@agentmemory/agentmemory 0.8.9 → 0.8.10

package/README.md

@@ -7,6 +7,14 @@
   Persistent memory for Claude Code, Cursor, Gemini CLI, OpenCode, and any MCP client.
 </p>
 
+<p align="center">
+  <a href="https://gist.github.com/rohitg00/2067ab416f7bbe447c1977edaaa681e2"><img src="https://img.shields.io/badge/Viral%20GitHub%20Gist-669%20stars%20%2F%2091%20forks-FF6B35?style=for-the-badge&logo=github&logoColor=white&labelColor=1a1a1a" alt="Design doc: 686 stars / 94 forks on the gist" /></a>
+</p>
+
+<p align="center">
+  <strong>The gist extends Karpathy's LLM Wiki pattern with confidence scoring, lifecycle, knowledge graphs, and hybrid search.<br/> agentmemory is the implementation.</strong>
+</p>
+
 <p align="center">
   <a href="https://www.npmjs.com/package/@agentmemory/agentmemory"><img src="https://img.shields.io/npm/v/@agentmemory/agentmemory?color=CB3837&label=npm&style=for-the-badge&logo=npm" alt="npm version" /></a>
   <a href="https://github.com/rohitg00/agentmemory/actions"><img src="https://img.shields.io/github/actions/workflow/status/rohitg00/agentmemory/ci.yml?label=tests&style=for-the-badge&logo=github" alt="CI" /></a>
@@ -723,6 +731,20 @@ Create `~/.agentmemory/.env`:
                                    # LLM provider to compress the
                                    # observation — expect significant
                                    # token spend on active sessions.
+# AGENTMEMORY_INJECT_CONTEXT=false # OFF by default (#143). When on:
+                                   # - SessionStart may inject ~1-2K
+                                   #   chars of project context into
+                                   #   the first turn of each session
+                                   #   (this is what actually reaches
+                                   #   the model — Claude Code treats
+                                   #   SessionStart stdout as context)
+                                   # - PreToolUse fires /agentmemory/enrich
+                                   #   on every file-touching tool call
+                                   #   (resource cleanup, not a token
+                                   #   fix — PreToolUse stdout is debug
+                                   #   log only per Claude Code docs)
+                                   # Observations are still captured via
+                                   # PostToolUse regardless of this flag.
 # GRAPH_EXTRACTION_ENABLED=false
 # CONSOLIDATION_ENABLED=true
 # LESSON_DECAY_ENABLED=true

package/dist/cli.mjs

@@ -284,12 +284,12 @@ async function main() {
 	p.intro("agentmemory");
 	if (skipEngine) {
 		p.log.info("Skipping engine check (--no-engine)");
-		await import("./src-DJpwR1mt.mjs");
+		await import("./src-65nK6f5B.mjs");
 		return;
 	}
 	if (await isEngineRunning()) {
 		p.log.success("iii-engine is running");
-		await import("./src-DJpwR1mt.mjs");
+		await import("./src-65nK6f5B.mjs");
 		return;
 	}
 	if (!await startEngine()) {
@@ -333,7 +333,7 @@ async function main() {
 		process.exit(1);
 	}
 	s.stop("iii-engine is ready");
-	await import("./src-DJpwR1mt.mjs");
+	await import("./src-65nK6f5B.mjs");
 }
 async function runStatus() {
 	const port = getRestPort();
@@ -557,7 +557,7 @@ async function runDemo() {
 	p.log.success("agentmemory is working. Point your agent at it and get back to coding.");
 }
 async function runMcp() {
-	await import("./standalone-DpxhaZLn.mjs");
+	await import("./standalone-CdWMLSak.mjs");
 }
 ({
 	status: runStatus,

