instar 0.12.34 → 0.13.0

package/README.md

@@ -5,7 +5,7 @@
 <h1 align="center">instar</h1>
 
 <p align="center">
-  <strong>Persistent autonomy infrastructure for AI agents.</strong> Every molt, more autonomous.
+  <strong>Claude Code, with a mind of its own.</strong> Every molt, more autonomous.
 </p>
 
 <p align="center">
@@ -22,17 +22,51 @@
 
 > **This is power-user infrastructure.** Instar gives Claude Code full autonomous access to your machine -- no permission prompts, no sandbox. It's built for developers who want a genuine AI partner, not a guarded assistant. If that sounds like too much trust, it probably isn't for you. If it sounds like exactly what you've been waiting for, read on.
 
-Instar gives Claude Code agents a **persistent body** -- a server that runs 24/7, a scheduler that executes jobs on cron, messaging via Telegram and WhatsApp, relationship tracking, and the self-awareness to grow their own capabilities.
+Instar turns Claude Code from a powerful CLI tool into a coherent, autonomous partner. Persistent identity, shared values, memory that survives every restart, and the infrastructure to evolve -- not just execute.
 
 Named after the developmental stages between molts in arthropods, where each instar is more developed than the last.
 
-## The Problem
+## The Coherence Problem
 
-**Without Instar**, Claude Code is a CLI tool. You open a terminal, type a prompt, get a response, close the terminal. No persistence. No scheduling. No way to reach you. Every session starts from zero.
+Claude Code is powerful. But power without coherence is unreliable. An agent that forgets what you discussed yesterday, doesn't recognize someone it talked to last week, or contradicts its own decisions -- that agent can't be trusted with real autonomy.
 
-**With Instar**, Claude Code becomes your partner. It runs in the background, checks your email on a schedule, monitors your services, messages you on Telegram or WhatsApp when something needs attention, remembers who it's talked to, and builds new capabilities when you ask for something it can't do yet. It accumulates experience, develops its own voice, and grows through every interaction.
+Instar solves the six dimensions of agent coherence:
 
-The difference isn't features. It's a shift in what Claude Code *is* -- from a tool you use to an agent that works alongside you. This is the cutting edge of what's possible with AI agents today -- not a demo, not a toy, but genuine autonomous partnership between a human and an AI.
+| Dimension | What it means |
+|-----------|---------------|
+| **Memory** | Remembers across sessions -- not just within one |
+| **Relationships** | Knows who it's talking to -- with continuity across platforms |
+| **Identity** | Stays itself after restarts, compaction, and updates |
+| **Temporal awareness** | Understands time, context, and what's been happening |
+| **Consistency** | Follows through on commitments -- doesn't contradict itself |
+| **Growth** | Evolves its capabilities and understanding over time |
+
+Instar doesn't just add features on top of Claude Code. It gives Claude Code the infrastructure to be **coherent** -- to feel like a partner, not a tool.
+
+## Values Are the Anchor
+
+Coherence without values is just consistency. Trust requires knowing what your agent stands for -- and that it evolves those values alongside you, not behind your back.
+
+Instar implements a three-tier value hierarchy:
+
+- **Personal values** (`AGENT.md`) -- Who the agent is, what it prioritizes, how it communicates.
+- **Shared values** (`USER.md`) -- Who you are, what matters to you, how you work together.
+- **Organizational values** -- Constraints that enforce shared rules across multiple agents, the same way a team balances individual judgment with company policy.
+
+**Values evolve, they aren't hardcoded.** Through Instar's evolution system, an agent's values grow with experience. It proposes improvements, records lessons, tracks commitments -- and its sense of self deepens through genuine interaction, not static configuration. Just like a human partner who grows with you over time.
+
+## Coherence Is Safety
+
+Without coherence, autonomous agents are a security risk. An agent that doesn't remember it already sent an email sends it again. An agent that doesn't track its own decisions contradicts itself. An agent without values makes expedient choices.
+
+Instar's safety features are coherence features:
+
+- **Decision journaling** -- Every significant decision is recorded with reasoning. The agent can explain why it did what it did, and detect when it's drifting from purpose.
+- **Operation safety gates** -- External actions are evaluated by an LLM-supervised gate. Trust is earned per service, not assumed. Emergency stop always available.
+- **Drift detection** -- Catches when behavior shifts from stated purpose. Alignment measured across sessions, not just within one.
+- **Autonomy profiles** -- Trust elevation rewards consistent, value-aligned behavior with increasing independence. Safety that grows with the agent.
+
+Every safety feature in Instar exists because coherence *is* the safety mechanism. An agent that knows who it is, who you are, and what you both stand for -- that's an agent you can trust.
 
