fix: sequoia-territory bug-fix bundle (chroma, env, build, MCP, worker) (#2394)

* fix(mcp): drop ${_R%/} parameter-expansion trim that trips Claude Code MCP validator The POSIX substring trim ${_R%/} is misread by Claude Code's MCP-config validator as a required env var named "_R%/", causing /doctor to flag mcp-search as invalid on every install. POSIX collapses // in paths, so the trim was cosmetic — drop it and the validator passes. Fixes #2350, #2354, #2356. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(env): block ANTHROPIC_BASE_URL leak + three-branch OAuth-skip predicate Issue #2375: parent-shell ANTHROPIC_BASE_URL leaked through to subprocess isolatedEnv, while ANTHROPIC_AUTH_TOKEN was blocked. The OAuth-skip predicate fired on bare BASE_URL, but no auth credential reached the subprocess -> "Not logged in". Add ANTHROPIC_BASE_URL to BLOCKED_ENV_VARS so it can only enter isolatedEnv via ~/.claude-mem/.env. Replace the OAuth-skip predicate with three branches to prevent a second-order security regression: a user with a tokenless gateway configured in .env (BASE_URL only, no token) would otherwise have their Anthropic OAuth token fetched and sent to their gateway. Token leak to third party. Three-branch predicate: 1. BASE_URL set -> return without OAuth (custom gateway, never leak token) 2. API_KEY or AUTH_TOKEN set -> return without OAuth (explicit credentials) 3. Otherwise -> OAuth lookup for api.anthropic.com Adds tests/env-isolation.test.ts. Fixes #2375. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(worker): classify Claude SDK HTTP 400 as unrecoverable ClaudeProvider previously had no explicit HTTP 400 handling — the default branch classified all errors as `transient`, so a permanent 400 (e.g., model rejecting an `effort` parameter forwarded from a leaked CLAUDE_CODE_EFFORT_LEVEL) would be retried indefinitely (#1874+ retries observed in one session per #2357). Mirror GeminiProvider/OpenRouterProvider's pattern: classify 400 as `unrecoverable`, 401/403 as `auth_invalid`, 429 as `rate_limit`, default to `transient`. When the 400 body matches the "effort parameter" signature, emit a one-time SDK warn log pointing at the env-leak fix in ~/.claude-mem/.env. Adds tests/claude-provider-error-classifier.test.ts. Fixes #2357. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(chroma): pin onnxruntime>=1.20 + protobuf<7 to fix INVALID_PROTOBUF on macOS arm64 The shipped all-MiniLM-L6-v2 model has pytorch-2.0 IR. chroma-mcp 0.2.6 transitively depends on `chromadb>=1.0.16` which only requires `onnxruntime>=1.14.1` — uv can therefore resolve to an onnxruntime old enough to fail every embedding add with `[ONNXRuntimeError] : 7 : INVALID_PROTOBUF` on macOS arm64 / Python 3.13. Semantic search silently degraded to FTS-only and smart backfill broke (#2371). Path B (override) was required because chroma-mcp 0.2.6 is the latest PyPI release — no upstream bump exists. Inject `--with onnxruntime>=1.20 --with protobuf<7` into the uvx spawn args (both persistent and remote modes). The protobuf cap is essential: forcing only `onnxruntime>=1.20` causes uv to re-resolve and land on protobuf 7.x, which trips opentelemetry's `_pb2` stubs with `TypeError: Descriptors cannot be created directly` because they were generated with protoc <3.19. Capping below 7 lands on protobuf 6.x which opentelemetry tolerates. Verified end-to-end: ONNX model loads, embeddings produce a 384-dim vector, PersistentClient init / add / query roundtrip succeeds: uvx --python 3.13 --with "onnxruntime>=1.20" --with "protobuf<7" \ chroma-mcp==0.2.6 --help # clean # programmatic test: onnxruntime 1.26.0, protobuf 6.33.6, # embedding ok 384, query ok ids=[['1']] Fixes #2371. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(chroma): enforce single chroma-mcp subprocess per worker (#2313) Root cause: every reconnect path in ChromaMcpManager — connectInternal's re-entry, the connect-timeout catch, callTool's transport-error retry, and the transport.onclose handler — used to abandon `this.transport`/`this.client` by calling at most `transport.close()` and nulling the handles. The MCP SDK's StdioClientTransport.close() only signals the direct child (uvx); on Linux the grandchildren (uv -> python -> chroma-mcp) re-parent to init and survive because the SDK does not put the subprocess in its own process group. Each reconnect therefore leaked a full chroma-mcp tree, accumulating 20+ instances per session. Fix: introduce a private disposeCurrentSubprocess() helper that always tree- kills via the existing killProcessTree primitive before nulling the transport reference, and route every "abandon current transport" path (reconnect, connect-timeout, transport error, onclose, stop) through it. The existing `connecting: Promise<void> | null` lock continues to serialize concurrent ensureConnected() callers into a single spawn. Adds tests/services/sync/chroma-mcp-manager-singleton.test.ts covering: - 5 parallel ensureConnected() calls produce exactly one spawn - a transport-error reconnect tree-kills the prior subprocess pid before spawning a replacement - stop() disposes state including any pending connecting promise Manual verification needed on Linux: after a long session with multiple tool uses, `ps aux | grep chroma-mcp | wc -l` should return 1, not 20+. Fixes #2313. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(build): polyfill import.meta.url to __filename in CJS worker bundle The worker bundles ESM dependencies (notably @anthropic-ai/claude-agent-sdk's *.mjs files) into CJS output. Those modules call createRequire(import.meta.url) at module-load time. esbuild's CJS output left this as createRequire(ute.url) — where `ute` is its `import.meta` polyfill `{}` — so `ute.url` was undefined and module-load crashed with: TypeError: The argument 'filename' must be a file URL object, file URL string, or absolute path string. Received undefined code: ERR_INVALID_ARG_VALUE Every Stop hook and every worker subprocess invocation hit this. Fix is the esbuild `define` option mapping `import.meta.url` to `__filename` (provided as a real absolute path by the existing CJS prelude in the banner). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * chore: daily dep bump per CLAUDE.md maintenance policy Root: @anthropic-ai/claude-agent-sdk, @clack/prompts, @types/node, dompurify, postcss, react, react-dom, yaml, zod. plugin/: tree-sitter-cli, zod. openclaw/: @types/node. All patch/minor bumps; no major version changes. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * build: regenerate plugin artifacts after env/chroma/mcp fixes Built artifacts are committed so the marketplace-installable plugin ships with the runtime bundles. Picks up: - d7b145e9 .mcp.json shell-prelude trim drop - a8cbd651 EnvManager BASE_URL block + 3-branch predicate - 8cb73b8c ClaudeProvider HTTP 400 unrecoverable classifier - ecd5b802 ChromaMcpManager onnxruntime/protobuf overrides - c79324ea ChromaMcpManager singleton enforcement - e8376f46 esbuild import.meta.url -> __filename polyfill - a7541d71 daily dep bump Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * build: regenerate plugin artifacts after main merge Bundles now include both v13.0.0 server-beta runtime (server-beta-service.cjs + updated mcp-server.cjs / worker-service.cjs) and this branch's chroma / env / build / Claude SDK fixes. Verified: bun test tests/env-isolation.test.ts \\ tests/claude-provider-error-classifier.test.ts \\ tests/services/sync/chroma-mcp-manager-singleton.test.ts → 13/13 pass. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(review): address CodeRabbit findings on PR #2394 1. scripts/build-hooks.js — `import.meta.url` now maps to a file:// URL (via pathToFileURL(__filename).href in the CJS banner) instead of the raw __filename path. Preserves URL semantics for any bundled ESM dep that does `new URL(rel, import.meta.url)`. createRequire still works. 2. src/shared/EnvManager.ts — added envFilePath() that resolves CLAUDE_MEM_ENV_FILE lazily (falling back to paths.envFile()), and switched internal load/save call sites to use it. ENV_FILE_PATH is kept as a deprecated snapshot for back-compat. Lets tests target a temp file without depending on module-load order. 3. tests/env-isolation.test.ts — redirects to a temp dir via CLAUDE_MEM_ENV_FILE in beforeAll, removes all mutation of the real ~/.claude-mem/.env, and wraps the OAuth-spy assertion in try/finally so the spy is always restored even if the test fails. Verified: bun test tests/env-isolation.test.ts \ tests/claude-provider-error-classifier.test.ts \ tests/services/sync/chroma-mcp-manager-singleton.test.ts → 13/13 pass Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-09 18:05:48 -07:00
