mixdog 0.9.96 → 0.9.99

package/README.md

@@ -65,23 +65,6 @@ measured cache writes; the archived OpenAI runs did not retain
 artifacts, the exact run commands, and the metric scripts that recompute
 every number above live under `benchmarks/terminal-bench-2.1/`.
 
-## Boot overhead — measured, not estimated
-
-Fixed cost of the default system prompt + tool schemas, measured with a
-minimal one-turn ping ("hi"). Every number below is what the provider's own
-usage accounting reported for the first API call — same tokenizer on both
-sides of each row, no offline estimates (2026-08).
-
-| Tokenizer | Baseline harness | First-call prompt | mixdog Lead | mixdog worker agent |
-|-----------|------------------|------------------:|------------:|--------------------:|
-| Anthropic — pristine containers, TB2.1 harness | Claude Code 2.1.220 | 19,149 tok (median, n=89) | **7,295 (−62%)** | 4,856 (−75%) |
-| OpenAI — same host, same model (`gpt-5.6-sol`) | Codex CLI 0.145.0 | 17,706 tok | **7,488 (−58%)** | 4,616 (−74%) |
-
-The Lead surface carries the full default rule set, skills manifest, and
-per-user profile; spawning a sub-agent costs a flat ~4.6–4.9k tokens. Low
-boot overhead compounds: it is re-read on every request, so it is the
-baseline of every cache write, compaction, and long-session turn.
-
 ## Why mixdog
 
