|
1 | 1 | # codex-mem Go Development Tracker |
2 | 2 |
|
3 | | -Last updated: 2026-03-19 |
| 3 | +Last updated: 2026-03-20 |
4 | 4 | Status: active |
5 | 5 |
|
6 | 6 | ## Purpose |
@@ -149,6 +149,34 @@ Immediate next tasks: |
149 | 149 |
|
150 | 150 | ## Decisions Log |
151 | 151 |
|
| 152 | +### 2026-03-20 Session Update |
| 153 | + |
| 154 | +- Completed: Continued the exploratory v2 documentation set by adding a maintainer-facing semantic-interfaces draft plus a migration-sequencing draft. The interface doc maps the v2 design into the current Go package boundaries, proposes additive retrieval, memory-domain, semantic-index, and runtime-resurfacing interfaces, and keeps `retrieval.Service` on a nil-safe lexical-first path. The migration doc then turns the same design into a rollout order: interfaces first, additive note-embedding metadata migration next, sidecar/backfill tooling after that, notes-only hybrid retrieval later, and implicit resurfacing only after hybrid retrieval itself is stable. |
| 155 | +- In progress: none. |
| 156 | +- Blockers: none. |
| 157 | +- Next step: decide whether to keep extending docs with an implementation-spike plan, or start the first Go slice for additive metadata fields and nil-safe semantic interfaces. |
| 158 | + |
| 159 | +### 2026-03-20 Session Update |
| 160 | + |
| 161 | +- Completed: Continued the exploratory v2 documentation set with two more design artifacts: a v2 conformance-scenarios draft and an operator-facing embedding backfill/health draft. The conformance doc defines the minimum scenario matrix for lexical-only baseline behavior, happy-path hybrid retrieval, degraded semantic states, runtime resurfacing controls, privacy/lifecycle guardrails, and sidecar rebuild recovery. The operator doc turns the storage draft into a rollout sequence, health-state model, expected summary fields, and repair-oriented workflows, while the roadmap and readmes now link both drafts. |
| 162 | +- In progress: none. |
| 163 | +- Blockers: none. |
| 164 | +- Next step: decide whether the next v2 design pass should sketch concrete Go interfaces and migration ordering, or stop design expansion and begin implementation spikes behind config gates. |
| 165 | + |
| 166 | +### 2026-03-20 Session Update |
| 167 | + |
| 168 | +- Completed: Continued the exploratory v2 documentation set by adding a dedicated embedding-storage and backfill draft. The new doc compares direct-in-SQLite, local sidecar, and external-backend shapes; recommends SQLite metadata plus a pluggable local sidecar index for the first rollout; defines minimal embedding lifecycle metadata, explicit one-shot backfill flow, repair semantics, and degraded retrieval rules; and links that design back into the v2 outline, roadmap, and doc indexes. |
| 169 | +- In progress: none. |
| 170 | +- Blockers: none. |
| 171 | +- Next step: decide whether the next v2 draft follow-up should add operator-facing backfill/health guidance or a conformance-scenarios draft for lexical-only, hybrid, and degraded retrieval modes. |
| 172 | + |
| 173 | +### 2026-03-20 Session Update |
| 174 | + |
| 175 | +- Completed: Continued the exploratory v2 documentation set by adding a dedicated runtime resurfacing draft plus a companion v2 config draft. The new docs define a suggested consult-scoring pipeline, prototype confidence thresholds, session-local suppression-cache cooling rules, a shaped working-context payload for implicit memory injection, and a notes-first early-v2 boundary that keeps handoffs lexical-only at first. The v2 outline, roadmap, and doc indexes now link to the new drafts so the design direction is easier to re-enter. |
| 176 | +- In progress: none. |
| 177 | +- Blockers: none. |
| 178 | +- Next step: decide whether the next v2 draft follow-up should focus on embedding storage/backfill design, conformance scenarios for hybrid retrieval, or pause until implementation work begins. |
| 179 | + |
152 | 180 | ### 2026-03-19 Session Update |
153 | 181 |
|
154 | 182 | - Completed: Added `--fail-on-partial` to `ingest-imports` so operators can keep `--continue-on-error` partial-success behavior, failed-output exports, and failed-manifest exports while still receiving a non-zero exit when any line in the batch failed. The ingest report now surfaces `fail_on_partial`, CLI coverage verifies that the partial report is still written before the command exits with an error, and the reusable `App.IngestImports` path now honors the same flag so embedded callers can receive the same report-plus-error behavior instead of only the CLI getting that control. Operator docs plus the README describe the automation-oriented workflow. |
|
0 commit comments