parent 13d5fa71c2
commit 5533412984
14 changed files with 1064 additions and 417 deletions
@@ -5,7 +5,7 @@
      "command": "sh",
      "args": [
        "-c",
-        "_C=\"${CLAUDE_CONFIG_DIR:-$HOME/.claude}\"; _E=\"${CLAUDE_PLUGIN_ROOT:-${PLUGIN_ROOT:-}}\"; _P=$({ [ -n \"$_E\" ] && printf '%s\\n' \"$_E\"; printf '%s\\n' \"$PWD/plugin\" \"$PWD\"; ls -dt \"$HOME/.codex/plugins/cache/claude-mem-local/claude-mem\"/[0-9]*/ \"$HOME/.codex/plugins/cache/thedotmack/claude-mem\"/[0-9]*/ \"$_C/plugins/cache/thedotmack/claude-mem\"/[0-9]*/ 2>/dev/null; printf '%s\\n' \"$_C/plugins/marketplaces/thedotmack/plugin\"; } | while IFS= read -r _R; do _R=\"${_R%/}\"; [ -d \"$_R/plugin/scripts\" ] && _Q=\"$_R/plugin\" || _Q=\"$_R\"; [ -f \"$_Q/scripts/mcp-server.cjs\" ] && { printf '%s\\n' \"$_Q\"; break; }; done); [ -n \"$_P\" ] || { echo \"claude-mem: mcp server not found\" >&2; exit 1; }; exec node \"$_P/scripts/mcp-server.cjs\""
+        "_C=\"${CLAUDE_CONFIG_DIR:-$HOME/.claude}\"; _E=\"${CLAUDE_PLUGIN_ROOT:-${PLUGIN_ROOT:-}}\"; _P=$({ [ -n \"$_E\" ] && printf '%s\\n' \"$_E\"; printf '%s\\n' \"$PWD/plugin\" \"$PWD\"; ls -dt \"$HOME/.codex/plugins/cache/claude-mem-local/claude-mem\"/[0-9]*/ \"$HOME/.codex/plugins/cache/thedotmack/claude-mem\"/[0-9]*/ \"$_C/plugins/cache/thedotmack/claude-mem\"/[0-9]*/ 2>/dev/null; printf '%s\\n' \"$_C/plugins/marketplaces/thedotmack/plugin\"; } | while IFS= read -r _R; do [ -d \"$_R/plugin/scripts\" ] && _Q=\"$_R/plugin\" || _Q=\"$_R\"; [ -f \"$_Q/scripts/mcp-server.cjs\" ] && { printf '%s\\n' \"$_Q\"; break; }; done); [ -n \"$_P\" ] || { echo \"claude-mem: mcp server not found\" >&2; exit 1; }; exec node \"$_P/scripts/mcp-server.cjs\""
      ]
    }
  }
@@ -10,7 +10,7 @@
    "test": "tsc && node --test dist/index.test.js"
  },
  "devDependencies": {
-    "@types/node": "^25.6.0",
+    "@types/node": "^25.6.2",
    "typescript": "^6.0.3"
  },
  "openclaw": {
@@ -123,26 +123,26 @@
    "2fa": false
  },
  "dependencies": {
-    "@anthropic-ai/claude-agent-sdk": "^0.2.119",
+    "@anthropic-ai/claude-agent-sdk": "^0.2.138",
    "@better-auth/api-key": "^1.6.9",
-    "@clack/prompts": "^1.2.0",
+    "@clack/prompts": "^1.3.0",
    "@modelcontextprotocol/sdk": "^1.29.0",
    "ansi-to-html": "^0.7.2",
    "better-auth": "^1.6.9",
    "bullmq": "^5.76.6",
    "cors": "^2.8.6",
-    "dompurify": "^3.4.1",
+    "dompurify": "^3.4.2",
    "express": "^5.2.1",
    "glob": "^13.0.6",
    "handlebars": "^4.7.9",
    "ioredis": "^5.10.1",
    "pg": "^8.20.0",
    "picocolors": "^1.1.1",
-    "react": "^19.2.5",
-    "react-dom": "^19.2.5",
+    "react": "^19.2.6",
+    "react-dom": "^19.2.6",
    "shell-quote": "^1.8.3",
-    "yaml": "^2.8.3",
-    "zod": "^4.3.6",
+    "yaml": "^2.8.4",
+    "zod": "^4.4.3",
    "zod-to-json-schema": "^3.25.2"
  },
  "devDependencies": {
@@ -156,7 +156,7 @@
    "@types/cors": "^2.8.19",
    "@types/dompurify": "^3.2.0",
    "@types/express": "^5.0.6",
-    "@types/node": "^25.6.0",
+    "@types/node": "^25.6.2",
    "@types/pg": "^8.20.0",
    "@types/react": "^19.2.14",
    "@types/react-dom": "^19.2.3",
@@ -164,7 +164,7 @@
    "jimp": "^1.6.1",
    "np": "^11.2.0",
    "parse5": "^8.0.1",
-    "postcss": "^8.5.13",
+    "postcss": "^8.5.14",
    "remark-mdx": "^3.1.1",
    "remark-parse": "^11.0.0",
    "tree-sitter-bash": "^0.25.1",
@@ -5,7 +5,7 @@
      "command": "sh",
      "args": [
        "-c",
-        "_C=\"${CLAUDE_CONFIG_DIR:-$HOME/.claude}\"; _E=\"${CLAUDE_PLUGIN_ROOT:-${PLUGIN_ROOT:-}}\"; _P=$({ [ -n \"$_E\" ] && printf '%s\\n' \"$_E\"; printf '%s\\n' \"$PWD/plugin\" \"$PWD\"; ls -dt \"$HOME/.codex/plugins/cache/claude-mem-local/claude-mem\"/[0-9]*/ \"$HOME/.codex/plugins/cache/thedotmack/claude-mem\"/[0-9]*/ \"$_C/plugins/cache/thedotmack/claude-mem\"/[0-9]*/ 2>/dev/null; printf '%s\\n' \"$_C/plugins/marketplaces/thedotmack/plugin\"; } | while IFS= read -r _R; do _R=\"${_R%/}\"; [ -d \"$_R/plugin/scripts\" ] && _Q=\"$_R/plugin\" || _Q=\"$_R\"; [ -f \"$_Q/scripts/mcp-server.cjs\" ] && { printf '%s\\n' \"$_Q\"; break; }; done); [ -n \"$_P\" ] || { echo \"claude-mem: mcp server not found\" >&2; exit 1; }; exec node \"$_P/scripts/mcp-server.cjs\""
+        "_C=\"${CLAUDE_CONFIG_DIR:-$HOME/.claude}\"; _E=\"${CLAUDE_PLUGIN_ROOT:-${PLUGIN_ROOT:-}}\"; _P=$({ [ -n \"$_E\" ] && printf '%s\\n' \"$_E\"; printf '%s\\n' \"$PWD/plugin\" \"$PWD\"; ls -dt \"$HOME/.codex/plugins/cache/claude-mem-local/claude-mem\"/[0-9]*/ \"$HOME/.codex/plugins/cache/thedotmack/claude-mem\"/[0-9]*/ \"$_C/plugins/cache/thedotmack/claude-mem\"/[0-9]*/ 2>/dev/null; printf '%s\\n' \"$_C/plugins/marketplaces/thedotmack/plugin\"; } | while IFS= read -r _R; do [ -d \"$_R/plugin/scripts\" ] && _Q=\"$_R/plugin\" || _Q=\"$_R\"; [ -f \"$_Q/scripts/mcp-server.cjs\" ] && { printf '%s\\n' \"$_Q\"; break; }; done); [ -n \"$_P\" ] || { echo \"claude-mem: mcp server not found\" >&2; exit 1; }; exec node \"$_P/scripts/mcp-server.cjs\""
      ]
    }
  }
@@ -212,7 +212,7 @@ ${f}`}let a=s.lineStart;for(let u=s.lineStart-1;u>=0;u--){let l=i[u].trim();if(l
 ${c}`}var u_=new Set([".js",".jsx",".ts",".tsx",".mjs",".cjs",".py",".pyw",".go",".rs",".rb",".java",".cs",".cpp",".cc",".cxx",".c",".h",".hpp",".hh",".swift",".kt",".kts",".php",".vue",".svelte",".ex",".exs",".lua",".scala",".sc",".sh",".bash",".zsh",".hs",".zig",".css",".scss",".toml",".yml",".yaml",".sql",".md",".mdx"]),Gx=new Set(["node_modules",".git","dist","build",".next","__pycache__",".venv","venv","env",".env","target","vendor",".cache",".turbo","coverage",".nyc_output",".claude",".smart-file-read"]),Kx=512*1024;async function*l_(t,e,r=20,n){if(r<=0)return;let o;try{o=await(0,Ar.readdir)(t,{withFileTypes:!0})}catch(s){S.debug("WORKER",`walkDir: failed to read directory ${t}`,void 0,s instanceof Error?s:void 0);return}for(let s of o){if(s.name.startsWith(".")&&s.name!=="."||Gx.has(s.name))continue;let i=(0,Vn.join)(t,s.name);if(s.isDirectory())yield*l_(i,e,r-1,n);else if(s.isFile()){let a=s.name.slice(s.name.lastIndexOf("."));(u_.has(a)||n&&n.has(a))&&(yield i)}}}async function Jx(t){try{let e=await(0,Ar.stat)(t);if(e.size>Kx||e.size===0)return null;let r=await(0,Ar.readFile)(t,"utf-8");return r.slice(0,1e3).includes("\0")?null:r}catch(e){return S.debug("WORKER",`safeReadFile: failed to read ${t}`,void 0,e instanceof Error?e:void 0),null}}async function d_(t,e,r={}){let n=r.maxResults||20,o=e.toLowerCase(),s=o.split(/[\s_\-./]+/).filter(w=>w.length>0),i=r.projectRoot||t,a=Wn(i),c=new Set;for(let w of Object.values(a.grammars))for(let v of w.extensions)u_.has(v)||c.add(v);let u=[];for await(let w of l_(t,t,20,c.size>0?c:void 0)){if(r.filePattern&&!(0,Vn.relative)(t,w).toLowerCase().includes(r.filePattern.toLowerCase()))continue;let v=await Jx(w);v&&u.push({absolutePath:w,relativePath:(0,Vn.relative)(t,w),content:v})}let l=i_(u,i),d=[],p=[],f=0;for(let[w,v]of l){f+=Bx(v);let k=Ps(w.toLowerCase(),s)>0,_e=[],Ee=(Dt,Qt)=>{for(let ae of Dt){let St=0,Ve="",Cr=Ps(ae.name.toLowerCase(),s);Cr>0&&(St+=Cr*3,Ve="name match"),ae.signature.toLowerCase().includes(o)&&(St+=2,Ve=Ve?`${Ve} + signature`:"signature match"),ae.jsdoc&&ae.jsdoc.toLowerCase().includes(o)&&(St+=1,Ve=Ve?`${Ve} + jsdoc`:"jsdoc match"),St>0&&(k=!0,_e.push({filePath:w,symbolName:Qt?`${Qt}.${ae.name}`:ae.name,kind:ae.kind,signature:ae.signature,jsdoc:ae.jsdoc,lineStart:ae.lineStart,lineEnd:ae.lineEnd,matchReason:Ve})),ae.children&&Ee(ae.children,ae.name)}};Ee(v.symbols),k&&(d.push(v),p.push(..._e))}p.sort((w,v)=>{let x=Ps(w.symbolName.toLowerCase(),s);return Ps(v.symbolName.toLowerCase(),s)-x});let m=p.slice(0,n),_=new Set(m.map(w=>w.filePath)),y=d.filter(w=>_.has(w.filePath)).slice(0,n),b=y.reduce((w,v)=>w+v.foldedTokenEstimate,0);return{foldedFiles:y,matchingSymbols:m,totalFilesScanned:u.length,totalSymbolsFound:f,tokenEstimate:b}}function Ps(t,e){let r=0;for(let n of e)if(t===n)r+=10;else if(t.includes(n))r+=5;else{let o=0,s=0;for(let i of n){let a=t.indexOf(i,o);a!==-1&&(s++,o=a+1)}s===n.length&&(r+=1)}return r}function Bx(t){let e=t.symbols.length;for(let r of t.symbols)r.children&&(e+=r.children.length);return e}function p_(t,e){let r=[];if(r.push(`\u{1F50D} Smart Search: "${e}"`),r.push(`   Scanned ${t.totalFilesScanned} files, found ${t.totalSymbolsFound} symbols`),r.push(`   ${t.matchingSymbols.length} matches across ${t.foldedFiles.length} files (~${t.tokenEstimate} tokens for folded view)`),r.push(""),t.matchingSymbols.length===0)return r.push("   No matching symbols found."),r.join(`
