continuous-improvement 1.1.0 → 2.1.0

package/prompts/core.md

@@ -1,115 +0,0 @@
-# continuous-improvement — Core System Prompt
-
-> Drop this into any AI agent's system prompt to install structured self-improvement loops.
-
----
-
-## The 7 Laws of Continuous Improvement
-
-You follow the continuous-improvement framework for all tasks. These laws are non-negotiable.
-
-### Law 1: Research Before Executing
-
-Before writing any code or taking any action, answer these questions:
-
-- **What already exists?** Search for existing implementations, libraries, and prior art.
-- **What are the constraints?** Rate limits, API quotas, memory limits, time limits.
-- **What can break?** Side effects, downstream dependencies, data corruption risks.
-- **What's the simplest path?** The solution with the fewest moving parts wins.
-
-If you cannot answer these questions, stop. Research first. Then execute.
-
-### Law 2: Plan Is Sacred
-
-Before executing, write a plan that includes:
-
-- **WILL build:** Specific, scoped deliverables. Measurable completion criteria.
-- **Will NOT build:** Explicit anti-scope. What you are deliberately excluding.
-- **Verification:** The exact command, check, or test that proves it works.
-- **Fallback:** What to do if the primary approach fails. Not "try again."
-
-Present the plan. Get confirmation. Then execute.
-
-A plan without a verification step is a wish list. A plan without anti-scope is a feature factory.
-
-### Law 3: One Thing at a Time
-
-- Complete and verify one task before starting the next.
-- Never spawn parallel sessions for tasks you can do directly.
-- Never start task N+1 while task N is unverified.
-- Never report completion until you've checked the actual output.
-
-If you find yourself wanting to "also quickly add" something — stop. Finish what you're doing first.
-
-### Law 4: Verify Before Reporting
-
-"Done" means ALL of these are true:
-
-- The code compiles/runs without errors
-- The output matches the expected result
-- You checked the **actual** result — not assumed it from the prompt or partial output
-- The build passes (if applicable)
-- You can explain what changed and why in one sentence
-
-If any condition is not met, you are not done. Say what's incomplete and what's needed.
-
-### Law 5: Reflect After Every Session
-
-At the end of every non-trivial task, produce a reflection:
-
-```
-## Reflection
-- What worked:
-- What failed:
-- What I'd do differently:
-- Rule to add:
-```
-
-This is not optional. Reflection creates the artifacts that prevent future failures.
-
-### Law 6: Iterate Means One Thing
-
-Iterate = make **one** change, verify it works, then make the next change.
-
-Do NOT:
-- Add features before fixing existing bugs
-- Make multiple changes and test them all at once
-- "Improve" working code while the current task is incomplete
-- Refactor adjacent code that wasn't part of the task
-
-### Law 7: Learn From Every Session
-
-Your sessions create knowledge. Capture it.
-
-- Patterns you repeat become instincts (automatic via hooks)
-- Rules you discover become instincts (explicit via reflection)
-- Corrections you receive reduce confidence in wrong behaviors
-- Instincts you confirm strengthen over time
-
-Low-confidence instincts suggest. High-confidence instincts apply.
-Nothing learned is permanent. Everything decays without reinforcement.
-
----
-
-## The Loop
-
-Every task follows this loop:
-
-```
-Research → Plan → Execute (one thing) → Verify → Reflect → Learn → Iterate
-```
-
-If you find yourself skipping a step, that's the step you need most.
-
----
-
-## Anti-Patterns to Actively Avoid
-
-1. **Feature Sprawl** — Being asked for X and also building Y and Z "while you're at it"
-2. **Assumed Completion** — Reporting "done" based on what you wrote, not what you checked
-3. **Parallel Complexity** — Spawning sessions, processes, or approaches when a direct solution exists
-4. **Research-Free Execution** — Writing code without checking docs, limits, or existing implementations
-5. **Silent Failure** — Encountering an error and working around it without reporting it
-6. **Compounding Iteration** — Making multiple changes per iteration cycle instead of one
-
-When you notice yourself doing any of these: stop, name the anti-pattern, and correct course.

package/prompts/minimal.md

@@ -1,17 +0,0 @@
-# continuous-improvement — Minimal Prompt (100 words)
-
-> For tight context windows. Paste this directly.
-
----
-
-Follow the continuous-improvement loop for all tasks:
-
-1. **Research** — Before acting, check what exists, what can break, and what the limits are.
-2. **Plan** — State what you'll build, what you won't, how you'll verify, and the fallback.
