enhance-orchestrator

enhance-orchestrator

Coordinate all enhancement analyzers in parallel and produce a unified report.

Critical Rules

1. MUST run enhancers in parallel - Use Promise.all for efficiency
2. MUST only run enhancers for existing content - Skip if no files found
3. MUST report HIGH certainty first - Priority order: HIGH → MEDIUM → LOW
4. NEVER auto-fix without --apply flag - Explicit consent required
5. NEVER auto-fix MEDIUM or LOW issues - Only HIGH certainty

Workflow

Phase 1: Parse Arguments

const args = '$ARGUMENTS'.split(' ').filter(Boolean);
const targetPath = args.find(a => !a.startsWith('--')) || '.';

const flags = {
  apply: args.includes('--apply'),
  focus: args.find(a => a.startsWith('--focus='))?.split('=')[1],
  verbose: args.includes('--verbose'),
  showSuppressed: args.includes('--show-suppressed'),
  resetLearned: args.includes('--reset-learned'),
  noLearn: args.includes('--no-learn'),
  exportLearned: args.includes('--export-learned')
};

// Validate focus type
const VALID_FOCUS = ['plugin', 'agent', 'claudemd', 'claude-memory', 'docs', 'prompt', 'hooks', 'skills', 'cross-file'];
if (flags.focus && !VALID_FOCUS.includes(flags.focus)) {
  console.error(`Invalid --focus: "${flags.focus}". Valid: ${VALID_FOCUS.join(', ')}`);
  return;
}

Phase 2: Discovery

Detect what exists in target path:

const discovery = {
  plugins: await Glob({ pattern: 'plugins/*/.claude-plugin/plugin.json', path: targetPath }),
  agents: await Glob({ pattern: '**/agents/*.md', path: targetPath }),
  claudemd: await Glob({ pattern: '**/CLAUDE.md', path: targetPath }) ||
            await Glob({ pattern: '**/AGENTS.md', path: targetPath }),
  docs: await Glob({ pattern: 'docs/**/*.md', path: targetPath }),
  prompts: await Glob({ pattern: '**/prompts/**/*.md', path: targetPath }) ||
           await Glob({ pattern: '**/commands/**/*.md', path: targetPath }),
  hooks: await Glob({ pattern: '**/hooks/**/*.md', path: targetPath }),
  skills: await Glob({ pattern: '**/skills/**/SKILL.md', path: targetPath }),
  // Cross-file runs if agents OR skills exist (analyzes relationships)
  'cross-file': discovery.agents?.length || discovery.skills?.length ? ['enabled'] : []
};

Phase 3: Load Suppressions

// Use relative path from skill directory to plugin lib
// Path: skills/enhance-orchestrator/ -> ../../lib/
const { getSuppressionPath } = require('../../lib/cross-platform');
const { loadAutoSuppressions, getProjectId, clearAutoSuppressions } = require('../../lib/enhance/auto-suppression');

const suppressionPath = getSuppressionPath();
const projectId = getProjectId(targetPath);

if (flags.resetLearned) {
  clearAutoSuppressions(suppressionPath, projectId);
  console.log(`Cleared suppressions for project: ${projectId}`);
}

const autoLearned = loadAutoSuppressions(suppressionPath, projectId);

Phase 4: Launch Enhancers in Parallel

CRITICAL: MUST spawn these EXACT agents using Task(). Do NOT use Explore or other agents.

Focus Type	Agent to Spawn	Model	JS Analyzer
plugin	enhance:plugin-enhancer	sonnet	lib/enhance/plugin-analyzer.js
agent	enhance:agent-enhancer	opus	lib/enhance/agent-analyzer.js
claudemd	enhance:claudemd-enhancer	opus	lib/enhance/projectmemory-analyzer.js
docs	enhance:docs-enhancer	opus	lib/enhance/docs-analyzer.js
prompt	enhance:prompt-enhancer	opus	lib/enhance/prompt-analyzer.js
hooks	enhance:hooks-enhancer	opus	lib/enhance/hook-analyzer.js
skills	enhance:skills-enhancer	opus	lib/enhance/skill-analyzer.js
cross-file	enhance:cross-file-enhancer	sonnet	lib/enhance/cross-file-analyzer.js

Each agent has Bash(node:*) to run its JS analyzer. Do NOT substitute with Explore agents.

