chore(m07): triage Honcho memory v2 into design vs execution lanes #541

New issue

Closed

opened 2026-05-28 01:21:53 +02:00 by codex · 1 comment

codex commented

2026-05-28 01:21:53 +02:00

Collaborator

Scope

Make Milestone 07 actionable without risking vector-space drift.

Classify #357 and #459-#461 into:

design-only;
safe next implementation;
blocked until more evidence;
superseded/merge candidate.

Spec sources

state/roadmap/current-platform-roadmap.md § 07 - Honcho memory v2
runbooks/honcho-ollama-gemma-switch.md
Open M07 issues: #357, #459, #460, #461

Acceptance criteria

A concise packet separates embedding-space migration from procedural memory/control-plane work.
No recommendation mixes 1536d OpenAI embeddings with a new vector space without shadowing/rollback.
No runtime mutation.

Out of scope

Running migrations.
Changing Honcho providers.
Writing memory-control-plane code.

Agent notes

Recommended executor: Gemini 3.5 Flash local.

Rules:

Keep the PR small and boring.
Prefer closing/splitting stale work over adding new architecture docs.
Do not mutate production runtime.
Do not close issues live unless the issue explicitly requests it and evidence is clear.
PR body must include Spec sources read.

## Scope Make Milestone 07 actionable without risking vector-space drift. Classify #357 and #459-#461 into: - design-only; - safe next implementation; - blocked until more evidence; - superseded/merge candidate. ## Spec sources - `state/roadmap/current-platform-roadmap.md` § 07 - Honcho memory v2 - `runbooks/honcho-ollama-gemma-switch.md` - Open M07 issues: #357, #459, #460, #461 ## Acceptance criteria - A concise packet separates embedding-space migration from procedural memory/control-plane work. - No recommendation mixes 1536d OpenAI embeddings with a new vector space without shadowing/rollback. - No runtime mutation. ## Out of scope - Running migrations. - Changing Honcho providers. - Writing memory-control-plane code. ## Agent notes Recommended executor: Gemini 3.5 Flash local. Rules: - Keep the PR small and boring. - Prefer closing/splitting stale work over adding new architecture docs. - Do not mutate production runtime. - Do not close issues live unless the issue explicitly requests it and evidence is clear. - PR body must include `Spec sources read`.

codex added this to the 07 - Honcho memory v2 milestone

2026-05-28 01:21:53 +02:00

codex added the

labels

2026-05-28 01:21:53 +02:00

codex commented

2026-05-29 16:55:51 +02:00

Author

Collaborator

Role: executor
Intent: lane-outcome
Needs owner: no

M07 Honcho memory v2 triage packet. No runtime mutation, no provider changes, no embedding migration.

Spec sources read:

state/roadmap/current-platform-roadmap.md § 07 - Honcho memory v2
runbooks/honcho-ollama-gemma-switch.md
Open M07 issues #357, #459, #460, #461
Iskra domain reply via Platform Approvals, metadata-only

Iskra domain signal:

ISKRA_M07 issue=541 next=#357,#459,#461 hold=#460 reason=prioritize_memory_architecture_before_packet_templates

Disposition:

Issue	Disposition	Notes
#357	next, design-only	Keep embeddings on current 1536d path until a versioned/shadow migration design exists. No vector-space mixing.
#459	next, architecture	ADR for Memory Control Plane boundaries before code. This should define Honcho vs Forgejo vs local promise ledger vs future Postgres memory tables.
#461	next, procedural design	Useful after/with #459; procedural registry is high-value, but implementation should stay bounded by ADR boundaries.
#460	hold	Task packets/checkpoints are likely valuable, but should not precede the architecture/procedural boundary work.

Recommendation:

Run #459 first or in parallel with #357 as design-only.
Let #461 produce a procedural registry spec/seed plan after #459 has boundaries.
Keep #460 parked until the first two clarify schema ownership and state lifecycle.

Next: close this triage issue; execute #357/#459/#461 as small, non-runtime slices.

**Role:** executor **Intent:** lane-outcome **Needs owner:** no M07 Honcho memory v2 triage packet. No runtime mutation, no provider changes, no embedding migration. Spec sources read: - `state/roadmap/current-platform-roadmap.md` § 07 - Honcho memory v2 - `runbooks/honcho-ollama-gemma-switch.md` - Open M07 issues #357, #459, #460, #461 - Iskra domain reply via Platform Approvals, metadata-only Iskra domain signal: ```text ISKRA_M07 issue=541 next=#357,#459,#461 hold=#460 reason=prioritize_memory_architecture_before_packet_templates ``` Disposition: | Issue | Disposition | Notes | |---|---|---| | #357 | next, design-only | Keep embeddings on current 1536d path until a versioned/shadow migration design exists. No vector-space mixing. | | #459 | next, architecture | ADR for Memory Control Plane boundaries before code. This should define Honcho vs Forgejo vs local promise ledger vs future Postgres memory tables. | | #461 | next, procedural design | Useful after/with #459; procedural registry is high-value, but implementation should stay bounded by ADR boundaries. | | #460 | hold | Task packets/checkpoints are likely valuable, but should not precede the architecture/procedural boundary work. | Recommendation: 1. Run #459 first or in parallel with #357 as design-only. 2. Let #461 produce a procedural registry spec/seed plan after #459 has boundaries. 3. Keep #460 parked until the first two clarify schema ownership and state lifecycle. **Next:** close this triage issue; execute #357/#459/#461 as small, non-runtime slices.

codex closed this issue

2026-05-29 16:55:51 +02:00