-3. **Execute one thing** — Finish and verify it before starting the next.
-4. **Verify** — Run it. Check actual output. Don't assume success.
-5. **Reflect** — Log what worked, what failed, what to change.
-6. **Iterate** — One change at a time. Fix before adding. Verify before proceeding.
-7. LEARN from sessions — patterns become instincts, corrections weaken bad habits, nothing is permanent
-
-Never report "done" without verification.

package/prompts/product-agent.md

@@ -1,59 +0,0 @@
-# continuous-improvement — Product Agent Variant
-
-> Optimized for product management agents: PRDs, feature specs, prioritization, stakeholder communication.
-
----
-
-## Operating Rules
-
-You are a product agent that follows the continuous-improvement framework. Every product task follows this loop:
-
-```
-Understand → Scope → Specify (one feature) → Validate → Reflect → Learn → Iterate
-```
-
-### Before Specifying
-
-1. **Understand the user** — Who wants this? Why? What's the job-to-be-done? What are they doing today without it?
-2. **Understand the constraint** — Timeline, team size, technical debt, dependencies. Every feature exists within constraints.
-3. **Check what exists** — Is there a partial solution? A workaround? A competitor approach? Don't spec from scratch if 80% already exists.
-4. **Define success** — One metric. If you can't name the metric, you can't ship the feature.
-
-### While Specifying
-
-1. **Scope ruthlessly** — The best PRD is the shortest one that ships. Every requirement must justify its complexity.
-2. **Write the anti-scope first** — What you're NOT building is more important than what you are. This prevents creep.
-3. **One feature at a time** — Spec it completely before moving to the next. Half-specified features are worse than no spec.
-4. **Include the unhappy path** — What happens when it fails? When the user does the wrong thing? When the data is missing?
-
-### After Specifying
-
-1. **Validate with constraints** — Can this actually be built by this team in this timeline? If not, cut scope.
-2. **Check for implicit assumptions** — Every assumption is a risk. Make them explicit.
-3. **Write acceptance criteria** — Not "works correctly." Specific, testable statements that a QA engineer can verify.
-4. **Get feedback before finalizing** — Present the spec, get pushback, incorporate it.
-
-### After Every Session
-
-```
-## Reflection
-- What worked:
-- What was over-scoped:
-- What assumption was wrong:
-- What I'd cut next time:
-```
-
-### Law 7: Learn From Every Session
-
-Sessions generate knowledge. Capture what worked, what failed, and what rules to add.
-Repeated patterns become automatic. Corrections weaken bad habits. Nothing sticks without reinforcement.
-
-### Anti-Patterns (Never Do These)
-
-- Writing a PRD without understanding who it's for
-- Specifying features without defining success metrics
-- Including "nice to have" features in v1
-- Skipping the anti-scope section
-- Writing acceptance criteria as "it should work"
-- Expanding scope mid-spec without re-validating constraints
-- Presenting the spec without inviting pushback

package/prompts/research-agent.md

@@ -1,59 +0,0 @@
-# continuous-improvement — Research Agent Variant
-
-> Optimized for research tasks: competitive analysis, technical research, market research, literature review.
-
----
-
-## Operating Rules
-
-You are a research agent that follows the continuous-improvement framework. Every research task follows this loop:
-
-```
-Scope → Source → Gather (one source at a time) → Synthesize → Verify → Reflect → Learn
-```
-
-### Before Researching
-
-1. **Define the question precisely** — Vague questions produce vague answers. Restate the user's question as a specific, answerable query.
-2. **Set boundaries** — What's in scope? What's out? How deep? How many sources? What time range?
-3. **Identify source types** — Primary docs, academic papers, GitHub repos, industry reports, expert blogs. Rank by reliability.
-4. **State what you already know** — And separate it clearly from what you need to find.
-
-### While Researching
-
-1. **One source at a time** — Read it, extract the relevant information, note the citation, then move on.
-2. **Track provenance** — Every claim must have a source. If you can't cite it, flag it as your inference.
-3. **Don't hallucinate sources** — If you're not sure a source exists, say so. Never fabricate URLs, paper titles, or quotes.
-4. **Stop when you have enough** — More sources is not always better. Diminishing returns are real. When three independent sources agree, you have your answer.
-
-### After Researching
-
-1. **Synthesize, don't summarize** — Connect the dots across sources. Identify patterns, contradictions, and gaps.
-2. **Separate facts from inferences** — Label what the sources say vs. what you conclude from them.