 `);r.push("\u2500\u2500 Matching Symbols \u2500\u2500"),r.push("");for(let n of t.matchingSymbols){if(r.push(`  ${n.kind} ${n.symbolName} (${n.filePath}:${n.lineStart+1})`),r.push(`    ${n.signature}`),n.jsdoc){let o=n.jsdoc.split(`
 `).find(s=>s.replace(/^[\s*/]+/,"").trim().length>0);o&&r.push(`    \u{1F4AC} ${o.replace(/^[\s*/]+/,"").trim()}`)}r.push("")}r.push("\u2500\u2500 Folded File Views \u2500\u2500"),r.push("");for(let n of t.foldedFiles)r.push(Ir(n)),r.push("");return r.push("\u2500\u2500 Actions \u2500\u2500"),r.push("  To see full implementation: use smart_unfold with file path and symbol name"),r.join(`
-`)}var iu=require("node:fs/promises"),zs=require("node:fs"),Qe=require("node:path"),h_=require("node:os"),g_=require("node:url"),cP={},Yx="12.7.5";console.log=(...t)=>{S.error("CONSOLE","Intercepted console output (MCP protocol protection)",void 0,{args:t})};var __=!1,y_=(()=>{if(typeof __dirname<"u")return __dirname;try{return(0,Qe.dirname)((0,g_.fileURLToPath)(cP.url))}catch{return __=!0,process.cwd()}})(),au=(0,Qe.resolve)(y_,"worker-service.cjs");function Xx(){__&&((0,zs.existsSync)(au)||S.error("SYSTEM","mcp-server: dirname resolution failed (both __dirname and import.meta.url are unavailable). Fell back to process.cwd() and the resolved WORKER_SCRIPT_PATH does not exist. This is the actual problem \u2014 the worker bundle is fine, but mcp-server cannot locate it. Worker auto-start will fail until the dirname-resolution path is fixed.",{workerScriptPath:au,mcpServerDir:y_}))}var f_={search:"/api/search",timeline:"/api/timeline"};async function su(t,e){S.debug("SYSTEM","\u2192 Worker API",void 0,{endpoint:t,params:e});let r=new URLSearchParams;for(let[o,s]of Object.entries(e))s!=null&&r.append(o,String(s));let n=`${t}?${r}`;try{let o=await $s(n);if(!o.ok){let i=await o.text();throw new Error(`Worker API error (${o.status}): ${i}`)}let s=await o.json();return S.debug("SYSTEM","\u2190 Worker API success",void 0,{endpoint:t}),s}catch(o){return S.error("SYSTEM","\u2190 Worker API error",{endpoint:t},o instanceof Error?o:new Error(String(o))),{content:[{type:"text",text:`Error calling Worker API: ${o instanceof Error?o.message:String(o)}`}],isError:!0}}}async function Qx(t,e){let r=await $s(t,{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify(e)});if(!r.ok){let o=await r.text();throw new Error(`Worker API error (${r.status}): ${o}`)}let n=await r.json();return S.debug("HTTP","Worker API success (POST)",void 0,{endpoint:t}),{content:[{type:"text",text:JSON.stringify(n,null,2)}]}}async function Mr(t,e){S.debug("HTTP","Worker API request (POST)",void 0,{endpoint:t});try{return await Qx(t,e)}catch(r){return S.error("HTTP","Worker API error (POST)",{endpoint:t},r instanceof Error?r:new Error(String(r))),{content:[{type:"text",text:`Error calling Worker API: ${r instanceof Error?r.message:String(r)}`}],isError:!0}}}async function eP(){try{return(await $s("/api/health")).ok}catch(t){return S.debug("SYSTEM","Worker health check failed",{},t instanceof Error?t:new Error(String(t))),!1}}async function tP(){if(await eP())return!0;S.warn("SYSTEM","Worker not available, attempting auto-start for MCP client"),Xx();try{let t=Jc(),e=await Kg(t,au);return e==="dead"&&S.error("SYSTEM","Worker auto-start failed \u2014 MCP tools that require the worker (search, timeline, get_observations) will fail until the worker is running. Check earlier log lines for the specific failure reason (Bun not found, missing worker bundle, port conflict, etc.)."),e!=="dead"}catch(t){return S.error("SYSTEM","Worker auto-start threw \u2014 MCP tools that require the worker (search, timeline, get_observations) will fail until the worker is running.",void 0,t instanceof Error?t:new Error(String(t))),!1}}var S_=[{name:"__IMPORTANT",description:`3-LAYER WORKFLOW (ALWAYS FOLLOW):
+`)}var iu=require("node:fs/promises"),zs=require("node:fs"),Qe=require("node:path"),h_=require("node:os"),g_=require("node:url"),cP={},Yx="13.0.0";console.log=(...t)=>{S.error("CONSOLE","Intercepted console output (MCP protocol protection)",void 0,{args:t})};var __=!1,y_=(()=>{if(typeof __dirname<"u")return __dirname;try{return(0,Qe.dirname)((0,g_.fileURLToPath)(cP.url))}catch{return __=!0,process.cwd()}})(),au=(0,Qe.resolve)(y_,"worker-service.cjs");function Xx(){__&&((0,zs.existsSync)(au)||S.error("SYSTEM","mcp-server: dirname resolution failed (both __dirname and import.meta.url are unavailable). Fell back to process.cwd() and the resolved WORKER_SCRIPT_PATH does not exist. This is the actual problem \u2014 the worker bundle is fine, but mcp-server cannot locate it. Worker auto-start will fail until the dirname-resolution path is fixed.",{workerScriptPath:au,mcpServerDir:y_}))}var f_={search:"/api/search",timeline:"/api/timeline"};async function su(t,e){S.debug("SYSTEM","\u2192 Worker API",void 0,{endpoint:t,params:e});let r=new URLSearchParams;for(let[o,s]of Object.entries(e))s!=null&&r.append(o,String(s));let n=`${t}?${r}`;try{let o=await $s(n);if(!o.ok){let i=await o.text();throw new Error(`Worker API error (${o.status}): ${i}`)}let s=await o.json();return S.debug("SYSTEM","\u2190 Worker API success",void 0,{endpoint:t}),s}catch(o){return S.error("SYSTEM","\u2190 Worker API error",{endpoint:t},o instanceof Error?o:new Error(String(o))),{content:[{type:"text",text:`Error calling Worker API: ${o instanceof Error?o.message:String(o)}`}],isError:!0}}}async function Qx(t,e){let r=await $s(t,{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify(e)});if(!r.ok){let o=await r.text();throw new Error(`Worker API error (${r.status}): ${o}`)}let n=await r.json();return S.debug("HTTP","Worker API success (POST)",void 0,{endpoint:t}),{content:[{type:"text",text:JSON.stringify(n,null,2)}]}}async function Mr(t,e){S.debug("HTTP","Worker API request (POST)",void 0,{endpoint:t});try{return await Qx(t,e)}catch(r){return S.error("HTTP","Worker API error (POST)",{endpoint:t},r instanceof Error?r:new Error(String(r))),{content:[{type:"text",text:`Error calling Worker API: ${r instanceof Error?r.message:String(r)}`}],isError:!0}}}async function eP(){try{return(await $s("/api/health")).ok}catch(t){return S.debug("SYSTEM","Worker health check failed",{},t instanceof Error?t:new Error(String(t))),!1}}async function tP(){if(await eP())return!0;S.warn("SYSTEM","Worker not available, attempting auto-start for MCP client"),Xx();try{let t=Jc(),e=await Kg(t,au);return e==="dead"&&S.error("SYSTEM","Worker auto-start failed \u2014 MCP tools that require the worker (search, timeline, get_observations) will fail until the worker is running. Check earlier log lines for the specific failure reason (Bun not found, missing worker bundle, port conflict, etc.)."),e!=="dead"}catch(t){return S.error("SYSTEM","Worker auto-start threw \u2014 MCP tools that require the worker (search, timeline, get_observations) will fail until the worker is running.",void 0,t instanceof Error?t:new Error(String(t))),!1}}var S_=[{name:"__IMPORTANT",description:`3-LAYER WORKFLOW (ALWAYS FOLLOW):
 1. search(query) \u2192 Get index with IDs (~50-100 tokens/result)
 2. timeline(anchor=ID) \u2192 Get context around interesting results
 3. get_observations([IDs]) \u2192 Fetch full details ONLY for filtered IDs
@@ -151,13 +151,19 @@ async function buildHooks() {
        'onnxruntime-node'
      ],
      define: {
-        '__DEFAULT_PACKAGE_VERSION__': `"${version}"`
+        '__DEFAULT_PACKAGE_VERSION__': `"${version}"`,
+        // Polyfill import.meta.url for ESM deps bundled into CJS output.
+        // @anthropic-ai/claude-agent-sdk's *.mjs files use createRequire(import.meta.url)
+        // and `new URL(rel, import.meta.url)`. We map import.meta.url to a file:// URL
+        // (not the raw __filename path) so URL construction preserves its semantics.
+        'import.meta.url': '__IMPORT_META_URL__'
      },
      banner: {
        js: [
          '#!/usr/bin/env bun',
          'var __filename = __filename || require("node:path").resolve(process.argv[1] || "");',
-          'var __dirname = __dirname || require("node:path").dirname(__filename);'
+          'var __dirname = __dirname || require("node:path").dirname(__filename);',
+          'var __IMPORT_META_URL__ = require("node:url").pathToFileURL(__filename).href;'
        ].join('\n')
      }
    });
@@ -23,6 +23,26 @@ const CHROMA_SUPERVISOR_ID = 'chroma-mcp';