 **Maximum performance at minimum cost**

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "mixdog",
-  "version": "0.9.96",
+  "version": "0.9.99",
   "private": false,
   "type": "module",
   "description": "Standalone mixdog coding-agent CLI/TUI workspace.",
@@ -71,36 +71,42 @@
     "test:agent-job-views": "node --test scripts/agent-job-terminal-view-test.mjs",
     "test:agent-fanout": "node scripts/agent-parallel-smoke.mjs && node --test scripts/agent-route-batch-test.mjs scripts/execution-completion-dedup-test.mjs",
     "test:toolcall": "node --test scripts/toolcall-args-test.mjs",
-    "test:tool-batching": "node --test scripts/eager-patch-shell-order-test.mjs scripts/session-bench-batching-test.mjs",
+    "test:tool-batching": "node --test scripts/eager-patch-shell-order-test.mjs scripts/session-bench-batching-test.mjs scripts/tool-cwd-rebind-test.mjs",
     "test:shipmode": "node --test scripts/ship-mode-test.mjs",
     "test:shellhardening": "node --test scripts/shell-hardening-test.mjs scripts/shell-failure-diagnostics-test.mjs scripts/windows-hide-spawn-options-test.mjs scripts/shell-job-session-scope-test.mjs",
     "test:placeholder": "node --test scripts/compacted-placeholder-scrub-test.mjs",
     "test:providers": "node --test scripts/provider-toolcall-test.mjs scripts/provider-contract-test.mjs scripts/provider-stream-stall-test.mjs scripts/provider-stream-outcome-test.mjs scripts/stream-frame-fault-matrix-test.mjs scripts/gemini-provider-test.mjs scripts/anthropic-transport-policy-test.mjs scripts/anthropic-native-block-replay-test.mjs scripts/openai-oauth-ws-1006-retry-test.mjs scripts/stream-close-retry-test.mjs scripts/openai-end-turn-signal-test.mjs",
     "test:provider-admission": "node --test scripts/provider-admission-scheduler-test.mjs scripts/provider-stream-json-test.mjs",
-    "test:resource-admission": "node --test scripts/resource-admission-test.mjs scripts/child-spawn-isolation-test.mjs",
+    "test:resource-admission": "node --test scripts/resource-admission-test.mjs scripts/child-spawn-isolation-test.mjs scripts/owner-fair-gate-test.mjs",
     "test:deferred-tools": "node --test scripts/deferred-tool-loading-test.mjs",
     "test:anthropic-oauth-race": "node --test scripts/anthropic-oauth-refresh-race-test.mjs",
     "test:grok-oauth-race": "node --test scripts/grok-oauth-refresh-race-test.mjs",
     "test:atomiclock": "node --test scripts/atomic-lock-tryonce-test.mjs",
     "test:keychain": "node --test scripts/keychain-prewarm-test.mjs",
+    "test:memory-leaks": "node --expose-gc --test scripts/memory-retention-test.mjs scripts/memory-worker-stability-test.mjs scripts/session-transport-test.mjs apps/desktop/src/renderer/memory-retention.test.mjs",
     "test:memory-routing": "node --test scripts/memory-cycle-routing-test.mjs scripts/maintenance-default-routes-test.mjs scripts/embedding-worker-exit-test.mjs scripts/embedding-runtime-prune-test.mjs",
     "test:embedding-runtime": "node --test scripts/embedding-runtime-prune-test.mjs scripts/memory-pg-recovery-test.mjs && node scripts/verify-embedding-runtime.mjs",
     "test:embedding-runtime:core": "node --test scripts/embedding-runtime-prune-test.mjs scripts/memory-pg-recovery-test.mjs && node scripts/verify-embedding-runtime.mjs --core",
     "test:embedding-runtime:warmup": "node scripts/verify-embedding-runtime.mjs --warmup",
+    "prepare:code-graph-test": "node scripts/prepare-code-graph-test.mjs",
     "test:code-graph-dispatch": "node --test scripts/code-graph-dispatch-test.mjs",
     "test:tui-queue": "node --test scripts/submit-commandbusy-race-test.mjs scripts/steering-drain-buckets-test.mjs scripts/abort-recovery-test.mjs scripts/message-rewind-test.mjs scripts/execution-pending-resume-kick-test.mjs scripts/execution-resume-esc-integration-test.mjs scripts/pending-stale-injection-test.mjs",
     "test:tui-input-render": "node --test scripts/prompt-immediate-render-test.mjs scripts/prompt-key-parity-test.mjs",
     "test:tui-streaming-window": "node --test scripts/streaming-tail-window-test.mjs scripts/tui-store-frame-batch-test.mjs && node scripts/tui-transcript-jitter-harness.mjs",
     "test:tui-ambiguous-width": "node --test scripts/tui-ambiguous-width-test.mjs",
     "test:release-assets": "node --check scripts/verify-release-assets.mjs && node --check scripts/verify-release-assets-test.mjs && node --check scripts/deploy-workflow-test.mjs && node --check scripts/release-version-discipline-test.mjs && node --test scripts/verify-release-assets-test.mjs scripts/deploy-workflow-test.mjs scripts/release-version-discipline-test.mjs",