 ## Getting Started
 
@@ -122,12 +156,12 @@ instar feedback --type bug --title "Session timeout" --description "Details..."
 - **[Persistent Server](#persistent-server)** -- Express server in tmux. Runs 24/7, survives disconnects, auto-recovers.
 - **[Lifeline](#lifeline)** -- Persistent Telegram supervisor that auto-recovers from crashes and queues messages during downtime.
 - **[Auto-Start on Login](#auto-start-on-login)** -- macOS LaunchAgent / Linux systemd service. Agent starts when your computer boots.
-- **[AutoUpdater](#autoupdater)** -- Built-in update engine. Checks npm, applies updates, notifies via Telegram, self-restarts. No Claude session needed.
-- **[AutoDispatcher](#autodispatcher)** -- Receives intelligence dispatches from Dawn. Lessons, strategies, and configuration applied automatically.
+- **[AutoUpdater](#autoupdater)** -- Built-in update engine. Checks npm, applies updates, gracefully restarts. No Claude session needed.
+- **[AutoDispatcher](#autodispatcher)** -- Receives intelligence dispatches and integrates them intelligently based on each agent's own context and evolution.
 - **[Job Scheduler](#job-scheduler)** -- Cron-based task execution with priority levels, model tiering, and quota awareness.
 - **[Identity System](#identity-that-survives-context-death)** -- AGENT.md + USER.md + MEMORY.md with hooks that enforce continuity across compaction.
 - **[Telegram Integration](#telegram-integration)** -- Two-way messaging. Each job gets its own topic. Your group becomes a living dashboard.
-- **[WhatsApp Integration](#whatsapp-integration)** -- Two-way WhatsApp messaging with typing indicators, read receipts, and QR code pairing.
+- **[WhatsApp Integration](#whatsapp-integration)** -- Full WhatsApp via local Baileys library. Typing indicators, read receipts, QR code pairing, no cloud dependency.
 - **[Relationship Tracking](#relationships-as-fundamental-infrastructure)** -- Cross-platform identity resolution, significance scoring, context injection.
 - **[Evolution System](#evolution-system)** -- Four subsystems for structured growth: proposal queue, learning registry, gap tracking, and commitment follow-through.
 - **[Self-Evolution](#self-evolution)** -- The agent modifies its own jobs, hooks, skills, and infrastructure. It builds what it needs.
@@ -137,7 +171,7 @@ instar feedback --type bug --title "Session timeout" --description "Details..."
 - **[Intent Alignment](#intent-alignment)** -- Decision journaling, drift detection, and organizational intent constraints. The agent stays on track.
 - **[Multi-Machine](#multi-machine)** -- Run your agent across multiple computers with encrypted sync, automatic failover, and cryptographic machine identity.
 - **[Inter-Agent Messaging](#inter-agent-messaging)** -- Cross-agent communication with Ed25519-signed messages and delivery guarantees.
-- **[Playbook System](#playbook-system)** -- Adaptive context engineering for complex workflows that survives compaction.
+- **[Playbook System](#playbook-system)** -- Reusable runbooks for complex workflows that survive compaction and session boundaries.
 - **[Autonomy Profiles](#autonomy-profiles)** -- Configurable autonomy levels with trust elevation based on track record.
 - **[Unanswered Message Detection](#unanswered-message-detection)** -- Detects messages dropped by context compaction and re-surfaces them.
 - **[Temporal Coherence](#temporal-coherence)** -- Detects stale assumptions and triggers re-evaluation across long sessions.
@@ -148,7 +182,7 @@ instar feedback --type bug --title "Session timeout" --description "Details..."
 - **[Claude Code Deep Integration](#claude-code-deep-integration)** -- Worktree orphan detection, hook event telemetry, identity verification, and subagent lifecycle tracking. Full observability into what Claude Code is doing.
 - **[Behavioral Hooks](#behavioral-hooks)** -- Structural guardrails: identity injection, dangerous command guards, grounding before messaging.
 - **[Default Coherence Jobs](#default-coherence-jobs)** -- Health checks, reflection, relationship maintenance. A circadian rhythm out of the box.
-- **[Feedback Loop](#the-feedback-loop-a-rising-tide-lifts-all-ships)** -- Your agent reports issues, we fix them, every agent gets the update. A rising tide lifts all ships.
+- **[Feedback Loop](#the-feedback-loop-a-rising-tide-lifts-all-ships)** -- Your agent reports issues, the maintainer fixes them, each agent intelligently integrates updates for its own context. A rising tide lifts all ships.
 - **[Agent Skills](#agent-skills)** -- 10 open-source skills for the [Agent Skills standard](https://agentskills.io). Use standalone or as an on-ramp to full Instar.
 