package/dist/hooks/pre-tool-use.mjs

@@ -1,5 +1,6 @@
 #!/usr/bin/env node
 //#region src/hooks/pre-tool-use.ts
+const INJECT_CONTEXT = process.env["AGENTMEMORY_INJECT_CONTEXT"] === "true";
 const REST_URL = process.env["AGENTMEMORY_URL"] || "http://localhost:3111";
 const SECRET = process.env["AGENTMEMORY_SECRET"] || "";
 function authHeaders() {
@@ -8,6 +9,7 @@ function authHeaders() {
 	return h;
 }
 async function main() {
+	if (!INJECT_CONTEXT) return;
 	let input = "";
 	for await (const chunk of process.stdin) input += chunk;
 	let data;

package/dist/hooks/pre-tool-use.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"pre-tool-use.mjs","names":[],"sources":["../../src/hooks/pre-tool-use.ts"],"sourcesContent":["#!/usr/bin/env node\n\nconst REST_URL = process.env[\"AGENTMEMORY_URL\"] || \"http://localhost:3111\";\nconst SECRET = process.env[\"AGENTMEMORY_SECRET\"] || \"\";\n\nfunction authHeaders(): Record<string, string> {\n  const h: Record<string, string> = { \"Content-Type\": \"application/json\" };\n  if (SECRET) h[\"Authorization\"] = `Bearer ${SECRET}`;\n  return h;\n}\n\nasync function main() {\n  let input = \"\";\n  for await (const chunk of process.stdin) {\n    input += chunk;\n  }\n\n  let data: Record<string, unknown>;\n  try {\n    data = JSON.parse(input);\n  } catch {\n    return;\n  }\n\n  const toolName = data.tool_name as string;\n  if (!toolName) return;\n\n  const fileTools = [\"Edit\", \"Write\", \"Read\", \"Glob\", \"Grep\"];\n  if (!fileTools.includes(toolName)) return;\n\n  const toolInput = (data.tool_input || {}) as Record<string, unknown>;\n  const files: string[] = [];\n  const fileKeys =\n    toolName === \"Grep\"\n      ? [\"path\", \"file\"]\n      : [\"file_path\", \"path\", \"file\", \"pattern\"];\n  for (const key of fileKeys) {\n    const val = toolInput[key];\n    if (typeof val === \"string\" && val.length > 0) files.push(val);\n  }\n  if (files.length === 0) return;\n\n  const terms: string[] = [];\n  if (toolName === \"Grep\" || toolName === \"Glob\") {\n    const pattern = toolInput[\"pattern\"];\n    if (typeof pattern === \"string\" && pattern.length > 0) {\n      terms.push(pattern);\n    }\n  }\n\n  const sessionId = (data.session_id as string) || \"unknown\";\n\n  try {\n    const res = await fetch(`${REST_URL}/agentmemory/enrich`, {\n      method: \"POST\",\n      headers: authHeaders(),\n      body: JSON.stringify({ sessionId, files, terms, toolName }),\n      signal: AbortSignal.timeout(2000),\n    });\n\n    if (res.ok) {\n      const result = (await res.json()) as { context?: string };\n      if (result.context) {\n        process.stdout.write(result.context);\n      }\n    }\n  } catch {\n    // don't block tool execution\n  }\n}\n\nmain();\n"],"mappings":";;AAEA,MAAM,WAAW,QAAQ,IAAI,sBAAsB;AACnD,MAAM,SAAS,QAAQ,IAAI,yBAAyB;AAEpD,SAAS,cAAsC;CAC7C,MAAM,IAA4B,EAAE,gBAAgB,oBAAoB;AACxE,KAAI,OAAQ,GAAE,mBAAmB,UAAU;AAC3C,QAAO;;AAGT,eAAe,OAAO;CACpB,IAAI,QAAQ;AACZ,YAAW,MAAM,SAAS,QAAQ,MAChC,UAAS;CAGX,IAAI;AACJ,KAAI;AACF,SAAO,KAAK,MAAM,MAAM;SAClB;AACN;;CAGF,MAAM,WAAW,KAAK;AACtB,KAAI,CAAC,SAAU;AAGf,KAAI,CADc;EAAC;EAAQ;EAAS;EAAQ;EAAQ;EAAO,CAC5C,SAAS,SAAS,CAAE;CAEnC,MAAM,YAAa,KAAK,cAAc,EAAE;CACxC,MAAM,QAAkB,EAAE;CAC1B,MAAM,WACJ,aAAa,SACT,CAAC,QAAQ,OAAO,GAChB;EAAC;EAAa;EAAQ;EAAQ;EAAU;AAC9C,MAAK,MAAM,OAAO,UAAU;EAC1B,MAAM,MAAM,UAAU;AACtB,MAAI,OAAO,QAAQ,YAAY,IAAI,SAAS,EAAG,OAAM,KAAK,IAAI;;AAEhE,KAAI,MAAM,WAAW,EAAG;CAExB,MAAM,QAAkB,EAAE;AAC1B,KAAI,aAAa,UAAU,aAAa,QAAQ;EAC9C,MAAM,UAAU,UAAU;AAC1B,MAAI,OAAO,YAAY,YAAY,QAAQ,SAAS,EAClD,OAAM,KAAK,QAAQ;;CAIvB,MAAM,YAAa,KAAK,cAAyB;AAEjD,KAAI;EACF,MAAM,MAAM,MAAM,MAAM,GAAG,SAAS,sBAAsB;GACxD,QAAQ;GACR,SAAS,aAAa;GACtB,MAAM,KAAK,UAAU;IAAE;IAAW;IAAO;IAAO;IAAU,CAAC;GAC3D,QAAQ,YAAY,QAAQ,IAAK;GAClC,CAAC;AAEF,MAAI,IAAI,IAAI;GACV,MAAM,SAAU,MAAM,IAAI,MAAM;AAChC,OAAI,OAAO,QACT,SAAQ,OAAO,MAAM,OAAO,QAAQ;;SAGlC;;AAKV,MAAM"}
+{"version":3,"file":"pre-tool-use.mjs","names":[],"sources":["../../src/hooks/pre-tool-use.ts"],"sourcesContent":["#!/usr/bin/env node\n\n// Pre-tool-use enrichment hook.\n//\n// THIS HOOK IS A NO-OP BY DEFAULT AS OF 0.8.10 (#143). Previously it\n// fired /agentmemory/enrich on every Edit/Write/Read/Glob/Grep tool call\n// and wrote up to 4000 chars of context to stdout. Claude Code reads\n// PreToolUse stdout and prepends it to the model's next turn, which meant\n// agentmemory was silently injecting ~1000 tokens into every tool turn\n// via the user's Claude Code session. On Claude Pro that burned entire\n// allocations in a handful of messages (@adrianricardo, #143).\n//\n// Users who explicitly want pre-tool enrichment opt in with:\n//   AGENTMEMORY_INJECT_CONTEXT=true   in ~/.agentmemory/.env\n// and restart Claude Code. Expect your session input token count to grow\n// proportionally with the number of file-touching tool calls per turn.\nconst INJECT_CONTEXT = process.env[\"AGENTMEMORY_INJECT_CONTEXT\"] === \"true\";\n\nconst REST_URL = process.env[\"AGENTMEMORY_URL\"] || \"http://localhost:3111\";\nconst SECRET = process.env[\"AGENTMEMORY_SECRET\"] || \"\";\n\nfunction authHeaders(): Record<string, string> {\n  const h: Record<string, string> = { \"Content-Type\": \"application/json\" };\n  if (SECRET) h[\"Authorization\"] = `Bearer ${SECRET}`;\n  return h;\n}\n\nasync function main() {\n  // Default off: exit immediately so we don't even open stdin. This keeps\n  // Claude Code's tool-call hot path as cheap as possible.