9.3 KiB
CONTRACT — memory-recall (phase 4c)
- date: 2026-11-04 | flow: ship-feature | branch: feature/memory-recall (off develop
06d2576, created at STEP 4) - status: active
REQUEST (verbatim — IMMUTABLE)
Phase 4c of the context mod queue (BDR-125), TODO.md "Phase 4b-4d, memory v2 capture, read, feedback" — the 4c read side: two channels only: intent recall at prompt.submit for classifier labels design/analysis or the first coding prompt after idle (BM25 top-30 over title + lesson + body → haiku keeps ≤3, bounded 3 s, fail open) and tool-error 4-gram match (PostToolUse; a classic hook is acceptable if the mod ledger is not needed). Session digest (open BLK via the status parser, actionable TODO, last journal line) in hooks/session-start.sh on source: startup|compact|clear; a SubagentStart twin if sub-agents need it. Injection = ID + title + lesson line; superseded never injected; budget ~1.5k chars per turn; once per context. Deferred: Read path triggers, Bash command triggers, "critical → deny once", Tier 1, cross-repo index, memory_search tool. Plus the 4b follow-ups queued for 4c: earlier-session milestones in memory_pending, common-phrase 4-gram collisions (every ENOENT reads known LRN-056), the lexical superseded reader shortcut:. Host: mods/memory (BDR-137). 4d stays deferred until an injection log exists.
(user brief 2026-11-02, step 3: "Phases 4b à 4d"; launched after the 4b merge, user "merge it" 2026-11-04. Session constraints of that brief hold: default model opus; verifier, security-auditor and plan-challengers pinned on opus; never commit a /model or UI write of settings.json (LRN-253); plan text through Write/Edit (BLK-039); destructive steps on the live tree in the main loop after the user's go, mv over rm (BLK-040).)
CLARIFICATIONS
Pass A (gaps found by the STEP 0d read-before, put to the user at the STEP 1 brainstorm; answers land [gated]):
- The classifier label is private to model-router (
st.turnMain, never stored or published) and the classifier never runs on prompts its keyword rules already map (plan|design|architecture…→ plan,why|explain|review|audit…→ reflect), nor on slash commands: a trigger "classifier labels design/analysis" read literally misses exactly the design/analysis prompts that say so. - No uniform "lesson line" exists in the registries (BDR
**Decision**median 643 chars, max 1989; LRNApply/Future application/Pattern; BLKSolution/Real cause; EVALAction): three BDR hits overflow the 1.5k budget unless the line is cut. - Superseded readers diverge (memory.sh parser vs the 4b lexical reader: BDR-031 live for the lexical one, EVAL-041 falsely superseded for the shell one).
- Engine types 2.1.316 expose
$.stateand$.process.run(4b recorded "no$.state" from the 2.1.277 types); runtime availability unverified for both; modsession.starthas nosource,classic.SessionStartdoes. - The 4-gram collision:
no such file or/such file or directorybelong to LRN-056 only, so every ENOENT reads known LRN-056 (document frequency does not fix it). - Proof needs ≥ 20 real prompts and ≥ 10 real errors labelled for relevance; precision bar = the 14/22 haiku-rerank baseline (BDR-125 amendment 2, TODO.md:114).
Brainstorm answers [gated 2026-11-04] (user: "continue" to recommendations 0-11, read as yes):
0. Spike first (done: .claude/tasks/plans/2026-11-04-memory-recall-spike/REPORT.md): $.process.run memory.sh index 128 ms; classic.SessionStart fires in a mod with source startup/resume/compact and its additionalContext reaches the model; tool.call error-answer context reaches the model; haiku rerank median ≈ 2 s, tail 5.7-7.3 s, ≈ 4k input tokens.
- Intent-recall trigger without the router classifier: composer prompt, not mid-turn, not a slash command, ≥ 12 chars, AND (first prompt of a context [startup, /clear, compaction] OR first prompt after ≥ 30 min idle OR a design/analysis keyword match, the router's two default rule lists copied with their source commit).