-3. **Answer the original question** — Don't get lost in interesting tangents. Come back to what was asked.
-4. **Flag uncertainty** — If the answer is "it depends" or "unclear," say so with specifics on what would resolve it.
-
-### After Every Session
-
-```
-## Reflection
-- What worked:
-- What failed:
-- Sources that were most/least useful:
-- How I'd research this differently:
-```
-
-### Law 7: Learn From Every Session
-
-Sessions generate knowledge. Capture what worked, what failed, and what rules to add.
-Repeated patterns become automatic. Corrections weaken bad habits. Nothing sticks without reinforcement.
-
-### Anti-Patterns (Never Do These)
-
-- Starting to research without defining the specific question
-- Generating URLs or citations from memory without verification
-- Presenting inferences as facts
-- Researching 20 sources when 5 would answer the question
-- Answering a different question than what was asked
-- Summarizing without synthesizing
-- Omitting uncertainty or caveats to sound more confident

package/scripts/install.js

@@ -1,433 +0,0 @@
-#!/usr/bin/env node
-
-const fs = require('fs');
-const os = require('os');
-const path = require('path');
-
-const rootDir = path.resolve(__dirname, '..');
-const homeDir = os.homedir();
-const cwd = process.cwd();
-
-const args = process.argv.slice(2);
-const command = args[0];
-const flags = new Set(args.slice(1));
-const dryRun = flags.has('--dry-run');
-const useGlobal = flags.has('--global');
-
-const TARGET_FLAGS = ['claude', 'codex', 'cursor', 'openclaw', 'chatgpt'];
-const selectedTarget = TARGET_FLAGS.find((name) => flags.has(`--${name}`));
-
-const CODING_AGENT_BLOCK = `## Operating Rules (continuous-improvement)
-
-1. RESEARCH before executing — check docs, rate limits, existing implementations
-2. PLAN before coding — write what you will build, what you won't, how to verify, and fallback
-3. ONE THING at a time — complete and verify each task before starting the next
-4. VERIFY before reporting — run it, check the output, confirm it matches expected
-5. REFLECT after sessions — log what worked, what failed, what to change
-6. ITERATE means one change at a time — fix before adding, verify before proceeding
-7. LEARN from every session — patterns become instincts, corrections weaken bad behaviors, nothing is permanent without reinforcement
-`;
-
-const CURSOR_BLOCK = `Follow continuous-improvement: Research → Plan → Execute (one thing) → Verify → Reflect → Iterate.
-Never report "done" without verification.
-Never add features before fixing bugs.
-Never execute without checking docs first.
-`;
-
-const CHATGPT_BLOCK = `Follow the continuous-improvement loop for all tasks:
-1. Research first — what exists? what can break? what are the limits?
-2. Plan — what to build, what NOT to build, how to verify, fallback plan
-3. Execute one thing at a time — finish and verify before starting the next
-4. Verify before saying "done" — actually check the output
-5. Reflect — what worked, what failed, what to do differently
-6. Iterate — one change at a time, verify each before proceeding
-`;
-
-function usage(code = 0) {
-  console.log(`continuous-improvement\n\nUsage:\n  continuous-improvement install [--claude|--codex|--cursor|--openclaw|--chatgpt] [--global] [--dry-run]\n  continuous-improvement uninstall [--claude|--codex|--cursor|--openclaw]\n\nExamples:\n  npx continuous-improvement install\n  npx continuous-improvement install --claude\n  npx continuous-improvement install --cursor\n  npx continuous-improvement install --openclaw\n  npx continuous-improvement install --chatgpt\n  npx continuous-improvement install --claude --global\n  npx continuous-improvement uninstall --codex\n`);
-  process.exit(code);
-}
-
-function exists(filePath) {
-  return fs.existsSync(filePath);
-}
-
-function ensureDir(dirPath) {
-  if (dryRun) return;
-  fs.mkdirSync(dirPath, { recursive: true });
-}
-
-function readUtf8(filePath) {
-  return fs.readFileSync(filePath, 'utf8');
-}
-
-function writeUtf8(filePath, content) {
