feat: Implement Vitest unit tests across agent modules and introduce …

.github/workflows/test.yml

@@ -0,0 +1,26 @@
+name: Tests
+
+on:
+  pull_request:
+    branches: [main]
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-node@v4
+        with:
+          node-version: '22'
+          cache: 'npm'
+          cache-dependency-path: droid-agent/app/package-lock.json
+
+      - name: Install dependencies
+        working-directory: droid-agent/app
+        run: npm ci
+
+      - name: Run tests
+        working-directory: droid-agent/app
+        run: npm test

droid-agent/AGENT.md

@@ -0,0 +1,43 @@
+# AGENT.md — Droid Agent Knowledge Index
+
+> This file is the master index of everything the agent knows. It is loaded before every request so the agent can quickly decide where to look and what context is available. Every update to memory, skills, incidents, infra, or learned patterns must update this file.
+
+---
+
+## Skills (7 files)
+
+Domain knowledge and runbooks the agent uses during investigations.
+
+| File | Size | First line |
+|------|------|------------|
+| `skills/docker.md` | 3706b | ```bash |
+| `skills/error-rate.md` | 2591b | When error rates spike, work through this in order: |
+| `skills/general-debugging.md` | 5554b | When something breaks, answer these questions in order: |
+| `skills/infra-sync.md` | 2157b | Discover everything about the developer's local and cloud infrastructure. Write  |
+| `skills/kubernetes-incidents.md` | 5085b | Run these immediately when something is wrong in a k8s cluster: |
+| `skills/kubernetes.md` | 4165b | ```bash |
+| `skills/performance.md` | 3781b | When latency spikes or performance degrades: |
+
+## Memory
+
+### context.md
+<!-- Edit this file with permanent facts about your infrastructure -->
+
+### README.md
+Persistent knowledge the agent reads on every message. Survives container restar
+
+---
+
+## How to use this index
+
+- **Before investigating**: Check `infra/summary.md` and `context.md` for known infrastructure topology
+- **For k8s issues**: Read `skills/kubernetes-incidents.md` + `infra/kubernetes.md`
+- **For Docker issues**: Read `skills/docker.md` + `infra/docker.md`
+- **For performance**: Read `skills/performance.md`
+- **For error spikes**: Read `skills/error-rate.md`
+- **For past incidents**: Check `incidents/` for similar issues and `learned/patterns.md` for known investigation flows
+- **For general triage**: Read `skills/general-debugging.md`
+
+---
+
+*Last updated: 2026-03-18T12:24:15.792Z*