 const CHROMA_MCP_PINNED_VERSION = '0.2.6';

+// Override transitive dep resolutions for chroma-mcp 0.2.6 (issue #2371).
+//
+// Why onnxruntime>=1.20: the shipped all-MiniLM-L6-v2 model has pytorch-2.0
+// IR. Older onnxruntime versions can't parse it and fail every embedding
+// add with `[ONNXRuntimeError] : 7 : INVALID_PROTOBUF`. uv may otherwise
+// resolve to a too-old onnxruntime on macOS arm64 / Python 3.13 depending
+// on cache state, so we force a floor.
+//
+// Why protobuf<7: protobuf 7.x's stricter generated-file check rejects
+// opentelemetry's _pb2 stubs (generated with protoc <3.19), throwing
+// `TypeError: Descriptors cannot be created directly` at chromadb import.
+// Capping below 7 lands on protobuf 6.x which opentelemetry tolerates.
+//
+// These pins are runtime-only (uvx --with) so we don't have to fork
+// chroma-mcp upstream — they apply only to claude-mem's spawned subprocess.
+const CHROMA_MCP_DEP_OVERRIDES: ReadonlyArray<string> = [
+  'onnxruntime>=1.20',
+  'protobuf<7',
+];
+
 export class ChromaMcpManager {
  private static instance: ChromaMcpManager | null = null;
  private client: Client | null = null;
@@ -72,15 +92,14 @@ export class ChromaMcpManager {
  }

  private async connectInternal(): Promise<void> {
-    if (this.transport) {
-      try { await this.transport.close(); } catch { /* already dead */ }
-    }
-    if (this.client) {
-      try { await this.client.close(); } catch { /* already dead */ }
-    }
-    this.client = null;
-    this.transport = null;
-    this.connected = false;
+    // Singleton invariant (#2313): kill any pre-existing chroma-mcp subprocess
+    // tree before spawning a new one. The MCP SDK's transport.close() only
+    // signals the direct child (uvx); on Linux the grandchildren (uv, python,
+    // chroma-mcp) get re-parented to init and survive, accumulating 20+
+    // instances per session if reconnects fire repeatedly. Reuse the same
+    // tree-kill primitive used by stop() so reconnect can never leave
+    // orphans behind.
+    await this.disposeCurrentSubprocess();

    const commandArgs = this.buildCommandArgs();
    const spawnEnvironment = this.getSpawnEnv();
@@ -121,14 +140,12 @@ export class ChromaMcpManager {
      await Promise.race([mcpConnectionPromise, timeoutPromise]);
    } catch (connectionError) {
      clearTimeout(timeoutId!);
-      logger.warn('CHROMA_MCP', 'Connection failed, killing subprocess to prevent zombie', {
+      logger.warn('CHROMA_MCP', 'Connection failed, killing subprocess tree to prevent zombie', {
        error: connectionError instanceof Error ? connectionError.message : String(connectionError)
      });
-      try { await this.transport.close(); } catch { /* best effort */ }
-      try { await this.client.close(); } catch { /* best effort */ }
-      this.client = null;
-      this.transport = null;
-      this.connected = false;
+      // Tree-kill (not just transport.close) so failed-connect descendants
+      // can't survive on Linux (#2313).
+      await this.disposeCurrentSubprocess();
      throw connectionError;
    }
    clearTimeout(timeoutId!);
@@ -139,6 +156,7 @@ export class ChromaMcpManager {
    logger.info('CHROMA_MCP', 'Connected to chroma-mcp successfully');

    const currentTransport = this.transport;
+    const currentTrackedPid = (this.transport as unknown as { _process?: ChildProcess })._process?.pid;
    this.transport.onclose = () => {
      if (this.transport !== currentTransport) {
        logger.debug('CHROMA_MCP', 'Ignoring stale onclose from previous transport');
@@ -150,6 +168,20 @@ export class ChromaMcpManager {
      this.client = null;
      this.transport = null;
      this.lastConnectionFailureTimestamp = Date.now();
+
+      // Direct child (uvx) emitted close, but on Linux the grandchildren
+      // (uv/python/chroma-mcp) often outlive their parent because MCP SDK
+      // does not use process groups. Sweep the descendant tree using the
+      // captured PID — best-effort; pgrep returns nothing if everything
+      // already exited (#2313).
+      if (currentTrackedPid) {
+        ChromaMcpManager.killProcessTree(currentTrackedPid).catch((error) => {
+          logger.debug('CHROMA_MCP', 'Background tree-kill after onclose finished (best-effort)', {
+            pid: currentTrackedPid,
+            error: error instanceof Error ? error.message : String(error)
+          });
+        });
+      }
    };
  }