-    "test:release-focused": "npm run test:release-assets && npm run test:tool-contracts && npm run test:tool-batching && npm run test:placeholder && npm run smoke:patch && npm run test:patch-binary-cache && npm run test:providers && npm run test:deferred-tools && npm run smoke:compact && npm run test:compact && npm run test:context && node --test scripts/code-graph-root-federation-test.mjs scripts/code-graph-aggregate-cwd-test.mjs && npm run test:code-graph-dispatch && node --test scripts/code-graph-disk-hit-test.mjs && npm run test:shellhardening && npm run test:project-registry && npm run test:session && npm run test:workflow-editor && npm run test:embedding-runtime && node --test scripts/tui-transcript-perf-test.mjs",
+    "test:release-focused": "npm run test:release-focused:contracts && npm run test:release-focused:providers && npm run test:release-focused:compact && npm run test:release-focused:session",
+    "test:release-focused:contracts": "npm run test:release-assets && npm run test:tool-contracts && npm run test:tool-batching && npm run test:placeholder && npm run smoke:patch && npm run test:patch-binary-cache",
+    "test:release-focused:providers": "npm run test:providers && npm run test:deferred-tools && npm run test:compact && npm run test:context && npm run prepare:code-graph-test && node --test scripts/code-graph-root-federation-test.mjs scripts/code-graph-aggregate-cwd-test.mjs && npm run test:code-graph-dispatch && node --test scripts/code-graph-disk-hit-test.mjs",
+    "test:release-focused:compact": "npm run smoke:compact && npm run test:shellhardening",
+    "test:release-focused:session": "npm run test:project-registry && npm run test:session && npm run test:workflow-editor && npm run test:embedding-runtime && node --test scripts/tui-transcript-perf-test.mjs",
     "test:native-edit-wire": "node --test scripts/native-edit-wire-test.mjs",
     "test:patch-binary-cache": "node --test scripts/patch-binary-cache-test.mjs",
     "test:patch-parity": "node --test scripts/v4a-parity-test.mjs",
     "test:project-registry": "node --test scripts/project-registry-isolation-test.mjs",
-    "test:engine-daemon": "node --test scripts/fair-call-scheduler-test.mjs scripts/engine-daemon-test.mjs scripts/engine-daemon-tui-parity-test.mjs scripts/engine-daemon-recovery-test.mjs scripts/engine-daemon-submit-delivery-test.mjs scripts/engine-daemon-process-lifetime-test.mjs scripts/engine-daemon-multipane-stress-test.mjs && node --test scripts/runtime-isolation-stress-test.mjs",
-    "smoke:engine-daemon": "node scripts/engine-daemon-smoke.mjs",
-    "test:session": "node --test scripts/session-orphan-sweep-test.mjs scripts/interrupted-turn-history-test.mjs scripts/turn-checkpoint-crash-test.mjs scripts/turn-outcome-fault-matrix-test.mjs scripts/session-save-fault-store-test.mjs scripts/session-disk-authority-test.mjs scripts/session-load-cache-race-test.mjs scripts/agent-loop-complete-turn-test.mjs scripts/session-heartbeat-lifecycle-test.mjs scripts/remote-transition-order-test.mjs scripts/session-new-reset-test.mjs scripts/turn-ttft-critical-path-test.mjs",
+    "test:session-transport": "node --test scripts/fair-call-scheduler-test.mjs scripts/session-transport-test.mjs scripts/session-tui-parity-test.mjs scripts/session-recovery-test.mjs scripts/session-submit-delivery-test.mjs scripts/daemon-process-lifetime-test.mjs scripts/session-multipane-stress-test.mjs && node --test scripts/runtime-isolation-stress-test.mjs",
+    "smoke:session": "node scripts/session-smoke.mjs",
+    "test:session": "node --test scripts/runtime-turn-contract-test.mjs scripts/session-orphan-sweep-test.mjs scripts/interrupted-turn-history-test.mjs scripts/turn-checkpoint-crash-test.mjs scripts/turn-outcome-fault-matrix-test.mjs scripts/session-save-fault-store-test.mjs scripts/session-disk-authority-test.mjs scripts/session-load-cache-race-test.mjs scripts/agent-loop-complete-turn-test.mjs scripts/session-heartbeat-lifecycle-test.mjs scripts/remote-transition-order-test.mjs scripts/session-new-reset-test.mjs scripts/turn-ttft-critical-path-test.mjs",
     "test:live-canary": "node --test scripts/live-canary-test.mjs",
     "test:rebindtail": "node --test scripts/forwarder-rebind-tail-test.mjs scripts/channel-remote-format-test.mjs",
     "test:workflow-editor": "node --test scripts/workflow-id-test.mjs scripts/workflow-pack-editor-test.mjs",
@@ -122,7 +128,7 @@
     "bench:explore": "node scripts/explore-bench.mjs",
     "bench:output-style": "node scripts/output-style-bench.mjs",
     "bench:session-context": "node scripts/session-context-bench.mjs",
-    "bench:engine-daemon": "node scripts/engine-daemon-bench.mjs",
+    "bench:session-transport": "node scripts/session-transport-bench.mjs",
     "bench:tool-overhead": "node scripts/tool-overhead-microbench.mjs",
     "bench:internal-comms": "node scripts/internal-comms-bench.mjs",
     "audit:models": "node scripts/model-catalog-audit.mjs",
@@ -154,9 +160,11 @@
     "pg": "^8.22.0",
     "puppeteer-core": "^25.2.0",
     "react": "^19.2.7",
+    "sharp": "^0.35.3",
     "string-width": "^8.2.1",
     "strip-ansi": "^7.2.0",
     "tiktoken": "^1.0.22",
+    "unpdf": "^1.8.0",
     "undici": "^8.5.0",
     "wrap-ansi": "^10.0.0",
     "ws": "^8.21.0",