\n  if (!INJECT_CONTEXT) return;\n\n  let input = \"\";\n  for await (const chunk of process.stdin) {\n    input += chunk;\n  }\n\n  let data: Record<string, unknown>;\n  try {\n    data = JSON.parse(input);\n  } catch {\n    return;\n  }\n\n  const toolName = data.tool_name as string;\n  if (!toolName) return;\n\n  const fileTools = [\"Edit\", \"Write\", \"Read\", \"Glob\", \"Grep\"];\n  if (!fileTools.includes(toolName)) return;\n\n  const toolInput = (data.tool_input || {}) as Record<string, unknown>;\n  const files: string[] = [];\n  const fileKeys =\n    toolName === \"Grep\"\n      ? [\"path\", \"file\"]\n      : [\"file_path\", \"path\", \"file\", \"pattern\"];\n  for (const key of fileKeys) {\n    const val = toolInput[key];\n    if (typeof val === \"string\" && val.length > 0) files.push(val);\n  }\n  if (files.length === 0) return;\n\n  const terms: string[] = [];\n  if (toolName === \"Grep\" || toolName === \"Glob\") {\n    const pattern = toolInput[\"pattern\"];\n    if (typeof pattern === \"string\" && pattern.length > 0) {\n      terms.push(pattern);\n    }\n  }\n\n  const sessionId = (data.session_id as string) || \"unknown\";\n\n  try {\n    const res = await fetch(`${REST_URL}/agentmemory/enrich`, {\n      method: \"POST\",\n      headers: authHeaders(),\n      body: JSON.stringify({ sessionId, files, terms, toolName }),\n      signal: AbortSignal.timeout(2000),\n    });\n\n    if (res.ok) {\n      const result = (await res.json()) as { context?: string };\n      if (result.context) {\n        process.stdout.write(result.context);\n      }\n    }\n  } catch {\n    // don't block tool execution\n  }\n}\n\nmain();\n"],"mappings":";;AAgBA,MAAM,iBAAiB,QAAQ,IAAI,kCAAkC;AAErE,MAAM,WAAW,QAAQ,IAAI,sBAAsB;AACnD,MAAM,SAAS,QAAQ,IAAI,yBAAyB;AAEpD,SAAS,cAAsC;CAC7C,MAAM,IAA4B,EAAE,gBAAgB,oBAAoB;AACxE,KAAI,OAAQ,GAAE,mBAAmB,UAAU;AAC3C,QAAO;;AAGT,eAAe,OAAO;AAGpB,KAAI,CAAC,eAAgB;CAErB,IAAI,QAAQ;AACZ,YAAW,MAAM,SAAS,QAAQ,MAChC,UAAS;CAGX,IAAI;AACJ,KAAI;AACF,SAAO,KAAK,MAAM,MAAM;SAClB;AACN;;CAGF,MAAM,WAAW,KAAK;AACtB,KAAI,CAAC,SAAU;AAGf,KAAI,CADc;EAAC;EAAQ;EAAS;EAAQ;EAAQ;EAAO,CAC5C,SAAS,SAAS,CAAE;CAEnC,MAAM,YAAa,KAAK,cAAc,EAAE;CACxC,MAAM,QAAkB,EAAE;CAC1B,MAAM,WACJ,aAAa,SACT,CAAC,QAAQ,OAAO,GAChB;EAAC;EAAa;EAAQ;EAAQ;EAAU;AAC9C,MAAK,MAAM,OAAO,UAAU;EAC1B,MAAM,MAAM,UAAU;AACtB,MAAI,OAAO,QAAQ,YAAY,IAAI,SAAS,EAAG,OAAM,KAAK,IAAI;;AAEhE,KAAI,MAAM,WAAW,EAAG;CAExB,MAAM,QAAkB,EAAE;AAC1B,KAAI,aAAa,UAAU,aAAa,QAAQ;EAC9C,MAAM,UAAU,UAAU;AAC1B,MAAI,OAAO,YAAY,YAAY,QAAQ,SAAS,EAClD,OAAM,KAAK,QAAQ;;CAIvB,MAAM,YAAa,KAAK,cAAyB;AAEjD,KAAI;EACF,MAAM,MAAM,MAAM,MAAM,GAAG,SAAS,sBAAsB;GACxD,QAAQ;GACR,SAAS,aAAa;GACtB,MAAM,KAAK,UAAU;IAAE;IAAW;IAAO;IAAO;IAAU,CAAC;GAC3D,QAAQ,YAAY,QAAQ,IAAK;GAClC,CAAC;AAEF,MAAI,IAAI,IAAI;GACV,MAAM,SAAU,MAAM,IAAI,MAAM;AAChC,OAAI,OAAO,QACT,SAAQ,OAAO,MAAM,OAAO,QAAQ;;SAGlC;;AAKV,MAAM"}

