ready	View Preview	Jul 3, 2026, 5:05 AM

💡 Tip: Enable Workflows to automatically generate PRs for you.

[aao-release-bot]

Argus is not auto-reviewing this PR because it modifies protected paths that require human review (.agents/playbook.md, .agents/routines/triage-prompt.md, .agents/wg/constitution.md, .agents/wg/seats.md, .github/ai-review/expert-adcp-reviewer.md, .github/workflows/ai-review.yml). A human reviewer should review and merge this PR; Argus will resume on subsequent PRs once these changes land on main.

[aao-release-bot]

Argus is not auto-reviewing this PR because it modifies protected paths that require human review (.agents/playbook.md, .agents/routines/triage-prompt.md, .agents/wg/constitution.md, .agents/wg/seats.md, .github/ai-review/expert-adcp-reviewer.md, .github/workflows/ai-review.yml). A human reviewer should review and merge this PR; Argus will resume on subsequent PRs once these changes land on main.

[aao-release-bot]

Argus is not auto-reviewing this PR because it modifies protected paths that require human review (.agents/playbook.md, .agents/routines/triage-prompt.md, .agents/wg/constitution.md, .agents/wg/seats.md, .github/ai-review/expert-adcp-reviewer.md, .github/workflows/ai-review.yml). A human reviewer should review and merge this PR; Argus will resume on subsequent PRs once these changes land on main.