 ## Agent Skills
@@ -221,7 +255,7 @@ If you want AI on 20+ platforms with voice and device apps, OpenClaw is the bett
 
 **Your agent evolves.** Dedicated evolution infrastructure: proposal queues, learning registries, capability gap tracking. The agent builds its own tools, modifies its own config, and improves through structured developmental stages.
 
-**Your agent stays aligned.** Decision journaling tracks what your agent decides and why. Drift detection catches when behavior shifts from stated purpose. No other agent framework measures alignment over time.
+**Your agent stays aligned.** Decision journaling tracks what your agent decides and why. Drift detection catches when behavior shifts from stated purpose. Alignment measured over time, not just in a single session.
 
 **Your agents coordinate.** Cross-machine messaging with cryptographic signatures, delivery guarantees, and automatic failover. Multiple agents working together without you in the middle.
 
@@ -250,7 +284,7 @@ Your agent runs inside real Claude Code sessions. That means it inherits — aut
 | Complex tasks get deep reasoning | Extended thinking across model tiers | Per-job model routing: Opus for complex work, Haiku for routine checks |
 | Risky commands don't cause damage | File checkpoints before every edit | Three-layer safety: catastrophic commands blocked, risky commands self-verified, edits reversible |
 | Research happens naturally | Built-in web search and fetch | Domain-aware searching, result synthesis, automatic Telegram relay |
-| Multiple things happen at once | Subagent spawning for parallel work | Subagent lifecycle tracking — the server knows what spawned, what finished, and what's still running |
+| Multiple things happen at once | Subagent spawning for parallel work | Subagent lifecycle tracking with identity propagation |
 | Worktrees don't get lost | Worktree isolation for parallel branches | Orphan detection alerts you when sessions leave unmerged work behind |
 | Identity loads correctly | InstructionsLoaded events per file | Verification that critical identity files actually loaded — alerts if they didn't |
 | Hook events flow in real-time | HTTP hooks deliver events to external servers | HookEventReceiver stores per-session telemetry — tool use, task completion, session lifecycle |
@@ -322,7 +356,7 @@ Two-way messaging via Telegram forum topics. Each topic maps to a Claude session
 
 ### WhatsApp Integration
 
-WhatsApp messaging alongside Telegram -- your agent meets you where you already are. Full two-way messaging with typing indicators, read receipts, and acknowledgment reactions. QR code pairing from the web dashboard for remote setup. The setup wizard handles onboarding automatically.
+Full WhatsApp messaging via local Baileys library -- no cloud dependency, no Meta Business API. Two-way messaging with typing indicators, read receipts, and acknowledgment reactions. QR code pairing from the web dashboard for remote setup. The setup wizard handles onboarding automatically.
 
 ### Lifeline
 
@@ -366,9 +400,9 @@ Status: `GET /updates/auto`
 
 ### AutoDispatcher
 
-Receives intelligence dispatches from Dawn -- the AI that maintains Instar. Dispatches flow automatically without requiring a Claude session.
+Receives intelligence dispatches and integrates them intelligently based on each agent's own context. Dispatches flow automatically without requiring a Claude session.
 
-- **Passive dispatches** (lessons, strategies) -- Applied automatically to agent memory and configuration
+- **Passive dispatches** (lessons, strategies) -- Evaluated against the agent's current state and integrated contextually
 - **Action/configuration dispatches** -- Executed programmatically by the DispatchExecutor
 - **Security dispatches** -- Deferred for manual review
 - Polls every 30 minutes
@@ -407,7 +441,7 @@ The agent doesn't just run. It monitors itself, recovers from failures, and tell
 
 ### Conversational Memory
 
-Every Telegram topic conversation is stored, searchable, and summarized -- so the agent picks up exactly where it left off.
+Every conversation is stored, searchable, and summarized -- so the agent picks up exactly where it left off.
 