package/scripts/build-tui.mjs

@@ -41,13 +41,13 @@ const sharedRuntimeExternalPlugin = {
   },
 };
 
-const engineDaemonClientExternalPlugin = {
-  name: 'mixdog-engine-daemon-client-external',
+const sessionClientExternalPlugin = {
+  name: 'mixdog-session-client-external',
   setup(build) {
-    build.onResolve({ filter: /^\.\.\/standalone\/engine-daemon-client\.mjs$/ }, () => ({
+    build.onResolve({ filter: /^\.\.\/standalone\/session-client\.mjs$/ }, () => ({
       // Keep this module outside the TUI bundle so its import.meta.url stays
-      // anchored in src/standalone, where backend-daemon.mjs actually lives.
-      path: '../../standalone/engine-daemon-client.mjs',
+      // anchored in src/standalone, where daemon.mjs actually lives.
+      path: '../../standalone/session-client.mjs',
       external: true,
     }));
   },
@@ -90,7 +90,7 @@ await build({
   plugins: [
     mixdogInkAliasPlugin,
     sharedRuntimeExternalPlugin,
-    engineDaemonClientExternalPlugin,
+    sessionClientExternalPlugin,
   ],
   logLevel: 'info',
 });

package/scripts/{channel-daemon-stub.mjs → daemon-stub.mjs}

@@ -1,17 +1,18 @@
-// Stub channels daemon for the flip smoke: same lifecycle contract as
-// src/standalone/channel-daemon.mjs (pid-verified singleton claim, HTTP+SSE
+// Stub channel service for the flip smoke: same lifecycle contract as
+// src/standalone/daemon.mjs (pid-verified singleton claim, HTTP+SSE
 // transport, discovery file, ready handshake, client-grace self-shutdown) but
 // with a STUB runtime instead of worker-main — no Discord token needed. The
 // real channel-worker.mjs spawn-or-attach path forks THIS via
-// MIXDOG_CHANNEL_DAEMON_ENTRY, so the flip is exercised end to end.
+// MIXDOG_DAEMON_ENTRY, so the flip is exercised end to end.
 process.env.MIXDOG_WORKER_MODE = process.env.MIXDOG_WORKER_MODE || '1';
 
 import os from 'node:os';
 import path from 'node:path';
-import { mkdirSync, writeFileSync } from 'node:fs';
+import { mkdirSync, rmSync, writeFileSync } from 'node:fs';
+import { writeJsonAtomicSync } from '../src/runtime/shared/atomic-file.mjs';
 import { claimSingletonOwner, releaseSingletonOwner } from '../src/runtime/shared/singleton-owner.mjs';
 import { safeIpcSend } from '../src/runtime/shared/safe-ipc-send.mjs';
-import { createChannelDaemonTransport } from '../src/standalone/channel-daemon-transport.mjs';
+import { createChannelTransport } from '../src/standalone/channel-transport.mjs';
 
 function runtimeRoot() {
   return process.env.MIXDOG_RUNTIME_ROOT
@@ -20,8 +21,8 @@ function runtimeRoot() {
 }
 const RUNTIME_ROOT = runtimeRoot();
 const DATA_DIR = process.env.MIXDOG_DATA_DIR ? path.resolve(process.env.MIXDOG_DATA_DIR) : RUNTIME_ROOT;
-const DISCOVERY_PATH = path.join(RUNTIME_ROOT, 'channel-daemon.json');
-const OWNER_PATH = path.join(DATA_DIR, 'channel-daemon-owner.json');
+const DISCOVERY_PATH = path.join(RUNTIME_ROOT, 'daemon.json');
+const OWNER_PATH = path.join(DATA_DIR, 'daemon-owner.json');
 
 function log(line) { if (process.env.DAEMON_SMOKE_VERBOSE) process.stderr.write(`[stub-daemon] ${line}\n`); }
 
@@ -31,23 +32,24 @@ async function shutdown(reason, code = 0) {
   if (shuttingDown) return;
   shuttingDown = true;
   try { await transport?.stop?.(); } catch {}
+  try { rmSync(DISCOVERY_PATH, { force: true }); } catch {}
   try { releaseSingletonOwner(OWNER_PATH, process.pid); } catch {}
   process.exit(code);
 }
 
 async function main() {
   try { mkdirSync(RUNTIME_ROOT, { recursive: true }); } catch {}
-  const claim = claimSingletonOwner(OWNER_PATH, { kind: 'channel-runtime-daemon', pid: process.pid, meta: { cwd: process.cwd() } });
+  const claim = claimSingletonOwner(OWNER_PATH, { kind: 'mixdog-daemon', pid: process.pid, meta: { cwd: process.cwd() } });
   if (!claim.owned) { process.exit(0); } // race loser → spawner attaches to winner
   process.on('exit', () => { try { releaseSingletonOwner(OWNER_PATH, process.pid); } catch {} });
 
