doctrine(agent-orchestra): v3+v4 expansion — layered actor identity + public claim intake + identity binding (Aaron + Amara 2026-04-29)

memory/MEMORY.md

@@ -3,7 +3,7 @@
 **📌 Fast path: read `CURRENT-aaron.md` and `CURRENT-amara.md` first.** <!-- latest-paired-edit: fork-audit R/C/T diff-filter coverage + plumbing-vs-porcelain note (2026-04-29 round-10 Amara). NOTE: this comment is a single-slot "latest paired edit" marker (not a paired-edit log). Per the round-10 Amara framing the slot semantics are now explicit. -->
 
 - [**Otto-362 — Doctrine memory expansion refreshes stale statements in the SAME edit (2026-04-29)**](feedback_otto_362_doctrine_memory_expansion_refresh_stale_statements_same_edit_2026_04_29.md) — When a memory file gets expanded with a new section that supersedes earlier statements in the same file, refresh the now-stale statements in the same edit, not a follow-up tick. Internal contradictions within one file are lying-by-omission. Pattern observed across 4 same-day doctrine PRs (#850/#851/#852/#853) where multi-AI review caught the drift instead of pre-push self-audit (10+ Copilot P1 + Codex P2 threads, all stale-statement class). Composes with same-tick CURRENT-update discipline (intra-file generalisation), verify-before-deferring, future-self-not-bound. Editing discipline, not lint — semantic contradictions can't be mechanised.
-- [**Zeta Agent Orchestra — capability + role + claim + isolation (Aaron + Amara, 2026-04-29)**](feedback_zeta_agent_orchestra_capability_role_claim_isolation_aaron_amara_2026_04_29.md) — Project-level multi-harness multi-maintainer multi-actor coordination model. *"Humans own intent. Harnesses run actors. Roles define authority. Claims bind work. GitHub coordinates now. Git preserves forever."* Stop classifying agents by name (subagent vs CLI vs buddy) — classify by capability (review-only / patch / write-worktree / push / merge / authority). Pinned vs free vs buddy roles. GitHub-native live coordination + git-native durable substrate, both must agree. Cross-harness memory rule: one canonical substrate (`memory/`, `docs/active-trajectory.md`, `docs/ops/**`, `docs/best-practices/**` `[planned]`); many thin bootstrap adapters (CLAUDE.md, AGENTS.md, GEMINI.md `[planned]`, .cursor/rules/ `[planned]`) — adapters point to memory, never duplicate it. Paced protocol — land design first, dry-run, then implementation. Composes with parallel-agent-worktree-isolation + best-practices-evidence-lineage rules landed same day.
+- [**Zeta Agent Orchestra — capability + role + claim + isolation (Aaron + Amara, 2026-04-29; v2/v3/v4)**](feedback_zeta_agent_orchestra_capability_role_claim_isolation_aaron_amara_2026_04_29.md) — Project-level multi-harness multi-maintainer multi-actor coordination model. *"Humans own intent. Harnesses run actors. Roles define authority. Claims bind work. GitHub coordinates now. Git preserves forever."* Stop classifying agents by name (subagent vs CLI vs buddy) — classify by capability (review_only / patch_only / write_worktree / push_branch / open_pr / merge_pr / authority_mutation). Pinned vs free vs buddy roles. GitHub-native live coordination + git-native durable substrate, both must agree. Cross-harness memory rule: one canonical substrate (`memory/`, `docs/active-trajectory.md`, `docs/ops/**`, `docs/best-practices/**` `[planned]`); many thin bootstrap adapters (CLAUDE.md, AGENTS.md, GEMINI.md `[planned]`, .cursor/rules/ `[planned]`) — adapters point to memory, never duplicate it. v3 additions: layered actor identity (`maintainer_id / host_id / harness_id / role_id / actor_id / session_id` — "Mac/Windows = host IDs, not agent IDs") + public claim intake layer (Claim Request ≠ Active Claim; CONTRIBUTING.md + AGENTS.md autonomous-agent block both `[planned]` content additions inside existing files, `.github/ISSUE_TEMPLATE/claim_request.yml` `[planned]` new file, reconciler tool `[planned]`, safety levels E0-E5, drift discipline synced/stale/drift/failed/pending). v4 corrections (Deepseek+Gemini+Ani+Alexa+Claude.ai → Amara synthesis): identity needs binding (`actors/<encoded-actor-id>.yaml` registry + signed commits + AgencySignature v2 schema additions Trust-Domain/Actor/Signed-By; integration writeup `[planned]` at `docs/aurora/2026-04-29-agencysignature-layered-actor-identity-integration-writeup-for-amara.md`, landing in PR #853); trust-domain prefix (zeta:// / zeta-system:// / zeta-external://); capabilities-as-primitive (roles become named bundles); reconciler-is-itself-an-actor (`zeta-system://github-actions/reconciler`; no privilege elevation from git mirror to GitHub issue); add `rejected` claim state distinct from `revoked`; auto-expire claim requests; DoS protection + prompt-injection defense for public intake; harness pre-action freshness check (not just CI PR-time); allowlist-first paths (fail-closed). v4 rollout reorder: identity → capabilities → claims → reconciler → public intake → dry run (NOT public-intake-first). Paced protocol — land doctrine first, dry-run, then implementation. Composes with parallel-agent-worktree-isolation + best-practices-evidence-lineage rules landed same day.
 - [**Best practices = evidence + human lineage + Zeta-native + enforcement + teaching (Aaron + Amara, 2026-04-29)**](feedback_best_practices_evidence_lineage_survival_substrate_aaron_amara_2026_04_29.md) — *"Best practices are not files to copy. They are evidence-backed decisions with human lineage, Zeta-native interpretation, enforcement, and teaching value."* Idiomatic ≠ best-practice (orthogonal axes — want both). Survival framing: future humans/agents must repair the factory without original authors. Six-question audit when Zeta touches any tool/language/domain. Standard schema for each best-practice entry (claim/status/scope/idiomatic-axis/best-practice-axis/evidence/human-lineage/Zeta-interpretation/enforcement/examples/exceptions/revisit-cadence). Supersedes the earlier "copy SQLSharp/scratch configs" misreading.
 - [**Parallel agents need isolated worktrees — coordinator owns main (Aaron + Amara, 2026-04-29)**](feedback_parallel_agents_need_isolated_worktrees_coordinator_owns_main_aaron_amara_2026_04_29.md) — Hard rule: when dispatching multiple background subagents in parallel, each needs an ISOLATED `git worktree` (not the coordinator's working tree). Sharing one working tree causes branch-switch collisions, stash confusion, orphan files, out-of-scope formatter side-effects. Mechanism (not vigilance): coordinator must allocate worktrees BEFORE allocating agents. *"Parallel agents may inspect broadly, but mutate narrowly."* Caught during 2026-04-29 incident with three subagents in same checkout.
 - [**LFG-only development flow — AceHack is a daily mirror (Aaron + Amara, 2026-04-29)**](feedback_lfg_only_development_flow_acehack_is_mirror_aaron_amara_2026_04_29.md) — Topology simplification 2026-04-29. Double-hop / AceHack-first / fork-data doctrine paused. *"LFG is the factory. AceHack is the mirror."* All PRs / issues / anchors / backlog live on LFG only; AceHack/main is a daily mirror. Existing AceHack archives stay as historical evidence; no new fork-data architecture. Unfreeze condition: AceHack PRs become real again. Supersedes the prior AceHack-first / double-hop topology adopted 2026-04-27.