package/dist/hooks/session-start.mjs

@@ -1,5 +1,6 @@
 #!/usr/bin/env node
 //#region src/hooks/session-start.ts
+const INJECT_CONTEXT = process.env["AGENTMEMORY_INJECT_CONTEXT"] === "true";
 const REST_URL = process.env["AGENTMEMORY_URL"] || "http://localhost:3111";
 const SECRET = process.env["AGENTMEMORY_SECRET"] || "";
 function authHeaders() {
@@ -29,7 +30,7 @@ async function main() {
 			}),
 			signal: AbortSignal.timeout(5e3)
 		});
-		if (res.ok) {
+		if (INJECT_CONTEXT && res.ok) {
 			const result = await res.json();
 			if (result.context) process.stdout.write(result.context);
 		}

package/dist/hooks/session-start.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"session-start.mjs","names":[],"sources":["../../src/hooks/session-start.ts"],"sourcesContent":["#!/usr/bin/env node\n\nconst REST_URL = process.env[\"AGENTMEMORY_URL\"] || \"http://localhost:3111\";\nconst SECRET = process.env[\"AGENTMEMORY_SECRET\"] || \"\";\n\nfunction authHeaders(): Record<string, string> {\n  const h: Record<string, string> = { \"Content-Type\": \"application/json\" };\n  if (SECRET) h[\"Authorization\"] = `Bearer ${SECRET}`;\n  return h;\n}\n\nasync function main() {\n  let input = \"\";\n  for await (const chunk of process.stdin) {\n    input += chunk;\n  }\n\n  let data: Record<string, unknown>;\n  try {\n    data = JSON.parse(input);\n  } catch {\n    return;\n  }\n\n  const sessionId =\n    (data.session_id as string) || `ses_${Date.now().toString(36)}`;\n  const project = (data.cwd as string) || process.cwd();\n\n  try {\n    const res = await fetch(`${REST_URL}/agentmemory/session/start`, {\n      method: \"POST\",\n      headers: authHeaders(),\n      body: JSON.stringify({ sessionId, project, cwd: project }),\n      signal: AbortSignal.timeout(5000),\n    });\n\n    if (res.ok) {\n      const result = (await res.json()) as { context?: string };\n      if (result.context) {\n        process.stdout.write(result.context);\n      }\n    }\n  } catch {\n    // silently fail -- don't block Claude Code startup\n  }\n}\n\nmain();\n"],"mappings":";;AAEA,MAAM,WAAW,QAAQ,IAAI,sBAAsB;AACnD,MAAM,SAAS,QAAQ,IAAI,yBAAyB;AAEpD,SAAS,cAAsC;CAC7C,MAAM,IAA4B,EAAE,gBAAgB,oBAAoB;AACxE,KAAI,OAAQ,GAAE,mBAAmB,UAAU;AAC3C,QAAO;;AAGT,eAAe,OAAO;CACpB,IAAI,QAAQ;AACZ,YAAW,MAAM,SAAS,QAAQ,MAChC,UAAS;CAGX,IAAI;AACJ,KAAI;AACF,SAAO,KAAK,MAAM,MAAM;SAClB;AACN;;CAGF,MAAM,YACH,KAAK,cAAyB,OAAO,KAAK,KAAK,CAAC,SAAS,GAAG;CAC/D,MAAM,UAAW,KAAK,OAAkB,QAAQ,KAAK;AAErD,KAAI;EACF,MAAM,MAAM,MAAM,MAAM,GAAG,SAAS,6BAA6B;GAC/D,QAAQ;GACR,SAAS,aAAa;GACtB,MAAM,KAAK,UAAU;IAAE;IAAW;IAAS,KAAK;IAAS,CAAC;GAC1D,QAAQ,YAAY,QAAQ,IAAK;GAClC,CAAC;AAEF,MAAI,IAAI,IAAI;GACV,MAAM,SAAU,MAAM,IAAI,MAAM;AAChC,OAAI,OAAO,QACT,SAAQ,OAAO,MAAM,OAAO,QAAQ;;SAGlC;;AAKV,MAAM"}