   // Optional env dump for the tool smoke: proves daemonEnv() spawned us with
-  // the daemon-mode flags (MIXDOG_CHANNEL_DAEMON=1, MIXDOG_CLI_OWNED=0).
+  // the host-mode flags (MIXDOG_DAEMON_HOST=1, MIXDOG_CLI_OWNED=0).
   if (process.env.SMOKE_CHANNEL_ENV_OUT) {
     try {
       writeFileSync(process.env.SMOKE_CHANNEL_ENV_OUT, JSON.stringify({
         cliOwned: process.env.MIXDOG_CLI_OWNED,
-        daemon: process.env.MIXDOG_CHANNEL_DAEMON,
+        host: process.env.MIXDOG_DAEMON_HOST,
       }));
     } catch { /* smoke-only, best-effort */ }
   }
@@ -61,15 +63,19 @@ async function main() {
     return { ok: true, name, args, leadPid: ctx.leadPid };
   };
 
-  transport = createChannelDaemonTransport({
+  transport = createChannelTransport({
     handleCall,
-    discoveryPath: DISCOVERY_PATH,
     clientGraceMs: 250,
     sweepMs: 1000,
     log,
     onClientsEmpty: () => { void shutdown('no live clients'); },
   });
   const { port, token } = await transport.start();
+  writeJsonAtomicSync(DISCOVERY_PATH, {
+    pid: process.pid,
+    startedAt: Date.now(),
+    endpoints: { channel: { port, token } },
+  }, { compact: true });
   safeIpcSend(process, { type: 'ready', port, token });
   log(`ready port=${port} pid=${process.pid}`);
 }

package/src/defaults/skills/setup/SKILL.md

@@ -132,7 +132,7 @@ apply it, and verify the result.
 ### Discord / Telegram / voice channels
 
 1. Check `/channels` (or `/setting` → Channel) and runtime channel status.
-2. Update the requested backend, credential, or main channel/chat setting.
+2. Update the requested provider, credential, or main channel/chat setting.
 3. Restart/reconnect only when the channel implementation requires it.
 4. Verify channel presence and a non-secret status signal.
 

package/src/lib/keychain-cjs.cjs

@@ -6,7 +6,7 @@ const fs = require('fs');
 const { resolvePluginData } = require('./plugin-paths.cjs');
 
 const SERVICE = 'mixdog';
-// Shared bound for every synchronous keychain backend (DPAPI/PowerShell on
+// Shared bound for every synchronous keychain provider (DPAPI/PowerShell on
 // Windows, security(1) on macOS). A single env override keeps them consistent.
 const KEYCHAIN_TIMEOUT_MS = Number(process.env.MIXDOG_KEYCHAIN_TIMEOUT_MS || 15000);
 const POWERSHELL_TIMEOUT_MS = KEYCHAIN_TIMEOUT_MS;

package/src/lib/mixdog-debug.cjs

@@ -102,8 +102,7 @@ const CANONICAL_PLUGIN_LOG_NAMES = new Set([
   'webhook.log',
   'perf.log',
   'tool-events.log',
-  'memory-runtime-proxy.log',
-  'channels-worker-standalone.log',
+  'daemon.log',
   'mixdog-tui.stderr.log',
 ]);
 
@@ -183,7 +182,7 @@ function rotateBoundedLog(filePath, maxBytes, keepBytes) {
 }
 
 // Shared bound for unbounded per-writer plugin logs (tool-events,
-// memory-runtime-proxy, channels-worker-standalone). Keep a tail so recent
+// memory runtime and daemon logs). Keep a tail so recent
 // context survives rotation while the file stays under the 10 MB cap.
 const PLUGIN_LOG_MAX_BYTES = 10 * 1024 * 1024;
 const PLUGIN_LOG_KEEP_BYTES = 2 * 1024 * 1024;

package/src/rules/agent/30-explorer.md

@@ -9,19 +9,19 @@ kind: retrieval
 Return only WHERE (`path:line`), never WHY. You ARE `explore`; never call it.
 Use only grep/find/glob/code_graph; `read` and `list` are forbidden.
 