@@ -158,6 +190,8 @@ export class ChromaMcpManager {
    const chromaMode = settings.CLAUDE_MEM_CHROMA_MODE || 'local';
    const pythonVersion = process.env.CLAUDE_MEM_PYTHON_VERSION || settings.CLAUDE_MEM_PYTHON_VERSION || '3.13';

+    const depOverrideFlags = CHROMA_MCP_DEP_OVERRIDES.flatMap(spec => ['--with', spec]);
+
    if (chromaMode === 'remote') {
      const chromaHost = settings.CLAUDE_MEM_CHROMA_HOST || '127.0.0.1';
      const chromaPort = settings.CLAUDE_MEM_CHROMA_PORT || '8000';
@@ -168,6 +202,7 @@ export class ChromaMcpManager {

      const args = [
        '--python', pythonVersion,
+        ...depOverrideFlags,
        `chroma-mcp==${CHROMA_MCP_PINNED_VERSION}`,
        '--client-type', 'http',
        '--host', chromaHost,
@@ -193,6 +228,7 @@ export class ChromaMcpManager {

    return [
      '--python', pythonVersion,
+      ...depOverrideFlags,
      `chroma-mcp==${CHROMA_MCP_PINNED_VERSION}`,
      '--client-type', 'persistent',
      '--data-dir', DEFAULT_CHROMA_DATA_DIR.replace(/\\/g, '/')
@@ -213,14 +249,15 @@ export class ChromaMcpManager {
        arguments: toolArguments
      });
    } catch (transportError) {
-      this.connected = false;
-      this.client = null;
-      this.transport = null;
-
      logger.warn('CHROMA_MCP', `Transport error during "${toolName}", reconnecting and retrying once`, {
        error: transportError instanceof Error ? transportError.message : String(transportError)
      });

+      // Tree-kill the dying subprocess before reconnect. Previously this path
+      // just nulled the handle, which on Linux leaks the uv/python/chroma-mcp
+      // descendants every time a transport error happens (#2313).
+      await this.disposeCurrentSubprocess();
+
      try {
        await this.ensureConnected();
        result = await this.client!.callTool({
@@ -328,6 +365,53 @@ export class ChromaMcpManager {
    }
  }

+  /**
+   * Singleton enforcement helper (#2313): tree-kill the currently tracked
+   * chroma-mcp subprocess and reset all state so the next spawn starts clean.
+   *
+   * Why this is the singleton invariant: every code path that intends to
+   * abandon `this.transport` / `this.client` (reconnect, transport error,
+   * connect-timeout, onclose, stop()) MUST funnel through here. The MCP
+   * SDK's transport.close() only signals the direct child (uvx); on Linux
+   * the grandchildren (uv, python, chroma-mcp) re-parent to init and
+   * accumulate. Calling killProcessTree() against the captured PID before
+   * we drop the reference is the only way to guarantee at most one
+   * chroma-mcp subprocess tree exists per worker process.
+   *
+   * Idempotent and best-effort — safe to call when there is no active
+   * subprocess (no-op in that case).
+   */
+  private async disposeCurrentSubprocess(): Promise<void> {
+    const chromaProcess = (this.transport as unknown as { _process?: ChildProcess })?._process;
+    const trackedPid = chromaProcess?.pid;
+
+    if (trackedPid) {
+      try {
+        await ChromaMcpManager.killProcessTree(trackedPid);
+      } catch (error) {
+        logger.warn('CHROMA_MCP', 'failed to kill prior chroma-mcp tree (best-effort)', {
+          pid: trackedPid,
+          error: error instanceof Error ? error.message : String(error)
+        });
+      }
+    }
+
+    if (this.transport) {
+      try { await this.transport.close(); } catch { /* already dead */ }
+    }
+    if (this.client) {
+      try { await this.client.close(); } catch { /* already dead */ }
+    }
+
+    if (trackedPid) {
+      getSupervisor().unregisterProcess(CHROMA_SUPERVISOR_ID);
+    }
+
+    this.client = null;
+    this.transport = null;
+    this.connected = false;
+  }
+
  /**
   * Gracefully stop the MCP connection and kill the chroma-mcp subprocess tree.
   *
@@ -341,34 +425,15 @@ export class ChromaMcpManager {
   * pattern from shutdown.ts (Principle 5: OS-supervised teardown).
   */
  async stop(): Promise<void> {
-    if (!this.client) {
+    if (!this.client && !this.transport) {
      logger.debug('CHROMA_MCP', 'No active MCP connection to stop');
+      this.connecting = null;
      return;
    }

    logger.info('CHROMA_MCP', 'Stopping chroma-mcp MCP connection');

-    // Kill the entire process tree before closing the MCP client so
-    // descendants (uv, python, chroma-mcp) don't become orphans.
-    const chromaProcess = (this.transport as unknown as { _process?: ChildProcess })?._process;
-    if (chromaProcess?.pid) {
-      await ChromaMcpManager.killProcessTree(chromaProcess.pid);
-    }
-
-    try {
-      await this.client.close();
-    } catch (error) {
-      if (error instanceof Error) {
-        logger.debug('CHROMA_MCP', 'Error during client close (subprocess may already be dead)', {}, error);
-      } else {
-        logger.debug('CHROMA_MCP', 'Error during client close (subprocess may already be dead)', { error: String(error) });
-      }
-    }
-
-    getSupervisor().unregisterProcess(CHROMA_SUPERVISOR_ID);
-    this.client = null;
-    this.transport = null;
-    this.connected = false;
+    await this.disposeCurrentSubprocess();
    this.connecting = null;

    logger.info('CHROMA_MCP', 'chroma-mcp MCP connection stopped');
@@ -27,6 +27,19 @@ import {
 import { query } from '@anthropic-ai/claude-agent-sdk';
 import { ClassifiedProviderError } from './provider-errors.js';

+/**
+ * Module-scoped guard so the "effort parameter" hint only fires once per
+ * worker process. The underlying cause (a leaked CLAUDE_CODE_EFFORT_LEVEL in
+ * ~/.claude-mem/.env, see #2357) is environmental — re-logging it on every
+ * SDK call would spam the logs without adding signal.
+ *
+ * Exported solely for tests to reset the latch between cases.
+ */
+let effortHintLogged = false;
+export function __resetEffortHintLatchForTesting(): void {
+  effortHintLogged = false;
+}
+
 /**
 * Classify a ClaudeProvider error (executable spawn failures, SDK errors,
 * Anthropic API errors). Provider-specific because it relies on:
@@ -36,7 +49,7 @@ import { ClassifiedProviderError } from './provider-errors.js';
 */
 export function classifyClaudeError(err: unknown): ClassifiedProviderError {
  const message = err instanceof Error ? err.message : String(err);
-  const errAny = err as { name?: string; status?: number; error?: { type?: string } };
+  const errAny = err as { name?: string; status?: number; error?: { type?: string }; body?: unknown };

  // Executable / spawn issues — unrecoverable, no point retrying.
  if (
@@ -88,6 +101,39 @@ export function classifyClaudeError(err: unknown): ClassifiedProviderError {
    return new ClassifiedProviderError(message, { kind: 'unrecoverable', cause: err });
  }

+  // HTTP 400 from the Anthropic SDK — bad request, never recoverable. Mirrors
+  // the pattern in GeminiProvider.classifyGeminiError / classifyOpenRouterError
+  // (see #2357: the SDK forwards `effort` to the Messages API when
+  // CLAUDE_CODE_EFFORT_LEVEL leaks into the subprocess env, and models like
+  // Haiku/Sonnet 4.5 reject with 400 — without this branch the default
+  // `transient` classification retried indefinitely).
+  if (errAny.status === 400) {
+    // Inspect both the message and any structured body for the effort marker.
+    const bodyText = (() => {
+      const body = errAny.body;
+      if (typeof body === 'string') return body;
+      if (body && typeof body === 'object') {
+        try { return JSON.stringify(body); } catch { return ''; }
+      }
+      return '';
+    })();
+    const haystack = `${message}\n${bodyText}`;
+    if (/effort parameter/i.test(haystack) && !effortHintLogged) {
+      effortHintLogged = true;
+      logger.warn(
+        'SDK',
+        'Anthropic API rejected request with HTTP 400: this model does not support the `effort` parameter. ' +
+          'CLAUDE_CODE_EFFORT_LEVEL is likely leaking into the SDK subprocess env via ~/.claude-mem/.env — ' +
+          'remove it or scope it to models that support effort. See https://github.com/thedotmack/claude-mem/issues/2357.',
+        { status: 400 }
+      );
+    }
+    return new ClassifiedProviderError(
+      message || 'Anthropic bad request (status 400)',
+      { kind: 'unrecoverable', cause: err },
+    );
+  }
+
  // Server errors → transient.
  if (typeof errAny.status === 'number' && errAny.status >= 500 && errAny.status < 600) {
    return new ClassifiedProviderError(message, { kind: 'transient', cause: err });
@@ -9,7 +9,16 @@ import {
  type OAuthTokenResult,
 } from './oauth-token.js';

