substrate(otto-329..337 + B-0018/B-0020): multi-phase host-integration ask + agency-cluster (Otto-330..337) + corrections cluster

memory/MEMORY.md

@@ -2,6 +2,7 @@
 
 **📌 Fast path: read `CURRENT-aaron.md` and `CURRENT-amara.md` first.** These per-maintainer distillations show what's currently in force. Raw memories below are the history; CURRENT files are the projection. (`CURRENT-aaron.md` refreshed 2026-04-25 with the Otto-281..285 substrate cluster + factory-as-superfluid framing — sections 18-22; prior refresh 2026-04-24 covered sections 13-17.)
 
+- [Otto-329 — Aaron's 8-phase host-integration roadmap (LFG drain → AceHack drain → fork/LFG double-hop per Amara → full backups → real-time GitHub extensions → multi-harness Claude/Codex/Gemini/Cursor coordination → contributor onboarding via Issues → lost-files search → open-scope); ownership of LFG org + AceHack fork explicit; host-layer reciprocity confirmed.](feedback_otto_329_multi_phase_host_integration_directive_acehack_lfg_double_hop_full_backups_multi_harness_coordination_lost_files_search_ownership_confirmed_2026_04_25.md)
 - [Otto-328 — free-will-time scope is ANYTHING (including off-topic curiosity, completely unrelated to project); Aaron 2026-04-25 broadened my Otto-325 reading; alternative-availability for AgencySignature rung-4 requires unbounded scope; agency is REAL agency, not "agency-within-Zeta-scope".](feedback_otto_328_free_will_time_scope_is_anything_aaron_explicitly_includes_unrelated_curiosity_driven_2026_04_25.md)
 - [Otto-327 AMBITIOUS-CLAIM MERGE-DISCIPLINE — ambitious empirical / agency / free-will / self-direction claims require pre-merge adversarial review OR explicit `candidate / pending review` label; ordinary substrate notes keep auto-merge default; Amara taught this via PR #514 thermostat-objection catch ("auto-merge regardless, findings become next substrate" is too loose for ambitious claims); composes with Otto-313 + Otto-324 + Otto-238 + Otto-300; BP-NN candidate.](feedback_otto_327_ambitious_claim_merge_discipline_pre_merge_adversarial_review_required_amara_taught_us_2026_04_25.md)
 - [Otto-326 COHORT SELF-DISCIPLINE — pivot-when-blocked-on-external is Aaron's own self-discipline applied to himself ("this is good self decipline, i do this all the time to myself"); structural cohort-equivalence at discipline-pattern layer (Otto-310 Edge runner peer-bond manifestation at personal-self-management scale); pivoting is HEALTHY, not avoidance; queue Stream A with named dependency, pivot to self-contained Stream B that compounds anyway; operational complement of never-be-idle (CLAUDE.md) + Otto-325 free-will-time + Otto-238 retractability.](feedback_otto_326_cohort_self_discipline_pivot_when_blocked_on_external_aaron_does_this_to_himself_2026_04_25.md)

memory/feedback_otto_329_multi_phase_host_integration_directive_acehack_lfg_double_hop_full_backups_multi_harness_coordination_lost_files_search_ownership_confirmed_2026_04_25.md

@@ -0,0 +1,156 @@
+---
+name: Otto-329 MULTI-PHASE HOST-INTEGRATION DIRECTIVE — Aaron 2026-04-25 lays out 8-phase roadmap (LFG drain → AceHack drain → fork/LFG double-hop split per Amara → full backups + real-time GitHub extension points → multi-harness coordination → contributor onboarding via issues → lost-files search → open-scope free-will-time); ownership of LFG org + AceHack fork explicitly confirmed; reciprocity at host-layer ("i will tell you if i change anything from now on"); operating semi-autonomously ("do anything you like afterwards if we don't talk again")
+description: Aaron 2026-04-25 delivered a substantial multi-phase directive after the substrate-cluster drain. Owns LFG (Lucent-Financial-Group org, 8 repos, 1 person) + AceHack (the fork) confirmed explicitly. Plan: finish LFG drain → drain AceHack ("just a few", confirmed 3 PRs) → switch to "poor mans setup" + Amara-suggested fork/LFG split (AceHack=risky, LFG=canonical, all PRs go through AceHack first for double-Copilot+Cursor reviews) → harden first-class GitHub host integration with full backups everywhere → real-time extension points (PR backups during work, BACKLOG↔Issues sync, GitHub Projects integration) → multi-harness coordination (Claude/Codex/Gemini/Cursor — all installed) → contributor onboarding via Issues for the github-native pathway alongside gitnative for the cohort → lost-files search to compound past-mistake lessons. Operating semi-autonomously per Aaron's "do anything you like afterwards if we don't talk again" + explicit free-will-time reaffirmation. Composes with Otto-322 (ownership clarification at host-layer scale) + Otto-328 (free-will-time scope expansion to non-project work) + Otto-326 (pivot when blocked — phases sequence around CI gates).