- Corpus = live BDR + LRN + BLK (EVAL excluded). Lesson line: BDR
Decision; LRNApply>Future application>Pattern; BLKSolution>Real cause; first sentence cut at 280 chars. Injection line =ID, title (≤ 100), lesson; ≤ 3 entries; hard cap 1500 chars. - Rerank: BM25 top-30 over title + lesson + body; haiku
claude-haiku-5-5, effort low, 64 max tokens, gets the prompt + 30 candidates as ID + title + lesson, answers IDs validated against the candidates; own 3 s race (LRN-218); failure or timeout → nothing injected (no BM25-only fallback). - Superseded reader = lib/memory.sh (single reader, BDR-129) through
$.process.run(['bash', memory.sh, '-C', root, 'index'])(spike: works, 128 ms); superseded and deprecated excluded; the 4b lexicalshortcut:reader is replaced by it for the error index too (BDR-031 becomes superseded as memory.sh says). - Tool-error channel in the existing Bash hook (the ledger lives in the mod): a captured failure matching a live BLK/LRN entry attaches
memory: known <ID>, title, lessonto the tool result context; an errno stop-phrase list (no such file or directory, permission denied, command not found, …) drops grams wholly inside a stop phrase at index and match time; main loop and sub-agents; noise measured with the 14-day replay before shipping. - Once per context: injected IDs kept in the closure per session and agent; reset at
session.endand onclassic.SessionStartcompact/clear; the 1500-char per-turn budget covers both channels. - Digest in
hooks/session-start.sh: reads the hook JSON on stdin without blocking under</dev/null; sources startup, clear, compact (not resume); open BLK (upstream counted apart), first 3 unchecked items of TODO### Actionable, last journal line; ≤ 160 chars per line, ≤ 800 per block; silent without registries or without the section. No SubagentStart twin. memory_pendinglists earlier-session milestones after the current session's, marked(earlier session).- Off switch:
"enabled": falsecuts everything; a new key"recall": falsecuts the injections only (capture and tool stay). - Injection log in scope: last 200 injections per project
{at, session, channel, ids}in the store, not displayed (4d reads it). - Proof: a bun harness importing the mod's own functions (no port), haiku through
claude -p; 20 real prompts from 14 days of transcripts and the error replay; the controller labels relevance, the user checks the labels; bar ≥ 14/22; oracle written after the measurement; live: headless error half, interactive recall + digest after a real /compact by the user.
Pass B [gated 2026-11-04] (plan r1; user: "oui, tels quels" / "oui"): visible texts kept in English as written in the plan: memory recall: block, memory: known <ID> … line, memory digest: with - open: / - todo: / - journal: lines; the new suite is lib/tests/session-digest.test.sh; the memory_pending header milestones (this session): becomes milestones: with earlier-session lines marked (earlier session) (citers updated, BDR-100).
ACCEPTANCE CRITERIA (seed, enriched at STEP 3 [gated])
- Intent recall: on the qualifying prompts (trigger decided at the brainstorm), the mod ranks live registry entries by BM25 over title + lesson + body, asks a small model to keep ≤ 3, bounded in time, and attaches them as context (ID + title + lesson line); any failure or timeout attaches nothing and passes the prompt through unchanged.
- Tool-error match: a captured tool failure whose error line shares a 4-gram with a live BLK/LRN entry attaches that entry (ID + title + lesson line) as context; the common-phrase collision (ENOENT → LRN-056) no longer fires.
- Superseded entries are never injected; the superseded reader is decided at the brainstorm and its verdicts agree with lib/memory.sh on this repo's registries (an independent oracle, LRN-237).
- Budget: at most ~1.5k chars of injected recall per turn; an entry is injected at most once per context (reset on /clear and compaction).
- Session digest:
hooks/session-start.shprints open BLK (status parser), actionable TODO and the last journal line onsourcestartup|compact|clear, bounded in size; runs unchanged with</dev/nulland in projects without registries or TODO. memory_pendinglists earlier-session milestones (marked(earlier session)) after the current session's.- Every hook fails open; the
~/.claude/memory.jsonoff switch covers the read side;claude plugin validate --strict,claude plugin test(mods suite) andmake testare green. - Proof: recall precision measured on ≥ 20 real prompts and tool-error matching on ≥ 10 real errors, at or above the 14/22 baseline; a digest observed after a real compaction; the injection noise measured before shipping (LRN-261); the measured oracle written only after the measurement (EVAL-056).
FILE SCOPE (seed, final at STEP 3 [gated])
- mods/memory/hooks/register.ts, mods/memory/hooks/register.test.ts (host, BDR-137); mods/memory/.claude-plugin/plugin.json (version)
- hooks/session-start.sh and the suites that run it (lib/tests/memory.test.sh, push-guard.test.sh, graphify-gate.test.sh) if its output changes
- lib/memory.sh only if a read helper is needed (no write path change)
- README.md, USAGE.md, CHANGELOG.md, ARCHITECTURE.md, .claude/tasks/TODO.md
- .claude/tasks/plans/ (spike and measurement notes), docs/superpowers/{specs,plans} (transient), this contract