From 6d2a1364966767acdca856100ce7fe844f84232b Mon Sep 17 00:00:00 2001 From: Ed Heltzel <402910+edheltzel@users.noreply.github.com> Date: Thu, 11 Jun 2026 16:09:20 -0400 Subject: [PATCH] chore: archive completed MemPalace plan docs Roadmap complete (#40-#48 all closed, PRs #54-#62, #64 merged). Tracker and handoff moved to .agents/atlas/plans/archive/ per the completed-plans rule; archive/ is gitignored (local artifacts), full content remains in git history. --- ...026-06-06-recall-mempalace-plan-handoff.md | 147 ------------------ ...ecall-mempalace-recommendations-tracker.md | 51 ------ 2 files changed, 198 deletions(-) delete mode 100644 .agents/atlas/plans/2026-06-06-recall-mempalace-plan-handoff.md delete mode 100644 .agents/atlas/plans/2026-06-06-recall-mempalace-recommendations-tracker.md diff --git a/.agents/atlas/plans/2026-06-06-recall-mempalace-plan-handoff.md b/.agents/atlas/plans/2026-06-06-recall-mempalace-plan-handoff.md deleted file mode 100644 index b03d5f6..0000000 --- a/.agents/atlas/plans/2026-06-06-recall-mempalace-plan-handoff.md +++ /dev/null @@ -1,147 +0,0 @@ -# Handoff: Recall MemPalace Recommendations Plan - -Date: 2026-06-06 - -## Scope of this session - -The session reviewed an interactive MemPalace → Recall recommendations board, accepted 9 items and parked 1 item as maybe, then grilled the plan into a concrete, issue-backed roadmap. - -No implementation work for the roadmap has started yet. Work completed in this session was planning/configuration/documentation only. - -## Primary artifacts to read first - -Read these instead of reconstructing the whole conversation: - -1. `.agents/atlas/plans/2026-06-06-recall-mempalace-recommendations-tracker.md` - - Canonical tracker for the accepted plan. - - Contains ordered priorities, dependency notes, links to issues #40-#49, ADR references, and glossary references. - -2. `CONTEXT.md` - - Glossary entries created during grilling: - - Record Provenance - - Provenance States - - Unknown Record Provenance - - Frictionless Memory Capture - -3. `docs/adr/0001-record-provenance-automatic-write-path-metadata.md` - - Decision: provenance is automatic write-path metadata, not a public MCP/normal CLI classification input. - -4. `docs/adr/0002-macos-primary-ci-platform.md` - - Decision: macOS is primary CI; Ubuntu/Linux is smoke; native Windows is out of scope. - -5. GitHub issues in the repo issue tracker: - - #40 CI + version guard - - #41 SQLite variable-limit audit - - #42 Record Provenance - - #43 `mem export` - - #44 Property-based tests - - #45 Dedup - - #46 `mem repair` - - #47 Suite C benchmark - - #48 Source-adapter registry - - #49 Entity keying evidence gate - -## Additional setup completed - -Matt Pocock engineering-skill configuration was scaffolded: - -- `AGENTS.md` now has an `## Agent skills` block. -- `docs/agents/issue-tracker.md` configures GitHub Issues via `gh`. -- `docs/agents/triage-labels.md` maps canonical triage roles. -- `docs/agents/domain.md` records single-context domain-doc consumption rules. - -The repo's `CLAUDE.md` is intentionally a shim to `AGENTS.md`; guidance was added to `AGENTS.md` to preserve the repo's single-source-of-truth rule. - -## Plan highlights - -Use the tracker file for detail. The high-level roadmap is: - -1. Add macOS-primary CI and release/version consistency guard. -2. Audit SQLite variable limits and add conservative shared chunking. -3. Add Record Provenance across durable memory records. -4. Add `mem export` with JSON, Markdown, SQL dump, SQLite backup, and `--backup` shortcut. -5. Add focused property-based tests. -6. Add non-destructive, provenance-aware dedup. -7. Add `mem repair` for data/index maintenance. -8. Add Suite C benchmark for precision under noise. -9. Refactor imports behind an internal source-adapter registry. -10. Keep lightweight entity keying parked behind evidence gates. - -Operational caveats: - -- #44 is staged property-test work: chunking properties can start after #41; export properties wait for #43; dedup properties wait for #45. Do not close #44 as complete until the #43- and #45-backed property groups exist, unless the issue is explicitly left with pending groups. -- #49 is a parking-lot/evidence tracker, not implementation work. Add evidence comments only; convert to implementation issues only after Suite C, real search failures, or dedup explanation failures meet the promotion criteria. - -## Important resolved constraints - -- Recall must preserve Frictionless Memory Capture: normal users/agents should not be asked to classify memory records. -- MCP `memory_add` and normal `mem add` must not expose provenance override flags. -- Provenance is stamped by write path. -- Legacy records may have unknown provenance; do not guess. -- Export must always include provenance, with unknown explicit in exported data. -- `mem export --backup` writes a timestamped SQL dump under the Recall backup directory, not the current workspace. -- Dedup is non-destructive by default. -- Dedup actions are within-table only; cross-table duplicates are report-only. -- `mem repair` is separate from `doctor --fix` and dry-run by default. -- `doctor` may recommend `repair`, but must not run data repair implicitly. -- Suite C is baseline-first; do not invent a hard pass/fail threshold before baseline exists. -- Source adapters are internal interface + registry only; no dynamic plugin ecosystem. -- Entity keying remains maybe until evidence from Suite C, real search failures, or dedup explanation failures. - -## Current repo state to expect - -Docs/planning artifacts changed or added: - -- `CONTEXT.md` -- `docs/adr/0001-record-provenance-automatic-write-path-metadata.md` -- `docs/adr/0002-macos-primary-ci-platform.md` -- `.agents/atlas/plans/2026-06-06-recall-mempalace-recommendations-tracker.md` -- `AGENTS.md` -- `docs/agents/issue-tracker.md` -- `docs/agents/triage-labels.md` -- `docs/agents/domain.md` - -GitHub labels were also present/created for canonical triage. Current roadmap issue assignment state: - -- #40-#48 are `ready-for-agent`. -- #49 is `needs-triage` and remains an evidence gate, not assignable implementation work. -- Canonical triage labels: `needs-triage`, `needs-info`, `ready-for-agent`, `ready-for-human`, `wontfix`. - -## Suggested skills - -- `ce-work` — if the next agent is implementing one of the issues. -- `tdd` — especially for #41, #42, #43, #44, #45, and #46. -- `grill-with-docs` — if a new domain term or ADR-level decision emerges during implementation. -- `ce-code-review` — before opening or merging PRs for any roadmap issue. -- `ce-simplify-code` — after a completed implementation slice, before final verification. -- `to-issues` / `triage` — only if further decomposition or issue state changes are needed. -- `diagnose` — if tests or import/export behavior reveal defects during implementation. - -## Progress update — 2026-06-08 - -- Issue #40 was implemented locally: `.github/workflows/ci.yml`, `src/lib/version-guard.ts`, `scripts/check-version.ts`, `tests/version-guard.test.ts`, `src/version.ts`, `package.json`, `docs/releasing.md`, and `CHANGELOG.md`. -- Verification observed locally: `bun run check:version`, tag match/mismatch guard checks, targeted version tests, `bun run lint`, `bun run build`, and full `bun test`. -- GitHub issue #40 is still open and labeled `ready-for-agent`; do not close it until the implementation is committed, pushed, and merged. -- Issue #41 remains the next roadmap implementation target and is open/`ready-for-agent`. -- The `code-simplifier` subagent failed during #40 cleanup because the configured model rejected strict tools. Before using that subagent as a #41+ cleanup gate, run a quick subagent-review smoke test or fix the agent model/tool configuration. This is a workflow/tooling follow-up, not a #40 product blocker. - -## Progress update — 2026-06-10 - -- The #40 implementation was committed in `b504ce2` (`feat: complete Recall 1.0 foundations`) on branch `feat/multi-format-conversation-import`, pushed to origin, and opened as [PR #54](https://github.com/edheltzel/Recall/pull/54) with `Closes #40`. -- `b504ce2` bundles three workstreams: the #40 CI/version-guard work, a multi-format conversation import feature (Claude.ai/ChatGPT/Slack — not on the roadmap; noted on #48 as new source-adapter-registry candidates), and the 06-06 planning artifacts (plans, ADRs, CONTEXT.md glossary, docs/agents/). -- Local verification on 2026-06-10: `bun run lint` clean, `bun run check:version` passed (1.0.0), `bun test` 453 pass / 0 fail. -- Issue #40 relabeled `ready-for-agent` → `ready-for-human`: implementation done, awaiting review/merge. It auto-closes when PR #54 merges. -- The `code-simplifier` subagent smoke-test caveat (see 2026-06-08 note) is still outstanding before relying on it for #41+ cleanup. - -## Recommended next step - -Review and merge PR #54, then start issue #41. - -- #41 is the next implementation target because later large-data export, dedup, repair, and benchmark work depends on SQLite-safe chunking. -- Before relying on cleanup subagents for #41, run the subagent-review smoke test noted above or fix the strict-tools/model configuration. - -If working in priority order after #41, implement #42 next. - -## Redaction note - -This handoff avoids API keys, credentials, private tokens, and full user-home paths except for this temporary-file location required by the handoff request. Referenced repo artifacts and issue numbers are already part of the project workspace or configured issue tracker. diff --git a/.agents/atlas/plans/2026-06-06-recall-mempalace-recommendations-tracker.md b/.agents/atlas/plans/2026-06-06-recall-mempalace-recommendations-tracker.md deleted