From 24bcfe6c0019349e1d0a404e9f593fe5d58cb534 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E5=B0=8F=E6=A2=A6?= Date: Tue, 11 Aug 2026 13:43:21 +0800 Subject: [PATCH] chore: neutralize team-internal identifiers on the public surface An outside reader clones this repo without our workspace, our people, or our drive layout. Names in comments read as noise; a documented example command pointing at E:\Project cannot be copied and run by anyone but us. - comments and test fixtures use neutral placeholder names - example and fixture paths move off this machine's drive layout - the benchmark README's example command no longer hardcodes a local checkout Behaviour is unchanged: these are comments, fixture strings and doc examples. meta-gate (14 cases) and supersede-shrink (24 cases) both still pass. The published MCP server identifier chinatsu-memory is deliberately untouched -- renaming it would break existing user configs, so that one is a product naming call, not a cleanup. Co-Authored-By: Claude Opus 5 --- backfill-embeddings.mjs | 4 ++-- benchmarks/longmemeval/README.md | 4 ++-- encoding-damage.integration.test.mjs | 2 +- high-signal-tokens.mjs | 4 ++-- index.mjs | 2 +- memory-health.test.mjs | 4 ++-- meta-gate.test.mjs | 8 ++++---- supersede-shrink.integration.test.mjs | 18 +++++++++--------- 8 files changed, 23 insertions(+), 23 deletions(-) diff --git a/backfill-embeddings.mjs b/backfill-embeddings.mjs index 7861d11..ff53562 100644 --- a/backfill-embeddings.mjs +++ b/backfill-embeddings.mjs @@ -28,7 +28,7 @@ const require = createRequire(import.meta.url) const __dirname = dirname(fileURLToPath(import.meta.url)) // ── Load .env.local if present ───────────────────────────── -// [2026-05-29 千夏 fix] .env.local 在 chinatsu-workspace/（memory/ 上一级）， +// [2026-05-29 fix] .env.local 在 workspace/（memory/ 上一级）， // 不在 memory/。原代码 resolve(__dirname,'.env.local') 指向 memory/.env.local（不存在）→ // 除非 shell 已 export env 否则 EMBEDDING_API_KEY 取不到直接 exit 1。两个路径都试。 for (const p of [resolve(__dirname, '..', '.env.local'), resolve(__dirname, '.env.local')]) { @@ -53,7 +53,7 @@ const CONCURRENCY = parseInt(getFlag('--concurrency') || '3', 10) const DRY_RUN = hasFlag('--dry-run') // ── Configuration ─────────────────────────────────────────── -// [2026-05-29 千夏 fix] 与 index.mjs:35-38 同一套解析：优先 engram.db（真实库）， +// [2026-05-29 fix] 与 index.mjs:35-38 同一套解析：优先 engram.db（真实库）， // 原默认 tokenmem.db 是 fresh-install 占位库（仅 10 行）→ 回填空转。 const DB_PATH = process.env.TOKENMEM_DB_PATH || (existsSync(resolve(__dirname, 'engram.db')) diff --git a/benchmarks/longmemeval/README.md b/benchmarks/longmemeval/README.md index b17c9b8..c43d951 100644 --- a/benchmarks/longmemeval/README.md +++ b/benchmarks/longmemeval/README.md @@ -44,8 +44,8 @@ reproducible result. node benchmarks/longmemeval/run.mjs ` --dataset C:\data\longmemeval_s_cleaned.json ` --per-category 5 ` - --engine baseline=E:\Project\mneme-baseline\index.mjs ` - --engine candidate=E:\Project\mneme\index.mjs ` + --engine baseline=C:\work\mneme-baseline\index.mjs ` + --engine candidate=C:\work\mneme\index.mjs ` --output C:\tmp\mneme-longmemeval-ab.json ``` diff --git a/encoding-damage.integration.test.mjs b/encoding-damage.integration.test.mjs index 387012e..4234672 100644 --- a/encoding-damage.integration.test.mjs +++ b/encoding-damage.integration.test.mjs @@ -38,7 +38,7 @@ const check = (label, cond, detail = '') => { check('ordinary prose with a question mark is not flagged', detectEncodingDamage('is the daemon up? it looked fine at 04:00') === null) check('intact CJK is not flagged', - detectEncodingDamage('daemon 重启脚本在 chinatsu-workspace/scripts 下，怎么跑？') === null) + detectEncodingDamage('daemon 重启脚本在 workspace/scripts 下，怎么跑？') === null) check('a short string is never flagged', detectEncodingDamage('??') === null) check('empty input is not flagged', detectEncodingDamage(null, undefined) === null) diff --git a/high-signal-tokens.mjs b/high-signal-tokens.mjs index 9aae84b..2ea093f 100644 --- a/high-signal-tokens.mjs +++ b/high-signal-tokens.mjs @@ -19,7 +19,7 @@ export const SHRINK_MIN_OLD_LEN = 200 // What ends a URL or path token. Half-width ':' and ',' stay legal (ports, // and trailing commas are stripped afterwards), but CJK text and CJK punctuation // terminate: Chinese prose runs straight into a path with no space, so without -// this a path swallows the rest of the sentence — "E:/Project/foo/。使用文档是". +// this a path swallows the rest of the sentence — "C:/work/foo/。