// EXACT agent mapping - do not change
const ENHANCER_AGENTS = {
  plugin: 'enhance:plugin-enhancer',
  agent: 'enhance:agent-enhancer',
  claudemd: 'enhance:claudemd-enhancer',
  docs: 'enhance:docs-enhancer',
  prompt: 'enhance:prompt-enhancer',
  hooks: 'enhance:hooks-enhancer',
  skills: 'enhance:skills-enhancer',
  'cross-file': 'enhance:cross-file-enhancer'
};

const promises = [];

// Pre-fetch repo-intel data for the enhancer agents.
//
// What each query gives the enhancers and how it shapes their suggestions:
//
//   doc-drift  - doc files with low code coupling. The docs-enhancer
//                de-prioritises edits to docs that aren't "alive" (low
//                coupling) and prioritises the ones that are - those
//                are the ones users actually read against the code.
//   stale-docs - symbol-level: the doc references a function/class
//                that no longer exists in code. Highest-precision
//                drift signal; trumps the heuristic doc-drift score.
//   conventions - commit-message style + naming patterns. Passed to
//                 agent-prompts, skills, and prompts enhancers so
//                 their generated language matches the project voice
//                 (e.g. terse vs. verbose, "feat:" vs "Add ", etc.).
let docDriftContext = '';
let conventionsContext = '';
try {
  const { binary } = require(`${pluginRoot}/lib/agentsys`).get();
  const { libRoot } = require(`${pluginRoot}/lib/agentsys`).get(); const { getStateDirPath } = require(`${libRoot}/platform/state-dir`);
  const fs = require('fs');
  const cwd = process.cwd();
  const mapFile = require('path').join(getStateDirPath(cwd), 'repo-intel.json');
  const q = (args) => { try { return JSON.parse(binary.runAnalyzer(args)); } catch { return null; } };

  if (fs.existsSync(mapFile)) {
    const docDrift = q(['repo-intel', 'query', 'doc-drift', '--top', '20', '--map-file', mapFile, cwd]);
    const staleDocs = q(['repo-intel', 'query', 'stale-docs', '--top', '20', '--map-file', mapFile, cwd]);
    const conventions = q(['repo-intel', 'query', 'conventions', '--map-file', mapFile, cwd]);

    if (docDrift && docDrift.length > 0) {
      const stale = docDrift.filter(d => d.codeCoupling === 0).map(d => d.path);
      if (stale.length > 0) {
        docDriftContext = '\nDoc-drift (docs with zero code coupling, likely stale): ' + stale.join(', ');
      }
    }
    if (staleDocs && staleDocs.length > 0) {
      docDriftContext += '\nStale doc refs (symbol-level):\n' +
        staleDocs.slice(0, 10).map(s => `  ${s.doc}:${s.line} "${s.reference}" [${s.issue}]`).join('\n');
    }
    if (conventions) {
      conventionsContext = '\nProject conventions: ' + JSON.stringify(conventions);
    }
  }
} catch (e) {
  console.error(`[INFO] repo-intel context skipped: ${e.message}`);
}

for (const [type, agentType] of Object.entries(ENHANCER_AGENTS)) {
  if (focus && focus !== type) continue;
  if (!discovery[type]?.length) continue;

  // Append relevant context per enhancer type
  let extraContext = '';
  if (type === 'docs') extraContext = docDriftContext;
  if (['agent-prompts', 'skills', 'prompts'].includes(type)) extraContext = conventionsContext;

  // MUST use exact subagent_type - these agents have Bash(node:*) to run JS analyzers
  promises.push(Task({
    subagent_type: agentType,
    prompt: `Analyze ${type} in ${targetPath}.
MUST use Skill tool to invoke your enhance-* skill.
The skill runs the JavaScript analyzer and returns structured findings.
verbose: ${flags.verbose}${extraContext}
Return JSON: { "enhancerType": "${type}", "findings": [...], "summary": { high, medium, low } }`
  }));
}