droid-agent/app/agent-md.js

@@ -0,0 +1,92 @@
+import { readFile, writeFile } from 'fs/promises';
+import { readAllMemory } from './memory.js';
+import { loadSkills } from './skills.js';
+
+const AGENT_MD_PATH = '/app/AGENT.md';
+
+export async function regenerateAgentMd() {
+  const skills = await loadSkills();
+  const memoryFiles = await readAllMemory();
+
+  // Group memory files by directory
+  const groups = {};
+  for (const f of memoryFiles) {
+    const slashIdx = f.path.indexOf('/');
+    const dir = slashIdx !== -1 ? f.path.substring(0, slashIdx) : '/';
+    if (!groups[dir]) groups[dir] = [];
+    groups[dir].push(f);
+  }
+
+  let md = `# AGENT.md — Droid Agent Knowledge Index
+
+> This file is the master index of everything the agent knows. It is loaded before every request so the agent can quickly decide where to look and what context is available. Every update to memory, skills, incidents, infra, or learned patterns must update this file.
+
+---
+
+## Skills (${skills.length} files)
+
+Domain knowledge and runbooks the agent uses during investigations.
+
+| File | Size | First line |
+|------|------|------------|
+`;
+
+  for (const skill of skills) {
+    const firstLine = (skill.content.split('\n').find(l => l.trim() && !l.startsWith('#')) || '').substring(0, 80);
+    md += `| \`skills/${skill.name}.md\` | ${skill.size}b | ${firstLine} |\n`;
+  }
+
+  md += `\n## Memory\n\n`;
+
+  // Root files
+  if (groups['/']) {
+    for (const f of groups['/']) {
+      const firstLine = (f.content.split('\n').find(l => l.trim() && !l.startsWith('#')) || '').substring(0, 80);
+      md += `### ${f.path}\n${firstLine}\n\n`;
+    }
+  }
+
+  // Directory groups
+  const dirOrder = Object.keys(groups).filter(d => d !== '/').sort();
+  for (const dir of dirOrder) {
+    const files = groups[dir];
+    md += `### ${dir}/ (${files.length} files)\n\n`;
+    md += `| File | Size | Summary |\n|------|------|---------|\n`;
+    for (const f of files) {
+      const fileName = f.path.substring(dir.length + 1);
+      // Extract first meaningful line or heading
+      const lines = f.content.split('\n');
+      let summary = '';
+      for (const line of lines) {
+        const trimmed = line.trim();
+        if (trimmed && !trimmed.startsWith('#') && !trimmed.startsWith('>') && !trimmed.startsWith('---') && !trimmed.startsWith('|')) {
+          summary = trimmed.substring(0, 100);
+          break;
+        }
+      }
+      md += `| \`${dir}/${fileName}\` | ${f.size}b | ${summary} |\n`;
+    }
+    md += `\n`;
+  }
+
+  md += `---
+
+## How to use this index
+
+- **Before investigating**: Check \`infra/summary.md\` and \`context.md\` for known infrastructure topology
+- **For k8s issues**: Read \`skills/kubernetes-incidents.md\` + \`infra/kubernetes.md\`
+- **For Docker issues**: Read \`skills/docker.md\` + \`infra/docker.md\`
+- **For performance**: Read \`skills/performance.md\`
+- **For error spikes**: Read \`skills/error-rate.md\`
+- **For past incidents**: Check \`incidents/\` for similar issues and \`learned/patterns.md\` for known investigation flows
+- **For general triage**: Read \`skills/general-debugging.md\`
+
+---
+
+*Last updated: ${new Date().toISOString()}*
+`;
+
+  await writeFile(AGENT_MD_PATH, md, 'utf-8');
+  console.log(`[agent-md] Regenerated AGENT.md (${md.length} chars, ${skills.length} skills, ${memoryFiles.length} memory files)`);
+  return md;
+}

droid-agent/app/agent.js

@@ -1,9 +1,21 @@
+import { readFile } from 'fs/promises';
 import { loadSkills, loadSkill } from './skills.js';
 import { readAllMemory, readMemory, writeMemory } from './memory.js';
 import { loadTools, loadMcpTools, executeTool } from './tools.js';
 import { getConversation, saveConversation } from './redis.js';
 import { logToolExecution, logIncident, upsertConversation, saveMessage, getRecentFeedback } from './db.js';
 import { createCompletion, buildImageContent, getProviderName, getModelId } from './provider.js';
+import { regenerateAgentMd } from './agent-md.js';
+
+const AGENT_MD_PATH = '/app/AGENT.md';
+
+async function loadAgentMd() {
+  try {
+    return await readFile(AGENT_MD_PATH, 'utf-8');
+  } catch {
+    return null;
+  }
+}
 
 const MAX_TOOL_ITERATIONS = 10;
 const MAX_SYNC_ITERATIONS = 40;
@@ -25,6 +37,15 @@ function log(emitter, type, data) {
 }
 
 async function buildSystemPrompt(emitter) {
+  // Load AGENT.md first — the master knowledge index
+  log(emitter, 'thinking', 'Loading AGENT.md knowledge index...');
+  const agentMd = await loadAgentMd();
+  if (agentMd) {
+    log(emitter, 'thinking', `AGENT.md loaded (${agentMd.length} chars)`);
+  } else {
+    log(emitter, 'thinking', 'AGENT.md not found — will operate without knowledge index');
+  }
+
   log(emitter, 'thinking', 'Loading skills...');
   const skills = await loadSkills();
   log(emitter, 'thinking', `Loaded ${skills.length} skills: ${skills.map(s => s.name).join(', ')}`);
@@ -47,6 +68,24 @@ CRITICAL RULES:
 