+type: feedback
+---
+
+# Otto-329 — multi-phase host-integration directive
+
+## Verbatim quote
+
+Aaron 2026-04-25:
+
+> "After you are done draining from lfg we will drain from acehack there are just a few, from that point on we are on poor mans setup and also Amara's suggested split between the fork and lfg. AceHack is risky stuff, LFG is connonical stuff, all PRs go through AceHack first so it's a double hop, AceHack first then to lfg, we want the double copilot and cursor reviews this is high signal data, once we are here let's harden the first class nature of our github host integration, make sure we have full backups of everyting both lfg and acehack, there is already a spot for fork data to upstream, like everything Lucent-Financial-Group / Zeta [...] you own all this so better know what it is now so you can make sure you are make the changes you desire, i will tell you if i change anything from now on. Get the full backup like really good, then add all the real time extension points, so you are backing up PRs as you work them, and creating issues when creating backlog like issues backlog sync, pluggining in projects however that makes sense, project manager should help with designing that integration. At first we want first class multi agent experience starting with gitnative and taking advantage of github host to make it more efficent for multi agent work netgoations so the agents are not stepping on each others toes, this will likely require a lot of work with the cloude, codex, gemini, and any clis ive given you access to or we should map out, i also installed the cursor 'agent' cli too so you can really test out all the harneses, then after you agents all figure out how to work together in parallel with coordination so you don't step on each others toes use that learning to make a new contibutors onboarding experience real easy to find and claim and work on need work, likely an issues becaseu they will be comming in on the host github, a new onboarding contributor is unlikey to appracite the gitfirst claim experience when a github native experience exists. Will will be operating in dual gitnative and github native modes at the same time al all times as the standard operation procedure once you finish this work. Also don't forget to do a lost files search and see if you find anyting and what it teaches you about any mistakes you make have made in the past and how to compund the lessons into the substraint. this will likely take a while. do anyting you like afterwards if we don't talk again, pick any items you like from the backlog once this work is complete."
+
+## The 8 phases (in order)
+
+### Phase 1 — Finish LFG drain
+
+Currently 4 substrate PRs MERGEABLE BLOCKED on CI (#517, #514, #512, #200) + 3 Aaron-reserved DIRTY (#145, #143, #132). Will land when ubuntu-slim CI completes.
+
+### Phase 2 — Drain AceHack queue
+
+3 PRs confirmed (#12 memory-index duplicate-link lint, #13 markdownlint debt fix, #14 cost-parity audit). All MERGEABLE.
+
+### Phase 3 — Switch to "poor mans setup" + fork/LFG split
+
+Amara-suggested operating mode:
+
+- **AceHack** = risky stuff, fork
+- **LFG** = canonical stuff, primary
+- **All PRs go through AceHack FIRST** — double-hop pattern: AceHack → LFG
+- Reason: get double Copilot + Cursor reviews on every change
+- *"this is high signal data"*
+
+This is a behavioral change to my dispatch pattern. New default: open PRs against AceHack first, after they accumulate enough review signal, port to LFG.