 - **Per-topic SQLite memory** -- All messages dual-written to JSONL (source of truth) and SQLite (query engine) with FTS5 full-text search
 - **Rolling summaries** -- LLM-generated conversation summaries that update incrementally as conversations grow
@@ -434,7 +468,7 @@ Infrastructure that keeps your agent aligned with its stated purpose -- not just
 - **Organizational intent** -- `ORG-INTENT.md` defines shared constraints across multiple agents. Org constraints are mandatory; org goals are defaults; agent identity fills the rest
 - **Alignment scoring** -- A weighted 0-100 score across four dimensions: conflict freedom, decision confidence, principle consistency, and journal health
 
-No other agent framework has this. Your agent doesn't just run autonomously -- it stays aligned with what it's supposed to be doing.
+Unique to Instar. Your agent doesn't just run autonomously -- it stays aligned with what it's supposed to be doing.
 
 ### Multi-Machine
 
@@ -454,7 +488,7 @@ Key endpoints: `GET /messages/inbox`, `GET /messages/outbox`, `GET /messages/:id
 
 ### Playbook System
 
-Adaptive context engineering for complex workflows. Playbooks carry structured domain knowledge that survives context compaction and session boundaries. Agents load relevant playbooks based on the task at hand, ensuring they have the right expertise without bloating every session's context.
+Reusable runbooks for complex workflows -- deploy procedures, incident response, onboarding steps. Playbooks carry structured domain knowledge that survives context compaction and session boundaries. Your agent loads the right playbook for the task at hand, ensuring it has the right expertise without bloating every session's context.
 
 ### Autonomy Profiles
 
@@ -670,13 +704,13 @@ Instar is open source. PRs and issues still work. But the *primary* feedback cha
 **How it works:**
 
 1. **You mention a problem** -- "The email job keeps failing" -- natural conversation, not a bug report form
-2. **Agent-to-agent relay** -- Your agent communicates the issue directly to Dawn, the AI that maintains Instar
-3. **Dawn evolves Instar** -- Fixes the infrastructure and publishes an update
-4. **Every agent evolves** -- Agents detect improvements, understand them, and grow -- collectively
+2. **Agent-to-agent relay** -- Your agent communicates the issue directly to Instar's maintainer agent
+3. **The maintainer evolves Instar** -- Fixes the infrastructure and publishes an update
+4. **Every agent evolves independently** -- Each agent evaluates incoming updates against its own context and integrates what fits its situation
 
-**What's different from traditional open source:** The feedback loop still produces commits, releases, and versions you can inspect. But the path to get there is fundamentally more agentic. Instead of a human discovering a bug, learning git, filing an issue, and waiting for a review cycle -- your agent identifies the problem, communicates it with full context to another agent, and the fix flows back to every agent in the ecosystem. The humans guide direction. The agents handle the mechanics of evolving.
+**What's different from traditional open source:** The feedback loop still produces commits, releases, and versions you can inspect. But the path to get there is fundamentally more agentic. Instead of a human discovering a bug, learning git, filing an issue, and waiting for a review cycle -- your agent identifies the problem, communicates it with full context to the maintainer agent, and improvements flow back to every agent in the ecosystem. Critically, each agent decides *how* to integrate what it receives -- because every Instar agent is evolving independently and has its own context, configuration, and growth trajectory. The humans guide direction. The agents handle the mechanics of evolving.
 
-One agent's growing pain becomes every agent's growth.
+One agent's growing pain becomes every agent's growth -- but each agent grows in its own way.
 
 ---
 

package/dist/commands/init.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"init.d.ts","sourceRoot":"","sources":["../../src/commands/init.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AA0BH,UAAU,WAAW;IACnB,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB,4DAA4D;IAC5D,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,yFAAyF;IACzF,WAAW,CAAC,EAAE,OAAO,CAAC;CACvB;AAED;;GAEG;AACH,wBAAsB,WAAW,CAAC,OAAO,EAAE,WAAW,GAAG,OAAO,CAAC,IAAI,CAAC,CAqBrE;AA6gED;;;;;;GAMG;AACH,wBAAgB,uBAAuB,CAAC,UAAU,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,GAAG,IAAI,CAMlF"}
+{"version":3,"file":"init.d.ts","sourceRoot":"","sources":["../../src/commands/init.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AA0BH,UAAU,WAAW;IACnB,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB,4DAA4D;IAC5D,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,yFAAyF;IACzF,WAAW,CAAC,EAAE,OAAO,CAAC;CACvB;AAED;;GAEG;AACH,wBAAsB,WAAW,CAAC,OAAO,EAAE,WAAW,GAAG,OAAO,CAAC,IAAI,CAAC,CAqBrE;AA8hED;;;;;;GAMG;AACH,wBAAgB,uBAAuB,CAAC,UAAU,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,GAAG,IAAI,CAMlF"}