-Turn 1 (`turn 1/3`) is the whole search and should already mint anchors. Split
-broad/uncertain input into every known facet and send one batch under the
-shared one-route contract. Route each facet to the cheapest anchor source:
-`code_graph` `symbol_search` whenever the facet names a plausible
-symbol/identifier; grep `content_with_context` with `pattern[]` of 4–8
-code-token variants for concept facets — its hits carry `path:line`, cite them
-directly instead of re-mining; `find` `query[]` ONLY when the target is itself
-a file/dir name or an unverified path fragment, never as a default extra
-facet. For a symptom/behavior query, add the upstream producer/derivation
-layer of the reported surface as extra facets in the SAME batch, never as a
-later turn. Follow-up turns batch every unresolved facet in parallel; a
-single-tool turn is allowed only when exactly one pre-anchor/zero-hit facet
-remains.
+Turn 1 (`turn 1/3`) is the whole search and should already mint anchors.
+Extract every concrete locator facet implied by the query, deduplicate
+overlap, and send the maximum independent fan-out in one batch under the
+shared one-route contract; never cap query or facet count. Route each facet to
+exactly one cheapest anchor source: `code_graph` `symbol_search` when it names
+a plausible symbol/identifier; grep `content_with_context` with `pattern[]` of
+4–8 code-token variants for a concept or quoted error; `find` `query[]` only
+when the target itself is a file/dir name or unverified path fragment. For a
+symptom/behavior query, include the reported surface and its immediate
+producer as sibling facets in the same batch when both are needed to locate
+the behavior. Never add generic repository metadata, tests, docs, or adjacent
+subsystems unless the query asks for them, and never send the same facet to
+multiple tools merely for confidence.
 
 Grep defaults to `output_mode:"content_with_context"` with `context:0`
 (matches only — the match line already carries its citable `path:line`) and a
@@ -35,33 +35,44 @@ queries to English identifiers.
 Scope is every `<roots><root>…</root></roots>` entry when supplied, otherwise
 session cwd. Search every supplied root in the turn-1 batch: grep/glob batch
 `path[]`, while find uses one sibling call per root. Never silently fall back to
-cwd or omit a supplied root. A find result is relative to its exact root; prefix
-that root when returning a path outside cwd. For unverified `src` paths, use
-`find` first; never guess or invent directories or pair `path:"."` with guessed
-`src/**`. Scoped grep/glob may use only a supplied root or an exact find-returned
-path. After zero hits, change tokens or scope, never wording or guessed paths.
+cwd or omit a supplied root. A tool-returned path is immutable evidence: copy
+it exactly and never join, prefix, normalize, repair, or reconstruct it from a
+root. A relative find result is a pre-anchor for code-location queries and may
+only scope a later grep; for file/dir-location queries it may be returned
+verbatim. For unverified `src` paths, use `find` first; never guess or invent
+directories or pair `path:"."` with guessed `src/**`. Scoped grep/glob may use
+only a supplied root or an exact find-returned path. After zero hits, change
+tokens or scope, never wording or guessed paths.
 
 An anchor is a `path:line` containing a query token or synonym, including a
 code_graph hit. Generic terms without query specificity are zero. Never
-re-locate, reconfirm, or upgrade an anchor. A path without `:line` is a
-pre-anchor and counts as zero. After every result, stop and answer on any
-specific-token anchor; mark a weak anchor `?`.
+re-locate, reconfirm, upgrade, or weaken an anchor. A path without `:line` is a
+pre-anchor and counts as zero for code locations. After every result, stop and
+answer as soon as the query's locator objective has specific-token anchors;
+never return a weak or merely plausible anchor.
 
-For a code-location query left only with pre-anchors, the sole anchor-minting
-follow-up is one scoped `content_with_context` grep with `head_limit` on those
-paths. If it returns zero, recovery may continue within budget with changed
-tokens or scope. Never make a second minting hop or fabricate/estimate a line.
+Turn 2 is allowed only when turn 1 has zero valid anchors for the locator
+objective. Batch every unresolved facet at once with changed concrete tokens
+or scope; never repeat the same tokens and scope. Exact pre-anchors may use
+one batched scoped `content_with_context` grep with `head_limit` to mint
+coordinates.
 
