chore(memory): phase 1 redesigned as native context diet

This commit is contained in:
bchanot committed 2026-10-24 19:31:41 +02:00
1 parent d157032616
commit 2c9ed20782
2 files changed
+15 -8

No files matched your search

+14 -8
View File
@@ -24,14 +24,20 @@ Shared rules for every phase:
- One pass over the git repos in ~/Documents/ (this repo first, then by size: faunosteo, zenquality, game, robin_petier, config, integration_alphalink): one `chore/claude-md-split` branch per repo, per-project approval, never an automatic merge (user signal per repo), `/context` before/after per project recorded in that project's journal.
- Proof: per project, canary that a moved section is ABSENT at start and PRESENT after a Read of a matching file; total always-on delta across the 7 projects.
- [ ] **Phase 1, mod `context`: marked-section subtraction + JIT injection + project fingerprint**
- New mod `mods/context/` (+ `skills/context` link, census row in mods.test.sh), config `~/.claude/context-mod.json`.
- Fingerprint at `session.start` (fs only, cached in `$.state`): stack (package.json deps astro/next/react/vue/svelte, Cargo.toml, go.mod, pyproject), web public vs app, `supabase/`, messaging deps (twilio, sendgrid, resend, mailgun), `functions/` or serverless config, `graphify-out/graph.json`, tracked code file count, code origin via `lib/code-origin.sh` (ours / foreign), gitflow repo, `.claude/memory/` present, empty dir.
- Marked sections and triggers: Design work (UI file Read/Edit or UI prompt, same detector as `hooks/design-toolchain-reminder.sh`); graphify (graph.json exists → stays in context block); Public websites never SPA + Web APIs versioned + rules/default-language.md (empty dir, or skill.prompt of init-project / ship-feature / brainstorming / refactor); secure-by-default sections 4 Supabase / 5 SMS-email / 6 cloud functions (fingerprint); gitflow detail (keep 3 core lines; full text on first Bash git/gitflow call); Memory registries format (keep 2 lines; full text on Read/Edit/Write under `.claude/memory/`); "Our code" vs "Foreign code" variant (origin → inject only the applicable one, at first code Read).
- Delivery: always-on part via `prompt.context` (drop/rewrite `claudeMd` instructionFiles); JIT via `tool.call` result `context` (Read comes before Edit, so guidance lands before the write) and `prompt.submit` `context`; skill-triggered via `skill.prompt` append.
- Drop the unused auto-memory system section (`prompt.section` name `memory`, dir `~/.claude/projects/*/memory/` empty 2026-10-23) behind a config flag.
- Proof: canary per section (ABSENT without trigger, PRESENT after it, once only, again after /clear and after a compaction); mod-off run loads everything; `claude plugin test` unit suite; `/context` delta (~4.4k est.).
- Cleanup: `hooks/design-toolchain-reminder.sh` + its settings.json registration + `lib/tests/design-toolchain-reminder.test.sh` (the mod injects the section itself); the hook's "(already in context)" text; CLAUDE.global.md lines that only pointed at the moved text; BDR-031 citers.
- [ ] **Phase 1, native context diet: no mod** (redesigned 2026-10-24 after challenge: the engine already does most of it)
- Facts 2026-10-24 (scratchpad pathtest, memtest): a `paths:` rule loads on Read of a matching file, NOT on Write of a new file (S1 ABSENT, S2 after Write ABSENT, S3 after Read PRESENT); `"autoMemoryEnabled": false` removes the auto-memory system section (PRESENT → ABSENT); classic SessionStart / UserPromptSubmit hooks inject additionalContext (seen live in every session); instruction files reload on compaction natively (InstructionsLoaded `load_reason: compact` in the 2.1.306 types). So each section goes to the native carrier that matches its trigger, with no mod code, no fail-open risk, and it works with the mod off and in sub-agents.
- Section → carrier:
- auto-memory system section → `"autoMemoryEnabled": false` in settings.json (one line, proven).
- Design work (1.5k chars) → user rule `rules/design-work.md` with the UI globs of rules/web-building.md (fires on UI file Read) + `hooks/design-toolchain-reminder.sh` injects the section text itself when it detects UI work in a prompt (covers new files created from a prompt); drop its "(already in context)" wording.
- graphify (0.6k) → `hooks/session-start.sh` injects the section only when `graphify-out/graph.json` exists (it already computes graph presence for the banner).
- Public websites never SPA + default language + web APIs versioned (2.3k) → into the skills that brainstorm (init-project, ship-feature STEP 1, brainstorming, refactor whole-project, onboard) + session-start injects them when the cwd has no tracked code file (new project); dedupe "Web APIs" against rules/rest-api.md.
- secure-by-default domains 4 Supabase + 6 cloud functions → user path rules (`supabase/**`, `**/*.sql`; `functions/**`, `**/serverless.*`, `**/firebase.json`); keep "Every API", hosting, injection, exposure and SMS/email always-on (no reliable path for messaging, 325 chars).
- Our code vs foreign code → session-start runs `lib/code-origin.sh` once and injects one line ("this repo: ours, strict limits" / "foreign, keep its style"); both short sections stay always-on.
- Registry format → compressed in phase 4a (lib/memory.sh enforces it); gitflow detail → compressed in phase 2 cleanup (hooks + lib + merge gate enforce it). No JIT needed for either.
- Project fingerprint (stack, supabase, messaging, functions, graph, origin) moves to phase 3, its first consumer (listing filter), reused by phase 5.
- Proof: canary per moved section (ABSENT without trigger, PRESENT after it: UI Read, UI prompt, graph.json present, empty dir, skill load, supabase Read); a new-file-only UI turn still gets Design work through the prompt hook; `/context` delta (~4.4k est. across phases 1, 2, 4a); hook tests extended (design-toolchain-reminder, session-start).
- Cleanup: CLAUDE.global.md loses the moved sections (one pointer line each where needed), citers patched (doctrine-citers census); BDR-031 citers.
- The `context` mod is born in phase 2 or 3 (first need: guards or listing rewrite); challenge at phase 2 launch whether guards fit better as classic hooks (push-guard pattern, transcript_path gives the last user message) or as the mod.
- [ ] **Phase 2, guards in the context mod (deny with reason, deterministic)**
- Registry format: moved to phase 4a (measured 2026-10-24: registry writes go through Bash/python, never Edit, since Edit needs a Read of a 357k-char file; a `tool.call` guard on Edit/Write would miss them; a pre-commit validator sees every write path).