 `;
 
+  // AGENT.md is the first thing the agent sees — the master knowledge index
+  if (agentMd) {
+    prompt += `<agent_knowledge_index>
+The following is your master knowledge index (AGENT.md). READ THIS FIRST before every request. It tells you exactly what skills, memory, infrastructure data, incident history, and learned patterns are available and where to find them. Use this to plan your approach before taking action.
+
+${agentMd}
+</agent_knowledge_index>
+
+IMPORTANT: Before responding to any user request, consult the knowledge index above to:
+1. Identify which skills are relevant to the issue
+2. Check if similar incidents have been investigated before
+3. Look up known infrastructure topology and configurations
+4. Review learned patterns for proven investigation sequences
+Then form an action plan and execute it.
+
+`;}
+
+
   if (skills.length > 0) {
     prompt += '<skills>\n';
     for (const skill of skills) {
@@ -297,6 +336,16 @@ export async function runAgent({ message, images, conversationId, emitter }) {
         }
       }
 
+      // Regenerate AGENT.md if any memory was written
+      if (memoryWrites.length > 0) {
+        try {
+          await regenerateAgentMd();
+          log(emitter, 'thinking', 'AGENT.md updated with new knowledge');
+        } catch (err) {
+          console.error('[agent-md] Regeneration failed:', err.message);
+        }
+      }
+
       let finalResponse = stripToolCalls(assistantText);
       finalResponse = stripMemoryWrites(finalResponse);
 
@@ -493,6 +542,16 @@ ${getToolInstructions()}`;
     log(emitter, 'thinking', `Fed ${results.length} tool result(s) back to model, looping...`);
   }
 
+  // Regenerate AGENT.md after sync
+  if (filesWritten.length > 0) {
+    try {
+      await regenerateAgentMd();
+      log(emitter, 'thinking', 'AGENT.md updated after sync');
+    } catch (err) {
+      console.error('[agent-md] Regeneration failed:', err.message);
+    }
+  }
+
   log(emitter, 'done', { message: `Sync complete: ${iterations} iterations, ${filesWritten.length} files written`, filesWritten });
   return { filesWritten };
 }
@@ -502,3 +561,6 @@ function extractField(content, fieldName) {
   const match = content.match(regex);
   return match ? match[1].trim() : null;
 }
+
+// Exported for testing
+export { parseToolCalls, parseMemoryWrites, stripToolCalls, stripMemoryWrites, extractField, tryParseJson, isValidJson };

droid-agent/app/learner.js

@@ -2,6 +2,7 @@ import { getUnanalyzedConversations, saveLearnerRun, getLastLearnerRun } from '.
 import { readMemory, writeMemory } from './memory.js';
 import { loadSkills } from './skills.js';
 import { createCompletion } from './provider.js';
+import { regenerateAgentMd } from './agent-md.js';
 
 const LEARNER_ENABLED = (process.env.LEARNER_ENABLED || 'true').toLowerCase() === 'true';
 const LEARNER_INTERVAL_MS = parseInt(process.env.LEARNER_INTERVAL_MS) || 3600000; // 1 hour
@@ -127,6 +128,16 @@ Your job is to produce FOUR sections separated by ===SPLIT===:
       filesWritten
     });
 
+    // Regenerate AGENT.md after learner writes
+    if (filesWritten.length > 0) {
+      try {
+        await regenerateAgentMd();
+        console.log('[learner] AGENT.md updated');
+      } catch (err) {
+        console.error('[agent-md] Regeneration failed:', err.message);
+      }
+    }
+
     console.log(`[learner] Cycle complete: analyzed ${conversations.length} conversations, wrote ${filesWritten.length} files`);
     return { conversationsAnalyzed: conversations.length, filesWritten };
   } catch (err) {