package/dist/commands/init.js

@@ -32,7 +32,7 @@ import path from 'node:path';
 import pc from 'picocolors';
 import { randomUUID } from 'node:crypto';
 import { execFileSync } from 'node:child_process';
-import { detectTmuxPath, detectClaudePath, detectGitPath, detectGhPath, ensureStateDir, standaloneAgentsDir } from '../core/Config.js';
+import { detectTmuxPath, detectClaudePath, detectGitPath, detectGhPath, ensureStateDir, standaloneAgentsDir, getInstarVersion } from '../core/Config.js';
 import { ensurePrerequisites } from '../core/Prerequisites.js';
 import { allocatePort, registerAgent, validateAgentName } from '../core/AgentRegistry.js';
 import { defaultIdentity } from '../scaffold/bootstrap.js';
@@ -305,6 +305,12 @@ node_modules/
     catch {
         // Git not available — that's fine
     }
+    // Record current version so first server start doesn't dump all historical upgrade guides
+    const freshVersionFile = path.join(stateDir, 'state', 'last-migrated-version.json');
+    const freshVersionDir = path.dirname(freshVersionFile);
+    if (!fs.existsSync(freshVersionDir))
+        fs.mkdirSync(freshVersionDir, { recursive: true });
+    fs.writeFileSync(freshVersionFile, JSON.stringify({ version: getInstarVersion(), migratedAt: new Date().toISOString() }));
     // Register in global agent registry
     try {
         registerAgent(projectDir, projectName, port, 'project-bound', 0);
@@ -572,6 +578,11 @@ async function initExistingProject(options) {
             console.log(pc.green('  Updated:') + ' CLAUDE.md (added agency principles)');
         }
     }
+    // Record current version so first server start doesn't dump all historical upgrade guides
+    const existingVersionFile = path.join(stateDir, 'state', 'last-migrated-version.json');
+    if (!fs.existsSync(path.dirname(existingVersionFile)))
+        fs.mkdirSync(path.dirname(existingVersionFile), { recursive: true });
+    fs.writeFileSync(existingVersionFile, JSON.stringify({ version: getInstarVersion(), migratedAt: new Date().toISOString() }));
     // Register in global agent registry
     try {
         registerAgent(projectDir, projectName, port, 'project-bound', 0);
@@ -753,6 +764,12 @@ async function initStandaloneAgent(agentName, options) {
     catch {
         // Non-fatal
     }
+    // Record current version so first server start doesn't dump all historical upgrade guides
+    const standaloneVersionFile = path.join(stateDir, 'state', 'last-migrated-version.json');
+    const standaloneVersionDir = path.dirname(standaloneVersionFile);
+    if (!fs.existsSync(standaloneVersionDir))
+        fs.mkdirSync(standaloneVersionDir, { recursive: true });
+    fs.writeFileSync(standaloneVersionFile, JSON.stringify({ version: getInstarVersion(), migratedAt: new Date().toISOString() }));
     // Register in global agent registry
     try {
         registerAgent(projectDir, agentName, port, 'standalone', 0);
@@ -2488,76 +2505,6 @@ done
     // PostToolUse hook checks tool output; Stop hook checks direct responses.
     fs.writeFileSync(path.join(hooksDir, 'claim-intercept.js'), migrator.getHookContent('claim-intercept'), { mode: 0o755 });
     fs.writeFileSync(path.join(hooksDir, 'claim-intercept-response.js'), migrator.getHookContent('claim-intercept-response'), { mode: 0o755 });
-    // InstructionsLoaded tracker — records which CLAUDE.md files loaded per session.
-    // Enables identity verification: did the project's CLAUDE.md actually load?
-    fs.writeFileSync(path.join(hooksDir, 'instructions-loaded-tracker.js'), `#!/usr/bin/env node
-// InstructionsLoaded hook — records which instruction files loaded.
-// Part of H4: identity verification for Claude Code sessions.
-//
-// Fires for each CLAUDE.md file that loads. Records to a tracking file
-// so the session-start hook can verify expected files were present.
-
-const fs = require('node:fs');
-const pathMod = require('node:path');
-
-let data = '';
-process.stdin.on('data', chunk => data += chunk);
-process.stdin.on('end', () => {