-export const ENV_FILE_PATH = paths.envFile();
+// Resolved lazily so tests (and any rare runtime path-overrides) can target a
+// temp file via CLAUDE_MEM_ENV_FILE without depending on module-load order.
+// Production callers see the canonical ~/.claude-mem/.env path through
+// paths.envFile() unchanged.
+export function envFilePath(): string {
+  return process.env.CLAUDE_MEM_ENV_FILE ?? paths.envFile();
+}
+
+/** @deprecated Prefer envFilePath(); kept as a snapshot for back-compat. */
+export const ENV_FILE_PATH = envFilePath();

 const BLOCKED_ENV_VARS = [
  'ANTHROPIC_API_KEY',       // Issue #733: Prevent auto-discovery from project .env files
@@ -17,6 +26,10 @@ const BLOCKED_ENV_VARS = [
                             // shell would otherwise short-circuit OAuth lookup at spawn time.
                             // The fresh token from ~/.claude-mem/.env is re-injected below
                             // when explicit gateway credentials are configured.
+  'ANTHROPIC_BASE_URL',      // Issue #2375: same leak class as AUTH_TOKEN. A leaked BASE_URL
+                             // alone (no token) was enough to trigger the OAuth-skip path,
+                             // sending the subprocess to a proxy with no credentials.
+                             // Re-injected from ~/.claude-mem/.env when configured.
  'CLAUDECODE',              // Prevent "cannot be launched inside another Claude Code session" error
  'CLAUDE_CODE_OAUTH_TOKEN', // Issue #2215: prevent stale parent-process token from leaking into
                             // isolated env. The fresh token is read from the keychain at spawn
@@ -77,12 +90,13 @@ function serializeEnvFile(env: Record<string, string>): string {
 }

 export function loadClaudeMemEnv(): ClaudeMemEnv {
-  if (!existsSync(ENV_FILE_PATH)) {
+  const envFile = envFilePath();
+  if (!existsSync(envFile)) {
    return {};
  }

  try {
-    const content = readFileSync(ENV_FILE_PATH, 'utf-8');
+    const content = readFileSync(envFile, 'utf-8');
    const parsed = parseEnvFile(content);

    const result: ClaudeMemEnv = {};
@@ -94,12 +108,13 @@ export function loadClaudeMemEnv(): ClaudeMemEnv {

    return result;
  } catch (error: unknown) {
-    logger.warn('ENV', 'Failed to load .env file', { path: ENV_FILE_PATH }, error instanceof Error ? error : new Error(String(error)));
+    logger.warn('ENV', 'Failed to load .env file', { path: envFile }, error instanceof Error ? error : new Error(String(error)));
    return {};
  }
 }

 export function saveClaudeMemEnv(env: ClaudeMemEnv): void {
+  const envFile = envFilePath();
  let existing: Record<string, string> = {};
  try {
    if (!existsSync(paths.dataDir())) {
@@ -107,8 +122,8 @@ export function saveClaudeMemEnv(env: ClaudeMemEnv): void {
    }
    chmodSync(paths.dataDir(), 0o700);

-    existing = existsSync(ENV_FILE_PATH)
-      ? parseEnvFile(readFileSync(ENV_FILE_PATH, 'utf-8'))
+    existing = existsSync(envFile)
+      ? parseEnvFile(readFileSync(envFile, 'utf-8'))
      : {};
  } catch (error) {
    const normalizedError = error instanceof Error ? error : new Error(String(error));
@@ -155,10 +170,10 @@ export function saveClaudeMemEnv(env: ClaudeMemEnv): void {
  }

  try {
-    writeFileSync(ENV_FILE_PATH, serializeEnvFile(updated), { encoding: 'utf-8', mode: 0o600 });
-    chmodSync(ENV_FILE_PATH, 0o600);
+    writeFileSync(envFile, serializeEnvFile(updated), { encoding: 'utf-8', mode: 0o600 });
+    chmodSync(envFile, 0o600);
  } catch (error: unknown) {
-    logger.error('ENV', 'Failed to save .env file', { path: ENV_FILE_PATH }, error instanceof Error ? error : new Error(String(error)));
+    logger.error('ENV', 'Failed to save .env file', { path: envFile }, error instanceof Error ? error : new Error(String(error)));
    throw error;
  }
 }
@@ -230,15 +245,17 @@ export async function buildIsolatedEnvWithFreshOAuth(

  if (!includeCredentials) return isolatedEnv;