使用文档是". const PATH_STOP = '\\s"\'`)）(（\\[\\]【】，、。；：！？…\\u4e00-\\u9fff\\u3000-\\u303f' const HIGH_SIGNAL_PATTERNS = [ @@ -57,7 +57,7 @@ export function extractHighSignalTokens(text) { * @returns {Array<{id,oldLen,newLen,ratio,dropped,droppedCount}>} one entry per suspicious pair */ // A token is still carried if the new text names it OR names something it is the -// tail of. Writing `memory/index.mjs` out and `E:/Project/ws/memory/index.mjs` in +// tail of. Writing `memory/index.mjs` out and `C:/work/ws/memory/index.mjs` in // is not a loss — it is the same file, said more precisely, and set difference // alone calls it a drop. // diff --git a/index.mjs b/index.mjs index cbfe22d..50fe080 100644 --- a/index.mjs +++ b/index.mjs @@ -338,7 +338,7 @@ export function initMemory() { // migration 003: decay_score + prior_versions (power-law decay + supersede paper trail). // Inline so upgrading an OLD db gets them without manually applying migrations/003 — // otherwise recall's `AND decay_score >= ?` (strict/cold-pool path) throws "no such column". - // (2026-06-10: caught during 爱芮 engram v1.1→v2.2 upgrade — 003 was the only file not inlined.) + // (2026-06-10: caught during an engram v1.1→v2.2 upgrade — 003 was the only file not inlined.) try { db.exec(`ALTER TABLE memories ADD COLUMN decay_score REAL NOT NULL DEFAULT 1.0`) log('Migration: added decay_score column') diff --git a/memory-health.test.mjs b/memory-health.test.mjs index 48eb778..f4470c9 100644 --- a/memory-health.test.mjs +++ b/memory-health.test.mjs @@ -265,7 +265,7 @@ function check(label, cond, detail = '') { const db = new Database(DB_PATH) const OPS = 'ops entry: http://10.0.0.5:9000 root, /api/svc/v1 routes, SVC_TOKEN and SVC_URL env, ' - + 'code at E:/Project/svc/, handler src/http.mjs, schema src/db.mjs. ' + + 'code at C:/work/svc/, handler src/http.mjs, schema src/db.mjs. ' + 'Padding so the row clears the 200B floor and has a real historical peak to measure against.' const priors = (content) => JSON.stringify([{ content, summary: null, merged_at: Date.now(), source_rowid: 1, created_at: Date.now() }]) const ins = db.prepare(`INSERT INTO memories (content, summary, memory_type, category, importance, memory_level, access_count, prior_versions, created_at, updated_at) VALUES (?,?,?,?,?,?,?,?,?,?)`) @@ -281,7 +281,7 @@ function check(label, cond, detail = '') { // low importance: shrank the same way but below the review bar ins.run('svc status: now v2.', 'low-imp shrink', 'long_term', 'general', 4, 'semi_abstract', 1, priors(OPS), t, t) // faithful rewrite: shorter prose, every identifier carried over - ins.run('http://10.0.0.5:9000 · /api/svc/v1 · SVC_TOKEN · SVC_URL · E:/Project/svc/ · src/http.mjs · src/db.mjs', + ins.run('http://10.0.0.5:9000 · /api/svc/v1 · SVC_TOKEN · SVC_URL · C:/work/svc/ · src/http.mjs · src/db.mjs', 'faithful tightening', 'long_term', 'general', 9, 'semi_abstract', 3, priors(OPS), t, t) const sh = detectShrinkVictims(db) diff --git a/meta-gate.test.mjs b/meta-gate.test.mjs index 0a8c426..f552604 100644 --- a/meta-gate.test.mjs +++ b/meta-gate.test.mjs @@ -3,7 +3,7 @@ // Configure the soft-binding vocabularies before importing the gate. process.env.MNEME_META_GATE_PROJECT_NAMES = 'AbyssDatabase,BlueBlob' -process.env.MNEME_META_GATE_PERSON_NAMES = '小天,千夏,爱芮,小梦' +process.env.MNEME_META_GATE_PERSON_NAMES = '张三,李四,王五,赵六' const { applyMetaGate } = await import('./meta-gate.mjs') @@ -12,12 +12,12 @@ const cases = [ { content: 'AbyssDatabase v0.2.0 released', expect: true, label: 'project+version' }, { content: '2026-05-05 ship notes for the mesh refactor', expect: true, label: 'iso_date' }, { content: 'continues from mem 6008 personality rules', expect: true, label: 'mem_ref' }, - { content: 'edit E:/Project/example/store.mjs to add the guard', expect: true, label: 'abs_path_windows' }, + { content: 'edit C:/work/example/store.mjs to add the guard', expect: true, label: 'abs_path_windows' }, { content: 'the daemon config lives at /etc/mneme/config.json', expect: true, label: 'abs_path_posix' }, { content: 'after commit 41171eb we must verify the committer', expect: true, label: 'commit_hash' }, // Downgrade (soft bindings — require env-configured vocabularies) - { content: '小天 and 千夏 aligned on the approach', expect: true, label: 'multi_person (env)' }, + { content: '张三 and 李四 aligned on the approach', expect: true, label: 'multi_person (env)' }, // Pass — signal-word cue and no bindings { content: 'cross-project rule: verify before you cite, always', expect: false, label: 'meta_signal_head' }, @@ -29,7 +29,7 @@ const cases = [ { content: 'listen on port 8080000 in the dev sandbox', expect: false, label: 'digit_only_not_hash' }, // Pass — single person mention below multi-person threshold - { content: 'when writing 小天 should not inflate importance to please anyone', expect: false, label: 'single_person' }, + { content: 'when writing 张三 should not inflate importance to please anyone', expect: false, label: 'single_person' }, ] let pass = 0, fail = 0 diff --git a/supersede-shrink.integration.test.mjs b/supersede-shrink.integration.test.mjs index 12750fe..a556a24 100644 --- a/supersede-shrink.integration.test.mjs +++ b/supersede-shrink.integration.test.mjs @@ -36,8 +36,8 @@ const FAT = `${marker} Key service ops entry. Service root http://100.87.169.100:8787 with agent API under /api/agent/v1. Set SERVICE_MANAGER_URL and SERVICE_MANAGER_TOKEN before calling. GET /api/agent/v1/whoami verifies identity; POST /keys/claim needs an Idempotency-Key. -Code lives at E:/Project/key-manager/, routes in src/http.mjs, schema in src/db.mjs. -Credentials under E:/Project/nas-setup/agents/. Inventory currently 300 available.` +Code lives at C:/work/key-manager/, routes in src/http.mjs, schema in src/db.mjs. +Credentials under C:/work/nas-setup/agents/. Inventory currently 300 available.` // ── Case 1: 1:1 supersede that keeps only the delta → warn on both axes ── { @@ -70,8 +70,8 @@ Credentials under E:/Project/nas-setup/agents/. Inventory currently 300 availabl check(missed.length === 0, 'case1 high-signal tokens reported as dropped', missed.length ? `not reported: ${missed.join(', ')}` : `${dropped.size} tokens`) - const winPath = [...dropped].some(t => /^E:[\\/]Project/.test(t)) - check(winPath, 'case1 windows drive path detected', [...dropped].filter(t => t.startsWith('E:')).join(', ')) + const winPath = [...dropped].some(t => /^C:[\\/]work/.test(t)) + check(winPath, 'case1 windows drive path detected', [...dropped].filter(t => t.startsWith('C:')).join(', ')) } // ── Case 2: full restatement + new fact → no warning ── @@ -239,8 +239,8 @@ dbRead.close() const { checkSupersedeShrink, isStillCarried } = await import('./high-signal-tokens.mjs') const carried = [ - ['memory/index.mjs', 'E:/Project/ws/memory/index.mjs'], - ['session-summarize.mjs', 'E:/Project/ws/memory/scripts/session-summarize.mjs'], + ['memory/index.mjs', 'C:/work/ws/memory/index.mjs'], + ['session-summarize.mjs', 'C:/work/ws/memory/scripts/session-summarize.mjs'], ['scripts/run.sh', 'C:/tools/scripts/run.sh'], ] for (const [short, long] of carried) { @@ -249,18 +249,18 @@ dbRead.close() // The boundary is a path separator. Same suffix, different file — must still warn. check('a different file that merely ends the same way is NOT carried', - !isStillCarried('send.mjs', new Set(['E:/Project/ws/feishu-send.mjs']))) + !isStillCarried('send.mjs', new Set(['C:/work/ws/feishu-send.mjs']))) check('a bare substring is not carried either', !isStillCarried('index.mjs', new Set(['reindex.mjs']))) const oldContent = 'runner lives at memory/index.mjs and the log rotates via scripts/run.sh, token in API_TOKEN, see https://ops.example.com/dash' - const newContent = 'runner lives at E:/Project/ws/memory/index.mjs and the log rotates via C:/tools/scripts/run.sh, token in API_TOKEN, see https://ops.example.com/dash — now also covers the nightly path' + const newContent = 'runner lives at C:/work/ws/memory/index.mjs and the log rotates via C:/tools/scripts/run.sh, token in API_TOKEN, see https://ops.example.com/dash — now also covers the nightly path' const clean = checkSupersedeShrink(newContent, [{ id: '1', content: oldContent }]) check('expanding relative paths to absolute raises no warning', clean.length === 0, JSON.stringify(clean)) // And the real loss still lands: same expansion, but the URL is gone. - const lossy = 'runner lives at E:/Project/ws/memory/index.mjs and the log rotates via C:/tools/scripts/run.sh, token in API_TOKEN' + const lossy = 'runner lives at C:/work/ws/memory/index.mjs and the log rotates via C:/tools/scripts/run.sh, token in API_TOKEN' const warned = checkSupersedeShrink(lossy, [{ id: '1', content: oldContent }]) check('a genuinely dropped identifier still warns', warned.length === 1 && warned[0].dropped.some(d => d.includes('ops.example.com')),