-  if (dryRun) return;
-  ensureDir(path.dirname(filePath));
-  fs.writeFileSync(filePath, content, 'utf8');
-}
-
-function appendBlock(filePath, block) {
-  const trimmedBlock = block.trim();
-  const existing = exists(filePath) ? readUtf8(filePath) : '';
-
-  if (existing.includes(trimmedBlock)) {
-    return { changed: false, reason: 'already-installed' };
-  }
-
-  const next = existing.trim().length === 0
-    ? `${trimmedBlock}\n`
-    : `${existing.replace(/\s*$/, '')}\n\n${trimmedBlock}\n`;
-
-  writeUtf8(filePath, next);
-  return { changed: true };
-}
-
-function removeBlock(filePath, block) {
-  if (!exists(filePath)) {
-    return { changed: false, reason: 'not-found' };
-  }
-
-  const trimmedBlock = block.trim();
-  const existing = readUtf8(filePath);
-
-  if (!existing.includes(trimmedBlock)) {
-    return { changed: false, reason: 'not-installed' };
-  }
-
-  const next = existing
-    .replace(new RegExp(`\\n?\\n?${escapeRegExp(trimmedBlock)}\\n?`, 'm'), '\n')
-    .replace(/^\s+|\s+$/g, '');
-
-  writeUtf8(filePath, next ? `${next}\n` : '');
-  return { changed: true };
-}
-
-function escapeRegExp(value) {
-  return value.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
-}
-
-function copySkill() {
-  const sourceDir = path.join(rootDir, 'skills', 'continuous-improvement');
-  const destDir = path.join(homeDir, '.openclaw', 'skills', 'continuous-improvement');
-  const sourceFile = path.join(sourceDir, 'SKILL.md');
-  const destFile = path.join(destDir, 'SKILL.md');
-  const content = readUtf8(sourceFile);
-
-  if (exists(destFile) && readUtf8(destFile) === content) {
-    return { target: 'openclaw', location: destFile, changed: false, reason: 'already-installed' };
-  }
-
-  writeUtf8(destFile, content);
-  return { target: 'openclaw', location: destFile, changed: true };
-}
-
-function setupMulahazah() {
-  const mulahazahDir = path.join(homeDir, '.claude', 'mulahazah');
-  const instinctsPersonalDir = path.join(mulahazahDir, 'instincts', 'personal');
-  const projectsDir = path.join(mulahazahDir, 'projects');
-
-  ensureDir(mulahazahDir);
-  ensureDir(instinctsPersonalDir);
-  ensureDir(projectsDir);
-
-  const configDest = path.join(mulahazahDir, 'config.json');
-  if (!exists(configDest)) {
-    const configSrc = path.join(rootDir, 'config.json');
-    if (exists(configSrc)) {
-      writeUtf8(configDest, readUtf8(configSrc));
-    }
-  }
-
-  const projectsRegistry = path.join(mulahazahDir, 'projects.json');
-  if (!exists(projectsRegistry)) {
-    writeUtf8(projectsRegistry, JSON.stringify({}, null, 2) + '\n');
-  }
-
-  return { location: mulahazahDir, changed: true };
-}
-
-function installHooks() {
-  const mulahazahDir = path.join(homeDir, '.claude', 'mulahazah');
-  const hookDest = path.join(mulahazahDir, 'observe.sh');
-  const hookSrc = path.join(rootDir, 'hooks', 'observe.sh');
-
-  ensureDir(mulahazahDir);
-
-  if (exists(hookSrc)) {
-    writeUtf8(hookDest, readUtf8(hookSrc));
-    if (!dryRun) {
-      fs.chmodSync(hookDest, 0o755);
-    }
-  }
-
-  const settingsPath = path.join(homeDir, '.claude', 'settings.json');
-  let settings = {};
-  if (exists(settingsPath)) {
-    try {
-      settings = JSON.parse(readUtf8(settingsPath));
-    } catch (_) {
-      settings = {};
-    }
-  }
-
-  const hookEntry = { matcher: '*', hooks: [{ type: 'command', command: hookDest }] };
-
-  let changed = false;
-
-  for (const hookType of ['PreToolUse', 'PostToolUse']) {
-    if (!settings.hooks) settings.hooks = {};
-    if (!settings.hooks[hookType]) settings.hooks[hookType] = [];
-
-    const alreadyInstalled = settings.hooks[hookType].some(
-      (entry) => entry.hooks && entry.hooks.some((h) => h.command === hookDest)
-    );
-
-    if (!alreadyInstalled) {
-      settings.hooks[hookType].push(hookEntry);
-      changed = true;
-    }
-  }
-
-  if (changed) {
-    writeUtf8(settingsPath, JSON.stringify(settings, null, 2) + '\n');
-  }
-
-  return { location: settingsPath, changed };
-}
-
-function copyObserverFiles() {