+{"version":3,"file":"session-start.mjs","names":[],"sources":["../../src/hooks/session-start.ts"],"sourcesContent":["#!/usr/bin/env node\n\n// Session-start hook.\n//\n// Always registers the session for observation tracking (so memories\n// captured on PostToolUse get attached to the right session). Only writes\n// project context to stdout — which Claude Code prepends to the very first\n// turn — when AGENTMEMORY_INJECT_CONTEXT=true. Default off as of 0.8.10\n// (#143); see pre-tool-use.ts for the full explanation.\nconst INJECT_CONTEXT = process.env[\"AGENTMEMORY_INJECT_CONTEXT\"] === \"true\";\n\nconst REST_URL = process.env[\"AGENTMEMORY_URL\"] || \"http://localhost:3111\";\nconst SECRET = process.env[\"AGENTMEMORY_SECRET\"] || \"\";\n\nfunction authHeaders(): Record<string, string> {\n  const h: Record<string, string> = { \"Content-Type\": \"application/json\" };\n  if (SECRET) h[\"Authorization\"] = `Bearer ${SECRET}`;\n  return h;\n}\n\nasync function main() {\n  let input = \"\";\n  for await (const chunk of process.stdin) {\n    input += chunk;\n  }\n\n  let data: Record<string, unknown>;\n  try {\n    data = JSON.parse(input);\n  } catch {\n    return;\n  }\n\n  const sessionId =\n    (data.session_id as string) || `ses_${Date.now().toString(36)}`;\n  const project = (data.cwd as string) || process.cwd();\n\n  try {\n    const res = await fetch(`${REST_URL}/agentmemory/session/start`, {\n      method: \"POST\",\n      headers: authHeaders(),\n      body: JSON.stringify({ sessionId, project, cwd: project }),\n      signal: AbortSignal.timeout(5000),\n    });\n\n    // Only write context to stdout when the user has explicitly opted\n    // into injection. Registering the session is cheap and doesn't touch\n    // Claude Code's input token window.\n    if (INJECT_CONTEXT && res.ok) {\n      const result = (await res.json()) as { context?: string };\n      if (result.context) {\n        process.stdout.write(result.context);\n      }\n    }\n  } catch {\n    // silently fail -- don't block Claude Code startup\n  }\n}\n\nmain();\n"],"mappings":";;AASA,MAAM,iBAAiB,QAAQ,IAAI,kCAAkC;AAErE,MAAM,WAAW,QAAQ,IAAI,sBAAsB;AACnD,MAAM,SAAS,QAAQ,IAAI,yBAAyB;AAEpD,SAAS,cAAsC;CAC7C,MAAM,IAA4B,EAAE,gBAAgB,oBAAoB;AACxE,KAAI,OAAQ,GAAE,mBAAmB,UAAU;AAC3C,QAAO;;AAGT,eAAe,OAAO;CACpB,IAAI,QAAQ;AACZ,YAAW,MAAM,SAAS,QAAQ,MAChC,UAAS;CAGX,IAAI;AACJ,KAAI;AACF,SAAO,KAAK,MAAM,MAAM;SAClB;AACN;;CAGF,MAAM,YACH,KAAK,cAAyB,OAAO,KAAK,KAAK,CAAC,SAAS,GAAG;CAC/D,MAAM,UAAW,KAAK,OAAkB,QAAQ,KAAK;AAErD,KAAI;EACF,MAAM,MAAM,MAAM,MAAM,GAAG,SAAS,6BAA6B;GAC/D,QAAQ;GACR,SAAS,aAAa;GACtB,MAAM,KAAK,UAAU;IAAE;IAAW;IAAS,KAAK;IAAS,CAAC;GAC1D,QAAQ,YAAY,QAAQ,IAAK;GAClC,CAAC;AAKF,MAAI,kBAAkB,IAAI,IAAI;GAC5B,MAAM,SAAU,MAAM,IAAI,MAAM;AAChC,OAAI,OAAO,QACT,SAAQ,OAAO,MAAM,OAAO,QAAQ;;SAGlC;;AAKV,MAAM"}