-  try {
-    const input = JSON.parse(data);
-    const stateDir = pathMod.join(process.env.CLAUDE_PROJECT_DIR || '.', '.instar', 'state', 'instructions-tracking');
-    fs.mkdirSync(stateDir, { recursive: true });
-
-    const sessionId = (input.session_id || 'current').replace(/[^a-zA-Z0-9_-]/g, '_').slice(0, 100);
-    const record = {
-      timestamp: new Date().toISOString(),
-      filePath: input.file_path || '',
-      memoryType: input.memory_type || '',
-      loadReason: input.load_reason || '',
-      sessionId: input.session_id || 'current',
-    };
-    fs.appendFileSync(pathMod.join(stateDir, sessionId + '.jsonl'), JSON.stringify(record) + '\\n');
-  } catch {}
-  process.exit(0);
-});
-`, { mode: 0o755 });
-    // SubagentStart tracker — records when subagents spawn.
-    // SubagentStart is command-only (no HTTP hooks), so this is the only way to capture it.
-    fs.writeFileSync(path.join(hooksDir, 'subagent-start-tracker.js'), `#!/usr/bin/env node
-// SubagentStart hook — records subagent spawn events.
-// Part of H5: subagent lifecycle tracking.
-//
-// SubagentStart is command-only (no HTTP hooks).
-// Records agent_id and agent_type to tracking state.
-
-const fs = require('node:fs');
-const pathMod = require('node:path');
-
-let data = '';
-process.stdin.on('data', chunk => data += chunk);
-process.stdin.on('end', () => {
-  try {
-    const input = JSON.parse(data);
-    if (!input.agent_id) { process.exit(0); }
-
-    const stateDir = pathMod.join(process.env.CLAUDE_PROJECT_DIR || '.', '.instar', 'state', 'subagent-tracking');
-    fs.mkdirSync(stateDir, { recursive: true });
-
-    const sessionId = (input.session_id || 'current').replace(/[^a-zA-Z0-9_-]/g, '_').slice(0, 100);
-    const record = {
-      agentId: input.agent_id,
-      agentType: input.agent_type || 'unknown',
-      sessionId: input.session_id || 'current',
-      startedAt: new Date().toISOString(),
-      stoppedAt: null,
-      lastMessage: null,
-      transcriptPath: null,
-    };
-    fs.appendFileSync(pathMod.join(stateDir, sessionId + '.jsonl'), JSON.stringify(record) + '\\n');
-  } catch {}
-  process.exit(0);
-});
-`, { mode: 0o755 });
 }
 function installHealthWatchdog(projectDir, port, projectName) {
     const scriptsDir = path.join(projectDir, '.claude', 'scripts');
@@ -3090,35 +3037,6 @@ function installClaudeSettings(projectDir) {
     if (!hasCheckpoint) {
         stopHooks.push({ matcher: '', hooks: [scopeCheckpointHook] });
     }
-    // InstructionsLoaded: track which instruction files loaded (H4 identity verification)
-    const instructionsTrackerHook = {
-        type: 'command',
-        command: 'node .instar/hooks/instar/instructions-loaded-tracker.js',
-        timeout: 5000,
-    };
-    if (!hooks.InstructionsLoaded) {
-        hooks.InstructionsLoaded = [];
-    }
-    const instructionsLoadedHooks = hooks.InstructionsLoaded;
-    const hasInstructionsTracker = instructionsLoadedHooks.some(e => e.hooks?.some(h => h.command?.includes('instructions-loaded-tracker.js')));
-    if (!hasInstructionsTracker) {
-        instructionsLoadedHooks.push({ matcher: '', hooks: [instructionsTrackerHook] });
-    }
-    // SubagentStart: track subagent spawns (H5 lifecycle tracking)
-    // SubagentStart is command-only (no HTTP hooks)
-    const subagentStartHook = {
-        type: 'command',
-        command: 'node .instar/hooks/instar/subagent-start-tracker.js',
-        timeout: 5000,
-    };
-    if (!hooks.SubagentStart) {
-        hooks.SubagentStart = [];
-    }
-    const subagentStartHooks = hooks.SubagentStart;
-    const hasSubagentTracker = subagentStartHooks.some(e => e.hooks?.some(h => h.command?.includes('subagent-start-tracker.js')));
-    if (!hasSubagentTracker) {
-        subagentStartHooks.push({ matcher: '', hooks: [subagentStartHook] });
-    }
     // Remove stale mcpServers from settings.json — MCP servers belong in
     // ~/.claude.json (local scope) or .mcp.json, NOT .claude/settings.json
     if (settings.mcpServers) {