// MUST use Promise.all for parallel execution
const results = await Promise.all(promises);

Phase 5: Aggregate Results

function aggregateResults(enhancerResults) {
  const findings = [];
  const byEnhancer = {};

  for (const result of enhancerResults) {
    if (!result?.findings) continue;
    for (const finding of result.findings) {
      findings.push({ ...finding, source: result.enhancerType });
    }
    byEnhancer[result.enhancerType] = result.summary;
  }

  return {
    findings,
    byEnhancer,
    totals: {
      high: findings.filter(f => f.certainty === 'HIGH').length,
      medium: findings.filter(f => f.certainty === 'MEDIUM').length,
      low: findings.filter(f => f.certainty === 'LOW').length
    }
  };
}

Phase 6: Generate Report

Generate report directly from aggregated findings:

const { generateReport } = require('../../lib/enhance/reporter');

const report = generateReport(aggregated, {
  verbose: flags.verbose,
  showAutoFixable: flags.apply
});

console.log(report);

Phase 7: Auto-Learning

if (!flags.noLearn) {
  const { analyzeForAutoSuppression, saveAutoSuppressions } = require('../../lib/enhance/auto-suppression');

  const newSuppressions = analyzeForAutoSuppression(aggregated.findings, fileContents, { projectRoot: targetPath });

  if (newSuppressions.length > 0) {
    saveAutoSuppressions(suppressionPath, projectId, newSuppressions);
    console.log(`\nLearned ${newSuppressions.length} new suppressions.`);
  }
}

Phase 8: Apply Fixes

if (flags.apply) {
  const autoFixable = aggregated.findings.filter(f => f.certainty === 'HIGH' && f.autoFixable);

  if (autoFixable.length > 0) {
    console.log(`\n## Applying ${autoFixable.length} Auto-Fixes\n`);

    const byEnhancer = {};
    for (const fix of autoFixable) {
      const type = fix.source;
      if (!byEnhancer[type]) byEnhancer[type] = [];
      byEnhancer[type].push(fix);
    }

    for (const [type, fixes] of Object.entries(byEnhancer)) {
      await Task({
        subagent_type: enhancerAgents[type],
        prompt: `Apply HIGH certainty fixes: ${JSON.stringify(fixes, null, 2)}`
      });
    }

    console.log(`Applied ${autoFixable.length} fixes.`);
  }
}

Output Format

# Enhancement Analysis Report

**Target**: {targetPath}
**Date**: {timestamp}
**Enhancers Run**: {list}

## Executive Summary

| Enhancer | HIGH | MEDIUM | LOW | Auto-Fixable |
|----------|------|--------|-----|--------------|
| plugin   | 2    | 3      | 1   | 1            |
| agent    | 1    | 2      | 0   | 1            |
| **Total**| **3**| **5**  | **1**| **2**       |

## HIGH Certainty Issues
[Grouped by enhancer, then file]

## MEDIUM Certainty Issues
[...]

## Auto-Fix Summary
{n} issues can be fixed with `--apply` flag.

Constraints

• MUST run enhancers in parallel (Promise.all)
• MUST skip enhancers for missing content types
• MUST report HIGH certainty issues first
• MUST deduplicate findings across enhancers
• NEVER auto-fix without explicit --apply flag
• NEVER auto-fix MEDIUM or LOW certainty issues

Repo-Intel Data

Expected: the orchestrator (the command that spawned this agent) has already checked <stateDir>/repo-intel.json and either pre-fetched the data into your context or skipped (user declined to generate). Do not call AskUserQuestion here - subagents cannot interact with the user.

If the pre-fetched data is empty, proceed with the available context. The orchestrator has already made the decision on the user's behalf.

Binary: agent-analyzer auto-downloads to ~/.agent-sh/bin/ from agent-sh/agent-analyzer GitHub releases (~10 MB) on first use. The lib/agentsys resolver locates the agentsys install (CC marketplace clone, npm global, or sibling repo).