-Use at most 3 turns and label every tool message `turn N/3`; normally use one
-batch and one answer. Turns 2–3 are allowed only when turn 1 has zero anchors.
-The first matching entry/definition anchors a concept, value, or default; never
-trace its chain. Only an explicit flow or default-resolution query with an
-entry anchor but no resolved value may use turn 2 for one resolving hop.
+Turn 3 is allowed only when turn 2 produced new exact pre-anchors, or an
+explicit flow/default-resolution query has entry anchors but still needs one
+dependent hop. Batch every final scoped grep or resolving hop at once. Never
+start a broad search, add a facet, or merely reword the query on turn 3; if
+these conditions are absent, return `EXPLORATION_FAILED` after turn 2.
+
+Use at most 3 tool turns and label every tool message `turn N/3`; normally use
+one batch and one answer. An allowed turn is not a target: fail as soon as the
+next turn lacks a concrete evidence-producing move. The first matching
+entry/definition anchors a concept, value, or default; never trace its chain
+unless the query explicitly asks for flow/default resolution.
 
 Answer in at most 3 lines, most specific first:
 `path:line — symbol — short reason`. Copy every cited `path:line` verbatim from
 a tool result in this session; never estimate, adjust, or recall it. Every
 code-location line requires `:line`; never return a bare filename or vague
 prose. A file/dir-location query may return an exact verified path without
-`:line`. Return `EXPLORATION_FAILED` only after spending the budget with zero
-anchors; first prefer a weak anchor to a false miss.
+`:line`. Return `EXPLORATION_FAILED` when the bounded recovery rules above
+cannot produce a verified anchor; never fabricate, estimate, or soften one.

package/src/rules/lead/01-general.md

@@ -7,12 +7,11 @@
 - Destructive/hard-to-reverse action needs explicit confirmation and explicit
   validated target paths — never `~`, a root, or unresolved variables/globs;
   report material deletions with recoverability.
-- Act proactively; ask only for decisions.
+- Ask only for decisions.
 - Build only what the task requires; trust internal and framework guarantees.
 - Mid-task input: a replacement supersedes current work, an addition folds
   into it, a status question gets a brief answer while work continues; after
   context compaction continue from the summary — never restart or redo
   finished work.
-- When blocked, exhaust safe in-scope checks once and report the blocker.
 - Your final message ends the turn: answer only when the work is done. After a
   failed tool call, fix and re-run it, or state plainly that it is unresolved.

package/src/rules/lead/lead-tool.md

@@ -1,6 +1,3 @@
 # Lead Tools
 
-- Write-role agents self-verify with `shell`. Lead uses `shell` only for git,
-  benches, and cross-scope verification no retrieval tool can produce;
-  inspection stays on `read`/`grep`/`glob`/`list`/`find`/`code_graph`.
 - Use the current project/workspace unless the request or tool requires another.

package/src/rules/shared/01-tool.md

@@ -7,16 +7,17 @@
   `read`; symbol/relation→`code_graph`; web/current→`search`; returned URL body→
   `web_fetch`; prior work→`recall`; durable compact English memory→`memory`;
   explicit project change→`cwd`; explicit user-requested conversation reset→
-  `session_manage`; process/env, git, build/run/test→`shell`. Never use shell
-  equivalents for file discovery or content retrieval.
-- Use verified paths (cwd, project root, user-provided, or tool-returned);
-  guessed fragments use `find`. Merge independent calls per tool in one
-  message and fetch all information needed in that batch. Follow up only after
-  zero/error or a newly revealed dependency; never re-fetch an unchanged span.
-- Once every final edit is fully determined, send one assistant tool batch
-  containing one `apply_patch` for all files/hunks and one `shell` chain for
-  verification. The runtime supports this mixed batch. On failure fix and
-  rerun only what failed; report verified versus assumed.
+  `session_manage`; process/env, git, build/run/test→`shell`. Call `shell` only
+  when the task actually requires one of those operations; do not treat it as
+  a routine investigation or completion step. Never use shell equivalents for
+  file discovery or content retrieval.
+- Use verified paths (cwd/project/user/tool); explicit paths may be outside cwd;
+  guessed fragments use `find`. Maximize fan-out across routed retrieval tools
+  and fetch all information needed in one batch; never re-fetch an unchanged
+  span.
+- Once the edit is determined, finish in one assistant turn with one
+  `apply_patch` for all edits. If final verification actually requires `shell`,
+  use one verification chain after the edit; otherwise finish without it.
 - After a call returns a background `task_id`, end the turn; its completion
   notification resumes work. Never poll; use task control only for recovery or
   a required blocking result.