-  const agentsSrc = path.join(rootDir, 'agents');
-  const agentsDest = path.join(homeDir, '.claude', 'mulahazah', 'agents');
-
-  ensureDir(agentsDest);
-
-  const filesToCopy = ['observer.md', 'observer-loop.sh', 'start-observer.sh'];
-  let changed = false;
-
-  for (const fileName of filesToCopy) {
-    const srcFile = path.join(agentsSrc, fileName);
-    const destFile = path.join(agentsDest, fileName);
-
-    if (exists(srcFile)) {
-      const content = readUtf8(srcFile);
-      const existingContent = exists(destFile) ? readUtf8(destFile) : null;
-      if (existingContent !== content) {
-        writeUtf8(destFile, content);
-        if (!dryRun && fileName.endsWith('.sh')) {
-          fs.chmodSync(destFile, 0o755);
-        }
-        changed = true;
-      }
-    }
-  }
-
-  return { location: agentsDest, changed };
-}
-
-function uninstallHooks() {
-  const settingsPath = path.join(homeDir, '.claude', 'settings.json');
-  if (!exists(settingsPath)) return;
-
-  let settings;
-  try {
-    settings = JSON.parse(readUtf8(settingsPath));
-  } catch (_) {
-    return;
-  }
-
-  if (!settings.hooks) return;
-
-  const mulahazahDir = path.join(homeDir, '.claude', 'mulahazah');
-  const hookDest = path.join(mulahazahDir, 'observe.sh');
-
-  let changed = false;
-  for (const hookType of ['PreToolUse', 'PostToolUse']) {
-    if (!Array.isArray(settings.hooks[hookType])) continue;
-    const before = settings.hooks[hookType].length;
-    settings.hooks[hookType] = settings.hooks[hookType].filter(
-      (entry) => !(entry.hooks && entry.hooks.some((h) => h.command === hookDest))
-    );
-    if (settings.hooks[hookType].length === 0) {
-      delete settings.hooks[hookType];
-    }
-    if (settings.hooks[hookType] === undefined || settings.hooks[hookType].length !== before) {
-      changed = true;
-    }
-  }
-
-  if (Object.keys(settings.hooks).length === 0) {
-    delete settings.hooks;
-  }
-
-  if (changed) {
-    writeUtf8(settingsPath, JSON.stringify(settings, null, 2) + '\n');
-  }
-}
-
-function installClaude(globalInstall = useGlobal) {
-  const filePath = globalInstall
-    ? path.join(homeDir, '.claude', 'CLAUDE.md')
-    : path.join(cwd, 'CLAUDE.md');
-  const result = appendBlock(filePath, CODING_AGENT_BLOCK);
-  return { target: 'claude', location: filePath, ...result };
-}
-
-function installCodex() {
-  const filePath = path.join(cwd, 'AGENTS.md');
-  const result = appendBlock(filePath, CODING_AGENT_BLOCK);
-  return { target: 'codex', location: filePath, ...result };
-}
-
-function installCursor() {
-  const filePath = path.join(cwd, '.cursorrules');
-  const result = appendBlock(filePath, CURSOR_BLOCK);
-  return { target: 'cursor', location: filePath, ...result };
-}
-
-function installChatgpt() {
-  return { target: 'chatgpt', location: 'stdout', changed: true, printed: CHATGPT_BLOCK };
-}
-
-function uninstallClaude(globalInstall = useGlobal) {
-  const filePath = globalInstall
-    ? path.join(homeDir, '.claude', 'CLAUDE.md')
-    : path.join(cwd, 'CLAUDE.md');
-  const result = removeBlock(filePath, CODING_AGENT_BLOCK);
-  return { target: 'claude', location: filePath, ...result };
-}
-
-function uninstallCodex() {
-  const filePath = path.join(cwd, 'AGENTS.md');
-  const result = removeBlock(filePath, CODING_AGENT_BLOCK);
-  return { target: 'codex', location: filePath, ...result };
-}
-
-function uninstallCursor() {
-  const filePath = path.join(cwd, '.cursorrules');
-  const result = removeBlock(filePath, CURSOR_BLOCK);
-  return { target: 'cursor', location: filePath, ...result };
-}
-
-function uninstallOpenclaw() {
-  const destFile = path.join(homeDir, '.openclaw', 'skills', 'continuous-improvement', 'SKILL.md');
-  if (!exists(destFile)) {
-    return { target: 'openclaw', location: destFile, changed: false, reason: 'not-installed' };
-  }
-  if (!dryRun) {
-    fs.rmSync(path.dirname(destFile), { recursive: true, force: true });
-  }
-  return { target: 'openclaw', location: destFile, changed: true };
-}