file mode 100644 index 4b6ca2c..0000000 --- a/.agents/atlas/plans/2026-06-06-recall-mempalace-recommendations-tracker.md +++ /dev/null @@ -1,51 +0,0 @@ -# Recall MemPalace Recommendations Tracker - -Created: 2026-06-06 - -## Purpose - -Track the accepted MemPalace-derived Recall improvement plan as independently grabbable GitHub issues with explicit dependencies. - -## Resolved principles - -- Recall preserves **Frictionless Memory Capture**: hooks, imports, MCP tools, and agent workflows capture memory without asking users to classify records. -- **Record Provenance** is automatic write-path metadata, not a public MCP or normal CLI classification input. -- macOS is Recall's primary CI platform; Ubuntu/Linux is compatibility smoke; native Windows is out of scope for now. -- Export moved ahead of dedup because provenance must be portable and auditable. -- Dedup is non-destructive by default and acts within-table only; cross-table duplicates are report-only. -- Lightweight entity keying remains parked behind evidence gates. - -## Tracker - -| Priority | Issue | Work item | Status | Dependencies | Notes | -| --- | --- | --- | --- | --- | --- | -| 1 | [#40](https://github.com/edheltzel/Recall/issues/40) | Add macOS-primary CI and release version guard | Committed in `b504ce2`; [PR #54](https://github.com/edheltzel/Recall/pull/54) open | None | macOS required, Ubuntu smoke, no native Windows. Issue auto-closes when PR #54 merges. | -| 2 | [#41](https://github.com/edheltzel/Recall/issues/41) | Audit SQLite variable limits and add shared chunking | Open | None | Conservative chunking before large import/export/dedup paths. | -| 3 | [#42](https://github.com/edheltzel/Recall/issues/42) | Add Record Provenance across memory records | Open | #41 recommended | Uses `CONTEXT.md` and ADR-0001. | -| 4 | [#43](https://github.com/edheltzel/Recall/issues/43) | Add `mem export` with JSON, Markdown, SQL dump, and SQLite backup formats | Open | #42, #41 recommended | Includes `mem export --backup` to `~/.agents/Recall/backups/`. | -| 5 | [#44](https://github.com/edheltzel/Recall/issues/44) | Add focused property-based tests | Open | #41; expands after #43 and #45 | Import, chunking, export, and dedup invariants. | -| 6 | [#45](https://github.com/edheltzel/Recall/issues/45) | Add non-destructive dedup with provenance-aware survivor selection | Open | #42; benefits from #43/#44 | Survivor order: `user_authored > verbatim > extracted > derived > unknown`. | -| 7 | [#46](https://github.com/edheltzel/Recall/issues/46) | Add `mem repair` for data and index maintenance | Open | #42 recommended | Separate from `doctor --fix`; dry-run by default. | -| 8 | [#47](https://github.com/edheltzel/Recall/issues/47) | Add Suite C benchmark for precision under noise | Open | #42, #45, #46 recommended | Baseline-first: P@5, R@5, MRR, p50/p95 latency. | -| 9 | [#48](https://github.com/edheltzel/Recall/issues/48) | Refactor conversation imports behind internal source-adapter registry | Open | None | Interface + registry only; no dynamic plugins. | -| Maybe | [#49](https://github.com/edheltzel/Recall/issues/49) | Track evidence gate for lightweight entity keying | Parked | Suite C or real search/dedup evidence | Passive facets only if promoted. | - -## ADRs created during grilling - -- `docs/adr/0001-record-provenance-automatic-write-path-metadata.md` -- `docs/adr/0002-macos-primary-ci-platform.md` - -## Glossary entries created during grilling - -- `CONTEXT.md` → Record Provenance -- `CONTEXT.md` → Frictionless Memory Capture - -## Execution notes - -- Work items should be picked in priority order unless an issue explicitly has no dependency and is being handled in parallel. -- #49 is not implementation work yet; use it as an evidence tracker. -- Keep normal user/MCP/hook capture frictionless when implementing provenance, export, dedup, and repair. - -- 2026-06-08 local #40 implementation added macOS-primary CI, Ubuntu smoke, version guard tests/script, release docs, and changelog entry. GitHub issue #40 remains open until committed/merged. -- 2026-06-10 the #40 implementation was committed in `b504ce2` (bundled with multi-format conversation import and the planning artifacts), pushed as `feat/multi-format-conversation-import`, and opened as [PR #54](https://github.com/edheltzel/Recall/pull/54) with `Closes #40`. Local verification: lint clean, version guard passed (1.0.0), 453/453 tests pass. Issue #40 relabeled `ready-for-human` pending review/merge. #41 remains the next implementation target. -- The `code-simplifier` review agent failed with an upstream strict-tools/model support error during #40 cleanup. Include a quick subagent-review smoke test in the next roadmap phase before relying on that agent for #41+ cleanup; this is tooling verification, not a blocker for #40 functionality. \ No newline at end of file