package/dist/index.mjs

@@ -159,6 +159,9 @@ function isConsolidationEnabled() {
 function isAutoCompressEnabled() {
 	return getMergedEnv()["AGENTMEMORY_AUTO_COMPRESS"] === "true";
 }
+function isContextInjectionEnabled() {
+	return getMergedEnv()["AGENTMEMORY_INJECT_CONTEXT"] === "true";
+}
 function getConsolidationDecayDays() {
 	return safeParseInt(getMergedEnv()["CONSOLIDATION_DECAY_DAYS"], 30);
 }
@@ -4127,7 +4130,7 @@ function registerAutoForgetFunction(sdk, kv) {
 
 //#endregion
 //#region src/version.ts
-const VERSION = "0.8.9";
+const VERSION = "0.8.10";
 
 //#endregion
 //#region src/functions/export-import.ts
@@ -4244,7 +4247,8 @@ function registerExportImportFunction(sdk, kv) {
 			"0.8.6",
 			"0.8.7",
 			"0.8.8",
-			"0.8.9"
+			"0.8.9",
+			"0.8.10"
 		]).has(importData.version)) return {
 			success: false,
 			error: `Unsupported export version: ${importData.version}`
@@ -10109,6 +10113,9 @@ function registerRetentionFunctions(sdk, kv) {
 		}
 		const scores = [];
 		const computeDecay = (createdAt) => Math.exp(-config.lambda * ((Date.now() - new Date(createdAt).getTime()) / (1e3 * 60 * 60 * 24)));
+		const pendingWrites = [];
+		let episodicScored = 0;
+		let semanticScored = 0;
 		for (const mem of memories) {
 			if (!mem.isLatest) continue;
 			const log = logsById.get(mem.id) ?? emptyAccessLog(mem.id);
@@ -10118,6 +10125,7 @@ function registerRetentionFunctions(sdk, kv) {
 			const score = Math.min(1, salience * temporalDecay + reinforcementBoost);
 			const entry = {
 				memoryId: mem.id,
+				source: "episodic",
 				score,
 				salience,
 				temporalDecay,
@@ -10126,7 +10134,8 @@ function registerRetentionFunctions(sdk, kv) {
 				accessCount: log.count
 			};
 			scores.push(entry);
-			await kv.set(KV.retentionScores, mem.id, entry);
+			pendingWrites.push([mem.id, entry]);
+			episodicScored++;
 		}
 		for (const sem of semanticMems) {
 			const log = logsById.get(sem.id) ?? emptyAccessLog(sem.id);
@@ -10149,6 +10158,7 @@ function registerRetentionFunctions(sdk, kv) {
 			const score = Math.min(1, salience * temporalDecay + reinforcementBoost);
 			const entry = {
 				memoryId: sem.id,
+				source: "semantic",
 				score,
 				salience,
 				temporalDecay,
@@ -10157,8 +10167,10 @@ function registerRetentionFunctions(sdk, kv) {
 				accessCount: effectiveCount
 			};
 			scores.push(entry);
-			await kv.set(KV.retentionScores, sem.id, entry);
+			pendingWrites.push([sem.id, entry]);
+			semanticScored++;
 		}
+		await Promise.all(pendingWrites.map(([id, entry]) => kv.set(KV.retentionScores, id, entry)));
 		scores.sort((a, b) => b.score - a.score);
 		const tiers = {
 			hot: scores.filter((s) => s.score >= config.tierThresholds.hot).length,
@@ -10170,6 +10182,13 @@ function registerRetentionFunctions(sdk, kv) {
 			total: scores.length,
 			...tiers
 		});
+		if (scores.length > 0) await recordAudit(kv, "retention_score", "mem::retention-score", [], {
+			total: scores.length,
+			episodic: episodicScored,
+			semantic: semanticScored,
+			tiers,
+			config
+		});