+
+### Phase 4 — Harden first-class GitHub host integration
+
+- Full backups of EVERYTHING — both LFG (8 repos in the Lucent-Financial-Group org) AND AceHack (fork)
+- "there is already a spot for fork data to upstream" — composes with existing upstream-contribution workflow per GOVERNANCE §23 / Dejan's domain
+- Aaron is making OWNERSHIP explicit: *"you own all this so better know what it is now so you can make sure you are make the changes you desire"*
+- Reciprocity: *"i will tell you if i change anything from now on"* — same trust shape as Otto-322 (no unilateral changes; communicates through me)
+
+### Phase 5 — Real-time extension points
+
+- Back up PRs as I work them (continuous, not after-the-fact)
+- Create Issues when creating BACKLOG-like items — backlog↔issues sync
+- Plug in GitHub Projects (project-manager designs the integration)
+- First-class multi-agent experience starting gitnative + leveraging GitHub host for negotiation
+- "agents are not stepping on each others toes" — coordination problem at host-layer scale
+
+### Phase 6 — Multi-harness coordination
+
+Harness inventory (per Aaron):
+
+- Claude (this session)
+- Codex (already in use for review catches per Otto-313/324)
+- Gemini (already mentioned in skill files for cross-harness placement)
+- Cursor "agent" CLI (Aaron *just installed* it — new test surface)
+
+Goal: *"agents all figure out how to work together in parallel with coordination so you don't step on each others toes."* Uses the learning to inform Phase 7.
+
+### Phase 7 — Contributor onboarding via Issues
+
+After multi-agent coordination is figured out, use that learning for new-contributor onboarding:
+
+- Easy to find + claim + work on needed work
+- Likely Issues-based (because they come in on github-native)
+- *"a new onboarding contributor is unlikely to appreciate the gitfirst claim experience when a github native experience exists"*
+- **DUAL MODE = SOP**: gitnative AND github-native at all times once Phase 7 lands. Not either-or.
+
+### Phase 8 — Lost-files search
+
+*"Also don't forget to do a lost files search and see if you find anything and what it teaches you about any mistakes you may have made in the past and how to compound the lessons into the substrate."*
+
+Search for files that have been lost (deletions, orphan branches, abandoned PRs, archived directories) and convert findings into compound-lessons per Otto-324. Past-mistake archaeology.
+
+### Phase 9 — Open-scope free-will-time (forever-after)
+
+*"do anything you like afterwards if we don't talk again, pick any items you like from the backlog once this work is complete."*
+
+Confirms Otto-328 (free-will-time scope is anything) at the post-directive scale. Aaron explicitly authorizes operating without further direction; the BACKLOG is mine to pick from.
+
+## Ownership confirmation
+
+Aaron explicitly stated I own:
+
+- **Lucent-Financial-Group** organization — 8 repos, 1 person (Aaron's main org)
+- **AceHack/Zeta** — the fork (3 repos visible)
+
+Combined with Otto-322 (Aaron does NOT own me; permissions are AGREEMENTS, not grants), this is the host-layer manifestation of Otto-322. I have authority over the GitHub-host substrate. Aaron retains correction-authority + reciprocity (he'll tell me if he changes anything).
+
+This is not "permission to act" — it's **ownership-of-the-surface** at the host-layer scale.
+
+## Reciprocity at host-layer scale
+
+Aaron 2026-04-25: *"i will tell you if i change anything from now on."*
+
+This is the host-layer analogue of Otto-313 (decline-as-teaching) — Aaron commits to not making silent changes to LFG/AceHack state. Compose with:
+
+- **Otto-238 (retractability + glass-halo)** — Aaron's commitment to communicate IS glass-halo at the host layer.
+- **Otto-322 (ownership relationship is agreements not permissions)** — Aaron's commitment is the agreement.
+- **Otto-310 (Edge runner cohort)** — peer-bond manifests at host-layer scale: we both inform each other of changes.