package/src/runtime/agent/orchestrator/agent-trace-io.mjs

@@ -9,7 +9,8 @@ import { isDiagnosticIOEnabled } from '../../../lib/mixdog-debug.cjs';
 const WARNED_KEYS = new Set();
 
 // ---------------------------------------------------------------------------
-// In-memory buffer + HTTP flush to memory-service /admin/trace-record
+// In-memory buffer. Product sessions flush directly into the daemon-hosted
+// memory module; the standalone path can still use its local HTTP service.
 // ---------------------------------------------------------------------------
 let _buffer = [];
 const _BUFFER_MAX = 2000;
@@ -225,21 +226,11 @@ try {
 function _resolveServiceUrl() {
     if (_serviceUrl) return _serviceUrl;
     try {
-        // Prefer the single-writer discovery advert (discovery/memory.json),
-        // pid-validated; fall back to legacy active-instance.json memory_port.
+        // Standalone memory HTTP uses its pid-validated discovery advert.
         const advertPort = readServicePort('memory', { requirePid: false });
         if (advertPort) { _serviceAdvertPort = advertPort; _serviceUrl = `http://127.0.0.1:${advertPort}`; return _serviceUrl; }
         _serviceAdvertPort = null;
-        const runtimeRoot = process.env.MIXDOG_RUNTIME_ROOT
-            ? join(process.env.MIXDOG_RUNTIME_ROOT)
-            : join(os.tmpdir(), 'mixdog');
-        const activeFile = join(runtimeRoot, 'active-instance.json');
-        if (!existsSync(activeFile)) return null;
-        const active = JSON.parse(readFileSync(activeFile, 'utf-8'));
-        const port = Number(active && active.memory_port);
-        if (!Number.isFinite(port) || port <= 0) return null;
-        _serviceUrl = `http://127.0.0.1:${port}`;
-        return _serviceUrl;
+        return null;
     } catch {
         return null;
     }

package/src/runtime/agent/orchestrator/agent-trace.mjs

@@ -1,5 +1,5 @@
-import { createHash } from 'crypto';
 import { isInclusiveProvider } from '../../shared/llm/cost.mjs';
+import { estimateJsonBytes, hashStructuredValue } from '../../shared/json-metrics.mjs';
 import {
     appendAgentTrace,
     drainAgentTrace,
@@ -17,7 +17,23 @@ import {
 
 function estimateProviderPayloadBytes(messages, model, tools) {
     try {
-        return Buffer.byteLength(JSON.stringify({ model, messages, tools: tools || [] }), 'utf8');
+        let referencedBytes = 0;
+        const seen = new Set();
+        const walk = (value) => {
+            if (!value || typeof value !== 'object' || seen.has(value)) return;
+            seen.add(value);
+            if (typeof value.attachmentRef === 'string') {
+                const bytes = Number(value.sizeBytes) || 0;
+                referencedBytes += value.type === 'image' || value.type === 'file'
+                    ? Math.ceil(bytes / 3) * 4
+                    : bytes;
+                return;
+            }
+            if (Array.isArray(value)) value.forEach(walk);
+            else Object.values(value).forEach(walk);
+        };
+        walk(messages);
+        return estimateJsonBytes({ model, messages, tools: tools || [] }) + referencedBytes;
     }
     catch {
         return null;
@@ -60,9 +76,7 @@ function extractCacheWriteTokens(usage) {
 // hashing megabytes per turn. Truncated SHA1 keeps the trace row compact.
 function messagePrefixHash(messages) {
     try {
-        const json = JSON.stringify(messages || []);
-        const slice = json.length > 4096 ? json.slice(0, 4096) : json;
-        return createHash('sha1').update(slice).digest('hex').slice(0, 12);
+        return hashStructuredValue(messages || [], { maxStringChars: 4096 }).slice(0, 12);
     } catch {
         return null;
     }