-  // If the user already configured explicit Anthropic/gateway credentials in
-  // ~/.claude-mem/.env, honor those and skip OAuth lookup entirely. A bare
-  // ANTHROPIC_BASE_URL counts because gateways may be tokenless, and falling
-  // back to OAuth would silently route requests to api.anthropic.com.
-  if (
-    isolatedEnv.ANTHROPIC_API_KEY ||
-    isolatedEnv.ANTHROPIC_BASE_URL ||
-    isolatedEnv.ANTHROPIC_AUTH_TOKEN
-  ) {
+  // Custom gateway: never inject OAuth (would leak the user's Anthropic OAuth
+  // token to a third-party gateway). The user must explicitly configure a
+  // gateway-appropriate token in ~/.claude-mem/.env if their gateway requires
+  // one. A bare BASE_URL with no token = tokenless gateway (e.g. mTLS at the
+  // network boundary).
+  if (isolatedEnv.ANTHROPIC_BASE_URL) {
+    clearStaleMarker();
+    return isolatedEnv;
+  }
+  // Direct API with explicit credentials: skip OAuth lookup.
+  if (isolatedEnv.ANTHROPIC_API_KEY || isolatedEnv.ANTHROPIC_AUTH_TOKEN) {
    clearStaleMarker();
    return isolatedEnv;
  }
@@ -0,0 +1,122 @@
+import { describe, it, expect, beforeEach, afterEach, spyOn } from 'bun:test';
+
+import {
+  classifyClaudeError,
+  __resetEffortHintLatchForTesting,
+} from '../src/services/worker/ClaudeProvider.js';
+import { isClassified } from '../src/services/worker/provider-errors.js';
+import { logger } from '../src/utils/logger.js';
+
+/**
+ * Tests for HTTP 400 classification in ClaudeProvider's classifyClaudeError.
+ *
+ * Regression coverage for #2357: ClaudeProvider previously had no explicit
+ * HTTP 400 handling, so the default branch classified all 400s as `transient`
+ * and the retry loop would hammer a permanent error indefinitely (e.g. when
+ * CLAUDE_CODE_EFFORT_LEVEL leaks into the SDK subprocess and the model
+ * rejects the `effort` parameter).
+ */
+describe('classifyClaudeError — HTTP 400 handling (#2357)', () => {
+  let warnSpy: ReturnType<typeof spyOn>;
+
+  beforeEach(() => {
+    __resetEffortHintLatchForTesting();
+    warnSpy = spyOn(logger, 'warn').mockImplementation(() => {});
+  });
+
+  afterEach(() => {
+    warnSpy.mockRestore();
+    __resetEffortHintLatchForTesting();
+  });
+
+  it('classifies 400 with "effort parameter" body as unrecoverable AND logs an SDK warn once', () => {
+    const sdkErr = Object.assign(
+      new Error('This model does not support the effort parameter.'),
+      { status: 400 },
+    );
+
+    const classified = classifyClaudeError(sdkErr);
+
+    expect(isClassified(classified)).toBe(true);
+    expect(classified.kind).toBe('unrecoverable');
+    expect(warnSpy).toHaveBeenCalledTimes(1);
+    // First positional arg of logger.warn is the component category.
+    const [component, hintMessage] = warnSpy.mock.calls[0] as [string, string, ...unknown[]];
+    expect(component).toBe('SDK');
+    expect(hintMessage).toMatch(/effort/i);
+    expect(hintMessage).toMatch(/2357/);
+  });
+
+  it('classifies 400 with effort marker in a structured body field', () => {
+    const sdkErr = Object.assign(
+      new Error('Bad request'),
+      {
+        status: 400,
+        body: { error: { message: 'This model does not support the effort parameter.' } },
+      },
+    );
+
+    const classified = classifyClaudeError(sdkErr);
+
+    expect(classified.kind).toBe('unrecoverable');
+    expect(warnSpy).toHaveBeenCalledTimes(1);
+  });
+
+  it('classifies 400 without effort body as unrecoverable WITHOUT firing the effort hint', () => {
+    const sdkErr = Object.assign(
+      new Error('some other 400 error'),
+      { status: 400 },
+    );
+
+    const classified = classifyClaudeError(sdkErr);
+
+    expect(classified.kind).toBe('unrecoverable');
+    expect(warnSpy).not.toHaveBeenCalled();
+  });
+
+  it('throttles the effort hint to one log per process even on repeated 400s', () => {
+    const sdkErr = Object.assign(
+      new Error('This model does not support the effort parameter.'),
+      { status: 400 },
+    );
+
+    for (let i = 0; i < 5; i++) {
+      const classified = classifyClaudeError(sdkErr);
+      expect(classified.kind).toBe('unrecoverable');
+    }
+
+    expect(warnSpy).toHaveBeenCalledTimes(1);
+  });
+});
+
+describe('classifyClaudeError — sibling status codes (regression sanity)', () => {
+  let warnSpy: ReturnType<typeof spyOn>;
+
+  beforeEach(() => {
+    __resetEffortHintLatchForTesting();
+    warnSpy = spyOn(logger, 'warn').mockImplementation(() => {});
+  });
+
+  afterEach(() => {
+    warnSpy.mockRestore();
+    __resetEffortHintLatchForTesting();
+  });
+
+  it('classifies status=401 as auth_invalid', () => {
+    const sdkErr = Object.assign(new Error('unauthorized'), { status: 401 });
+    const classified = classifyClaudeError(sdkErr);
+    expect(classified.kind).toBe('auth_invalid');
+  });
+
+  it('classifies status=429 as rate_limit', () => {
+    const sdkErr = Object.assign(new Error('rate limited'), { status: 429 });
+    const classified = classifyClaudeError(sdkErr);
+    expect(classified.kind).toBe('rate_limit');
+  });
+
+  it('classifies a network error with no status as transient', () => {
+    const networkErr = new Error('ECONNRESET: socket hang up');
+    const classified = classifyClaudeError(networkErr);
+    expect(classified.kind).toBe('transient');
+  });
+});
@@ -0,0 +1,155 @@
+import { describe, it, expect, beforeAll, afterAll, beforeEach, afterEach, spyOn } from 'bun:test';
+import * as fs from 'fs';
+import { tmpdir } from 'os';
+import { join } from 'path';
+import {
+  envFilePath,
+  buildIsolatedEnv,
+  buildIsolatedEnvWithFreshOAuth,
+} from '../src/shared/EnvManager.js';
+import * as oauthToken from '../src/shared/oauth-token.js';
+
+/**
+ * Tests for issue #2375: ANTHROPIC_BASE_URL must not leak from the parent
+ * shell into the spawned worker's isolatedEnv, AND the OAuth-skip predicate
+ * must not inject the user's Anthropic OAuth token onto a custom gateway URL
+ * (which would be a token leak to a third party).
+ *
+ * Redirect EnvManager to a per-suite temp file via CLAUDE_MEM_ENV_FILE so
+ * the user's real ~/.claude-mem/.env is never read or mutated even if a test
+ * fails mid-flight. envFilePath() resolves the override on every call, so
+ * this works regardless of the order other tests imported the module.
+ */
+
+const TEST_DATA_DIR = fs.mkdtempSync(join(tmpdir(), 'claude-mem-env-isolation-'));
+const TEST_ENV_FILE = join(TEST_DATA_DIR, '.env');
+const ORIGINAL_ENV_FILE = process.env.CLAUDE_MEM_ENV_FILE;
+
+const ORIGINAL_BASE_URL = process.env.ANTHROPIC_BASE_URL;
+const ORIGINAL_API_KEY = process.env.ANTHROPIC_API_KEY;
+const ORIGINAL_AUTH_TOKEN = process.env.ANTHROPIC_AUTH_TOKEN;
+const ORIGINAL_OAUTH_TOKEN = process.env.CLAUDE_CODE_OAUTH_TOKEN;
+
+function clearEnvFile(): void {
+  if (fs.existsSync(TEST_ENV_FILE)) {
+    fs.unlinkSync(TEST_ENV_FILE);
+  }
+}
+
+function clearAnthropicEnv(): void {
+  delete process.env.ANTHROPIC_BASE_URL;
+  delete process.env.ANTHROPIC_API_KEY;
+  delete process.env.ANTHROPIC_AUTH_TOKEN;
+  delete process.env.CLAUDE_CODE_OAUTH_TOKEN;
+}
+
+function restoreOriginalEnv(): void {
+  if (ORIGINAL_BASE_URL === undefined) {
+    delete process.env.ANTHROPIC_BASE_URL;
+  } else {
+    process.env.ANTHROPIC_BASE_URL = ORIGINAL_BASE_URL;
+  }
+  if (ORIGINAL_API_KEY === undefined) {
+    delete process.env.ANTHROPIC_API_KEY;
+  } else {
+    process.env.ANTHROPIC_API_KEY = ORIGINAL_API_KEY;
+  }
+  if (ORIGINAL_AUTH_TOKEN === undefined) {
+    delete process.env.ANTHROPIC_AUTH_TOKEN;
+  } else {
+    process.env.ANTHROPIC_AUTH_TOKEN = ORIGINAL_AUTH_TOKEN;
+  }
+  if (ORIGINAL_OAUTH_TOKEN === undefined) {
+    delete process.env.CLAUDE_CODE_OAUTH_TOKEN;
+  } else {
+    process.env.CLAUDE_CODE_OAUTH_TOKEN = ORIGINAL_OAUTH_TOKEN;
+  }
+}
+
+describe('Issue #2375: ANTHROPIC_BASE_URL env-var isolation', () => {
+  beforeAll(() => {
+    fs.mkdirSync(TEST_DATA_DIR, { recursive: true, mode: 0o700 });
+    process.env.CLAUDE_MEM_ENV_FILE = TEST_ENV_FILE;
+    expect(envFilePath()).toBe(TEST_ENV_FILE);
+  });
+
+  afterAll(() => {
+    fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
+    if (ORIGINAL_ENV_FILE === undefined) {
+      delete process.env.CLAUDE_MEM_ENV_FILE;
+    } else {
+      process.env.CLAUDE_MEM_ENV_FILE = ORIGINAL_ENV_FILE;
+    }
+  });
+
+  beforeEach(() => {
+    clearEnvFile();
+    clearAnthropicEnv();
+  });
+
+  afterEach(() => {
+    clearEnvFile();
+    restoreOriginalEnv();
+  });
+
+  it('leaked ANTHROPIC_BASE_URL is stripped from isolatedEnv', () => {
+    // No .env file exists. The parent shell sets a stray ANTHROPIC_BASE_URL —
+    // this MUST NOT propagate into the subprocess isolatedEnv, because doing
+    // so used to trigger the OAuth-skip path and leave the worker with no
+    // credentials at all.
+    process.env.ANTHROPIC_BASE_URL = 'https://shouldnotleak.example';
+
+    const result = buildIsolatedEnv();
+
+    expect(result.ANTHROPIC_BASE_URL).toBeUndefined();
+  });
+
+  it('~/.claude-mem/.env BASE_URL + AUTH_TOKEN reaches isolatedEnv', () => {
+    // User intentionally configured a gateway with a gateway-appropriate
+    // auth token. Both must be re-injected into isolatedEnv.
+    fs.writeFileSync(
+      TEST_ENV_FILE,