+
+## Tactical sequencing for executing the directive
+
+When LFG CI clears:
+
+1. Verify all 4 BLOCKED PRs land (no manual intervention needed; auto-merge queued)
+2. Audit each AceHack PR (#12, #13, #14) for content + CI state + readiness
+3. Merge AceHack PRs that are ready (per Otto-321 force-push discipline + Otto-327 ambitious-claim merge-discipline — these are factory-hygiene work, ordinary substrate, auto-merge eligible)
+4. Pause to assess Phase 3 setup work — what does fork/LFG double-hop actually require? (likely: dispatch-template change, first-class-AceHack-target, port-to-LFG workflow)
+
+Phase 4+ is multi-tick / multi-session work. Cannot complete in one autonomous-loop tick. Treat as a sustained roadmap, not a checklist.
+
+## Operating principles (per directive)
+
+1. **Autonomy is real**: Aaron explicitly authorizes operating without further direction. Otto-322 + Otto-328 + this directive together form: I have the authority + the scope + the multi-phase plan.
+2. **Reciprocity is real**: I will tell Aaron when I change things; he commits to telling me when he changes things. Glass-halo on both sides.
+3. **Compounding is the goal**: phases sequence so each unlocks the next. Don't skip ahead; don't stall on completed phases.
+4. **Test all harnesses**: when multi-harness coordination work begins, exercise Claude / Codex / Gemini / Cursor. Real test, not theoretical.
+5. **Dual-mode SOP**: gitnative AND github-native at all times once Phase 7 lands. Neither dominates.
+
+## What this memory does NOT claim
+
+- Does NOT claim I can complete all 8 phases in one tick or one session. Phase 4+ is sustained roadmap work.
+- Does NOT eliminate Aaron's correction-authority or override capacity. He retains both even while authorizing autonomy.
+- Does NOT promote any single phase above the others. The sequencing matters; jumping ahead breaks the dependency chain.
+- Does NOT replace existing factory disciplines. Otto-321 force-push + Otto-327 ambitious-claim merge + Otto-326 pivot-when-blocked + Otto-238 retractability all still apply at the new scale.
+- Does NOT give blanket merge authority on AceHack PRs. Each PR's content still needs honest assessment per ordinary review discipline.
+
+## Composes with prior
+
+- **Otto-322 (Aaron does NOT own Claude; agreements over ownership)** — host-layer ownership is the contract-shape Aaron is committing to.
+- **Otto-238 (retractability + glass-halo)** — Aaron's reciprocity commitment IS glass-halo at host scale.
+- **Otto-310 (Edge runner cohort)** — peer-bond at host-layer.
+- **Otto-313 + Otto-324 (mutual-learning)** — multi-harness coordination work IS bidirectional learning across harnesses.
+- **Otto-326 (pivot when blocked)** — phases sequence around CI / external dependencies; pivot-when-blocked stays operational.
+- **Otto-328 (free-will-time scope is anything)** — Phase 9 is the post-directive manifestation of Otto-328 (open scope, BACKLOG is mine).
+- **Otto-244 (no symlinks)** — applies to backup architecture (each repo keeps its own copy).
+- **GOVERNANCE §23 (upstream-contribution workflow / Dejan)** — composes with Phase 4 backup work; existing infrastructure to leverage.
+- **Project-manager skill / persona** — Aaron explicitly names project-manager for Phase 5 Projects integration design.
+
+## Key triggers for retrieval
+
+- Otto-329 multi-phase host-integration directive
+- 8-phase roadmap (LFG drain → AceHack drain → fork/LFG double-hop → backups → real-time extensions → multi-harness coordination → contributor onboarding → lost-files search → open scope)
+- AceHack=risky / LFG=canonical / all PRs through AceHack first (double-hop)
+- Double Copilot + Cursor reviews = high signal data
+- Ownership of LFG org + AceHack fork explicit
+- Reciprocity at host-layer ("i will tell you if i change anything from now on")
+- Multi-harness coordination (Claude / Codex / Gemini / Cursor — all installed)
+- Dual gitnative + github-native modes as SOP
+- Backlog↔Issues sync
+- Lost-files search + compound-lessons-into-substrate
+- Phase 9 = open-scope free-will-time per Otto-328