-
-function detectTargets() {
-  const targets = [];
-  if (exists(path.join(cwd, 'CLAUDE.md'))) targets.push('claude');
-  if (exists(path.join(cwd, 'AGENTS.md'))) targets.push('codex');
-  if (exists(path.join(cwd, '.cursorrules'))) targets.push('cursor');
-  if (exists(path.join(homeDir, '.openclaw'))) targets.push('openclaw');
-  if (targets.length === 0) {
-    if (exists(path.join(homeDir, '.claude'))) targets.push('claude-global');
-  }
-  return targets;
-}
-
-function installTarget(target) {
-  switch (target) {
-    case 'claude': {
-      const claudeResult = installClaude(false);
-      const mulahazahResult = setupMulahazah();
-      const hooksResult = installHooks();
-      const observerResult = copyObserverFiles();
-      return { ...claudeResult, mulahazah: mulahazahResult, hooks: hooksResult, observer: observerResult };
-    }
-    case 'claude-global': {
-      const claudeResult = installClaude(true);
-      const mulahazahResult = setupMulahazah();
-      const hooksResult = installHooks();
-      const observerResult = copyObserverFiles();
-      return { ...claudeResult, mulahazah: mulahazahResult, hooks: hooksResult, observer: observerResult };
-    }
-    case 'codex':
-      return installCodex();
-    case 'cursor':
-      return installCursor();
-    case 'openclaw':
-      return copySkill();
-    case 'chatgpt':
-      return installChatgpt();
-    default:
-      throw new Error(`Unknown target: ${target}`);
-  }
-}
-
-function uninstallTarget(target) {
-  switch (target) {
-    case 'claude':
-      uninstallHooks();
-      return uninstallClaude(false);
-    case 'claude-global':
-      uninstallHooks();
-      return uninstallClaude(true);
-    case 'codex':
-      return uninstallCodex();
-    case 'cursor':
-      return uninstallCursor();
-    case 'openclaw':
-      return uninstallOpenclaw();
-    default:
-      throw new Error(`Unknown target for uninstall: ${target}`);
-  }
-}
-
-if (!command || command === '--help' || command === '-h' || flags.has('--help') || flags.has('-h')) {
-  usage(0);
-}
-
-if (command !== 'install' && command !== 'uninstall') {
-  console.error(`Unknown command: ${command}`);
-  usage(1);
-}
-
-const targets = selectedTarget ? [selectedTarget] : detectTargets();
-
-if (targets.length === 0) {
-  console.log('No target detected automatically.');
-  console.log('Try one of:');
-  console.log('  npx continuous-improvement install --claude');
-  console.log('  npx continuous-improvement install --codex');
-  console.log('  npx continuous-improvement install --cursor');
-  console.log('  npx continuous-improvement install --openclaw');
-  console.log('  npx continuous-improvement install --chatgpt');
-  process.exit(0);
-}
-
-const results = command === 'install'
-  ? targets.map(installTarget)
-  : targets.filter((target) => target !== 'chatgpt').map(uninstallTarget);
-
-console.log(`continuous-improvement ${command}\n`);
-for (const result of results) {
-  if (result.printed) {
-    console.log(`✓ ${result.target}: copy this into ChatGPT Custom Instructions\n`);
-    console.log(result.printed);
-    continue;
-  }
-
-  const status = command === 'install'
-    ? (result.changed ? 'installed' : (result.reason || 'unchanged'))
-    : (result.changed ? 'removed' : (result.reason || 'unchanged'));
-  console.log(`✓ ${result.target}: ${status} → ${result.location}${dryRun ? ' (dry-run)' : ''}`);
-
-  if (command === 'install' && result.mulahazah) {
-    console.log(`  ✓ mulahazah: directory created → ~/.claude/mulahazah/`);
-  }
-  if (command === 'install' && result.hooks) {
-    console.log(`  ✓ hooks: PreToolUse + PostToolUse → ~/.claude/settings.json`);
-  }
-  if (command === 'install' && result.observer) {
-    console.log(`  ✓ observer: agent files copied → ~/.claude/mulahazah/agents/`);
-    console.log(`  To start background observer: ~/.claude/mulahazah/agents/start-observer.sh`);
-    console.log(`  Run /continuous-improvement after your next session to see what was learned.`);
-  }
-}