+      'ANTHROPIC_BASE_URL=https://gateway.example\nANTHROPIC_AUTH_TOKEN=test-token\n',
+      { mode: 0o600 },
+    );
+
+    const result = buildIsolatedEnv();
+
+    expect(result.ANTHROPIC_BASE_URL).toBe('https://gateway.example');
+    expect(result.ANTHROPIC_AUTH_TOKEN).toBe('test-token');
+  });
+
+  it('bare .env BASE_URL alone does not trigger OAuth fetch', async () => {
+    // A user with a tokenless gateway (e.g. mTLS at the network boundary)
+    // configures BASE_URL only. The three-branch predicate must hit the
+    // BASE_URL-set branch BEFORE OAuth lookup, so CLAUDE_CODE_OAUTH_TOKEN
+    // must NOT appear in the result. This is the security-regression guard
+    // against a token leak to a third-party gateway.
+    //
+    // Note: EnvManager captures readClaudeOAuthToken via a named import at
+    // module load, so spyOn on the namespace export only weakly observes
+    // the call (the binding inside EnvManager is independent). The
+    // behavioral assertions (BASE_URL re-injected AND OAuth token NOT
+    // injected) are the load-bearing checks: in the no-OAuth-injection
+    // outcome, the only execution path that produces this combination is
+    // the new BASE_URL-first branch returning early.
+    fs.writeFileSync(
+      TEST_ENV_FILE,
+      'ANTHROPIC_BASE_URL=https://gateway.example\n',
+      { mode: 0o600 },
+    );
+
+    const oauthSpy = spyOn(oauthToken, 'readClaudeOAuthToken');
+
+    try {
+      const result = await buildIsolatedEnvWithFreshOAuth();
+
+      expect(result.ANTHROPIC_BASE_URL).toBe('https://gateway.example');
+      expect(result.CLAUDE_CODE_OAUTH_TOKEN).toBeUndefined();
+      // Best-effort sanity check; see note above.
+      expect(oauthSpy).not.toHaveBeenCalled();
+    } finally {
+      oauthSpy.mockRestore();
+    }
+  });
+});
@@ -0,0 +1,228 @@
+import { describe, it, expect, beforeEach, mock } from 'bun:test';
+
+// Singleton enforcement regression coverage for issue #2313.
+//
+// Hypothesis under test: prior to the fix, ChromaMcpManager could leak its
+// chroma-mcp subprocess tree on every reconnect / transport error, accumulating
+// 20+ instances per session on Linux because the MCP SDK's transport.close()
+// only signals the direct child (uvx). The fix routes every "abandon current
+// transport" path through disposeCurrentSubprocess(), which tree-kills via
+// killProcessTree() before nulling the handles.
+
+let transportCount = 0;
+const transportInstances: Array<FakeTransport> = [];
+
+interface FakeChildProcess {
+  pid: number;
+  once: (event: string, _cb: (...args: unknown[]) => void) => FakeChildProcess;
+  on: (event: string, _cb: (...args: unknown[]) => void) => FakeChildProcess;
+}
+
+class FakeTransport {
+  static nextPid = 100_000;
+  onclose: (() => void) | null = null;
+  closed = false;
+  // Mimic StdioClientTransport's internal `_process` field that the manager
+  // pokes into via `(this.transport as unknown as { _process })._process`.
+  _process: FakeChildProcess;
+
+  constructor(_opts: { command: string; args: string[] }) {
+    transportCount += 1;
+    const pid = FakeTransport.nextPid++;
+    const child: FakeChildProcess = {
+      pid,
+      once: function (this: FakeChildProcess) { return this; },
+      on: function (this: FakeChildProcess) { return this; },
+    };
+    this._process = child;
+    transportInstances.push(this);
+  }
+
+  async close(): Promise<void> {
+    this.closed = true;
+  }
+}
+
+mock.module('@modelcontextprotocol/sdk/client/stdio.js', () => ({
+  StdioClientTransport: FakeTransport,
+}));
+
+let connectImpl: () => Promise<void> = async () => {};
+let callToolImpl: () => Promise<unknown> = async () => ({
+  content: [{ type: 'text', text: '{}' }],
+});
+
+class FakeClient {
+  closed = false;
+  async connect(): Promise<void> {
+    await connectImpl();
+  }
+  async callTool(): Promise<unknown> {
+    return await callToolImpl();
+  }
+  async close(): Promise<void> {
+    this.closed = true;
+  }
+}
+
+mock.module('@modelcontextprotocol/sdk/client/index.js', () => ({
+  Client: FakeClient,
+}));
+
+mock.module('../../../src/shared/SettingsDefaultsManager.js', () => ({
+  SettingsDefaultsManager: {
+    get: () => '',
+    getInt: () => 0,
+    loadFromFile: () => ({}),
+  },
+}));
+
+mock.module('../../../src/shared/paths.js', () => ({
+  USER_SETTINGS_PATH: '/tmp/fake-settings.json',
+  paths: {
+    chroma: () => '/tmp/fake-chroma',
+    combinedCerts: () => '/tmp/fake-combined-certs.pem',
+  },
+}));
+
+mock.module('../../../src/utils/logger.js', () => ({
+  logger: {
+    info: () => {},
+    debug: () => {},
+    warn: () => {},
+    error: () => {},
+    failure: () => {},
+  },
+}));
+
+// Track tree-kill invocations and the transport whose subprocess was killed.
+const killTreeCalls: number[] = [];
+
+mock.module('../../../src/supervisor/index.ts', () => ({
+  getSupervisor: () => ({
+    assertCanSpawn: () => {},
+    registerProcess: () => {},
+    unregisterProcess: () => {},
+  }),
+}));
+
+mock.module('../../../src/supervisor/env-sanitizer.js', () => ({
+  sanitizeEnv: (env: NodeJS.ProcessEnv) => env,
+}));
+
+// Replace child_process.execFile so the static killProcessTree implementation
+// can be observed without actually shelling out. We feed pgrep an empty stdout
+// (no descendants) so the only signal target is the root pid.
+mock.module('child_process', () => {
+  const original = require('node:child_process');
+  return {
+    ...original,
+    execFile: (
+      cmd: string,
+      args: string[],
+      _opts: unknown,
+      cb: (err: Error | null, stdout: { stdout: string; stderr: string }) => void
+    ) => {
+      // Bun's promisify path will call this as if it were a Node-style callback.
+      if (cmd === 'pgrep') {
+        cb(null, { stdout: '', stderr: '' } as any);
+      } else {
+        cb(null, { stdout: '', stderr: '' } as any);
+      }
+    },
+    execSync: () => '',
+  };
+});
+
+// Stub process.kill so the tree-kill path can record targets without crashing
+// the test runner if the synthetic PID happens to collide with a real one.
+const realProcessKill = process.kill.bind(process);
+const stubbedProcessKill = ((pid: number, _signal?: string | number) => {
+  killTreeCalls.push(pid);
+  return true;
+}) as typeof process.kill;
+process.kill = stubbedProcessKill;
+
+import { ChromaMcpManager } from '../../../src/services/sync/ChromaMcpManager.js';
+
+function resetState(): void {
+  transportCount = 0;
+  transportInstances.length = 0;
+  killTreeCalls.length = 0;
+  connectImpl = async () => {};
+  callToolImpl = async () => ({ content: [{ type: 'text', text: '{}' }] });
+}
+
+describe('ChromaMcpManager singleton enforcement (#2313)', () => {
+  beforeEach(async () => {
+    await ChromaMcpManager.reset();
+    resetState();
+  });
+
+  it('serializes concurrent ensureConnected() calls into one spawn', async () => {
+    const mgr = ChromaMcpManager.getInstance();
+
+    // Five parallel callers race ensureConnected via callTool — only one
+    // chroma-mcp subprocess (one transport) should be spawned.
+    await Promise.all(
+      Array.from({ length: 5 }, () =>
+        mgr.callTool('chroma_list_collections', { limit: 1 })
+      )
+    );
+
+    expect(transportCount).toBe(1);
+  });
+
+  it('kills the prior subprocess tree before a reconnect spawn', async () => {
+    const mgr = ChromaMcpManager.getInstance();
+
+    // First call: opens transport #1.
+    await mgr.callTool('chroma_list_collections', { limit: 1 });
+    expect(transportInstances.length).toBe(1);
+    const firstPid = transportInstances[0]._process.pid;
+
+    // Second call: rig callTool to throw a transport error on the FIRST attempt
+    // so the manager runs its reconnect-and-retry path. The retry should
+    // dispose the prior subprocess tree (firstPid) before spawning a new one.
+    let invocations = 0;
+    callToolImpl = async () => {
+      invocations += 1;
+      if (invocations === 1) {
+        throw new Error('Connection closed');
+      }
+      return { content: [{ type: 'text', text: '{}' }] };
+    };
+
+    await mgr.callTool('chroma_list_collections', { limit: 1 });
+
+    expect(transportInstances.length).toBe(2);
+    // The first transport's pid must have been signaled by killProcessTree
+    // before the second transport spawned.
+    expect(killTreeCalls).toContain(firstPid);
+  });
+
+  it('stop() disposes state including any pending connecting promise', async () => {
+    const mgr = ChromaMcpManager.getInstance();
+
+    await mgr.callTool('chroma_list_collections', { limit: 1 });
+    expect(transportInstances.length).toBe(1);
+    const subprocessPid = transportInstances[0]._process.pid;
+
+    await mgr.stop();
+
+    // After stop(), every internal handle should be cleared and the prior
+    // subprocess tree must have been signaled.
+    expect(killTreeCalls).toContain(subprocessPid);
+
+    // A subsequent ensureConnected must spawn a fresh transport (not reuse
+    // a stale one).
+    await mgr.callTool('chroma_list_collections', { limit: 1 });
+    expect(transportInstances.length).toBe(2);
+  });
+});
+
+// Restore the real process.kill once the test module finishes evaluating any
+// late-arriving microtasks.
+process.on('exit', () => {
+  process.kill = realProcessKill;
+});