 		return {
 			success: true,
 			total: scores.length,
@@ -10195,21 +10214,53 @@ function registerRetentionFunctions(sdk, kv) {
 			}))
 		};
 		let evicted = 0;
+		let evictedEpisodic = 0;
+		let evictedSemantic = 0;
+		const evictedIds = [];
 		for (const candidate of candidates) try {
-			await kv.delete(KV.memories, candidate.memoryId);
+			let scope;
+			let resolvedSource;
+			if (candidate.source === "semantic") {
+				scope = KV.semantic;
+				resolvedSource = "semantic";
+			} else if (candidate.source === "episodic") {
+				scope = KV.memories;
+				resolvedSource = "episodic";
+			} else if (await kv.get(KV.memories, candidate.memoryId) !== null) {
+				scope = KV.memories;
+				resolvedSource = "episodic";
+			} else {
+				scope = KV.semantic;
+				resolvedSource = "semantic";
+			}
+			await kv.delete(scope, candidate.memoryId);
 			await kv.delete(KV.retentionScores, candidate.memoryId);
 			await deleteAccessLog(kv, candidate.memoryId);
 			evicted++;
+			evictedIds.push(candidate.memoryId);
+			if (resolvedSource === "semantic") evictedSemantic++;
+			else evictedEpisodic++;
 		} catch {
 			continue;
 		}
+		if (evicted > 0) await recordAudit(kv, "delete", "mem::retention-evict", evictedIds, {
+			threshold,
+			evicted,
+			evictedEpisodic,
+			evictedSemantic,
+			reason: "retention score below threshold"
+		});
 		ctx.logger.info("Retention-based eviction complete", {
 			evicted,
+			evictedEpisodic,
+			evictedSemantic,
 			threshold
 		});
 		return {
 			success: true,
-			evicted
+			evicted,
+			evictedEpisodic,
+			evictedSemantic
 		};
 	});
 }
@@ -15082,6 +15133,8 @@ async function main() {
 	console.log(`[agentmemory] Consolidation pipeline: registered (CONSOLIDATION_ENABLED=${isConsolidationEnabled() ? "true" : "false"})`);
 	if (isAutoCompressEnabled()) console.log(`[agentmemory] WARNING: AGENTMEMORY_AUTO_COMPRESS=true — every PostToolUse observation will be sent to your LLM provider for compression. This spends API tokens proportional to your session tool-use frequency (see #138). Set AGENTMEMORY_AUTO_COMPRESS=false to disable.`);
 	else console.log(`[agentmemory] Auto-compress: OFF (default, #138) — observations indexed via zero-LLM synthetic compression. Set AGENTMEMORY_AUTO_COMPRESS=true to opt-in to LLM-powered summaries (uses your API key).`);
+	if (isContextInjectionEnabled()) console.log(`[agentmemory] WARNING: AGENTMEMORY_INJECT_CONTEXT=true — the PreToolUse and SessionStart hooks will inject up to ~4000 chars of memory context into every tool turn. On Claude Pro this burns session tokens proportional to your tool-call frequency (see #143). Set AGENTMEMORY_INJECT_CONTEXT=false to disable.`);
+	else console.log(`[agentmemory] Context injection: OFF (default, #143) — hooks capture observations but do not inject context into Claude Code's conversation. Set AGENTMEMORY_INJECT_CONTEXT=true to opt-in (warning: expect your Claude Pro allocation to drain faster).`);
 	const teamConfig = loadTeamConfig();
 	if (teamConfig) {
 		registerTeamFunction(sdk, kv, teamConfig);