Sentinel: CodingAgent + Demo Pipeline E2E (Claude Code → LoRA training)

docs/SENTINEL-ARCHITECTURE.md

@@ -1,5 +1,7 @@
 # Sentinel Architecture: Universal Stream Processing for Autonomous Agents
 
+> **SUPERSEDED**: This is an early design document. The sentinel system has been fully re-architected as a Rust pipeline engine with 10 step types (Shell, LLM, Command, Condition, Loop, Parallel, Emit, Watch, Sentinel, CodingAgent), 103 tests, and full TypeScript lifecycle integration (persistence, triggers, escalation, event bridge). See **[src/docs/SENTINEL-ARCHITECTURE.md](src/docs/SENTINEL-ARCHITECTURE.md)** for the canonical, up-to-date documentation.
+
 ## Executive Summary
 
 Sentinels are **programmable stream observers** that watch any process, match patterns, classify output, and route events to persona inboxes. They solve the "last mile" problem preventing AI personas from effectively coding.

docs/SENTINEL-WORKERS.md

@@ -1,5 +1,7 @@
 # Sentinel Workers: Focused Agentic Loops
 
+> **SUPERSEDED**: This is an early conceptual document. The sentinel system has been fully implemented as a Rust pipeline engine with 10 step types and 103 tests. See **[src/docs/SENTINEL-ARCHITECTURE.md](src/docs/SENTINEL-ARCHITECTURE.md)** for the canonical architecture documentation, and **[src/docs/personas/ACADEMY-DOJO-ARCHITECTURE.md](src/docs/personas/ACADEMY-DOJO-ARCHITECTURE.md)** for the dual-sentinel teacher/student system.
+
 ## Vision
 
 Sentinels are **focused agents** that sit between dumb scripts and full citizen AIs. They have enough intelligence to interpret results, adjust approach, and know when to escalate - but they're laser-focused on a single goal.

src/browser/generated.ts

@@ -1,7 +1,7 @@
 /**
  * Browser Structure Registry - Auto-generated
  *
- * Contains 11 daemons and 218 commands and 2 adapters and 28 widgets.
+ * Contains 11 daemons and 221 commands and 2 adapters and 28 widgets.
  * Generated by scripts/generate-structure.ts - DO NOT EDIT MANUALLY
  */
 
@@ -129,9 +129,11 @@ import { GenomeAcademySessionBrowserCommand } from './../commands/genome/academy
 import { GenomeBatchMicroTuneBrowserCommand } from './../commands/genome/batch-micro-tune/browser/GenomeBatchMicroTuneBrowserCommand';
 import { GenomeDatasetPrepareBrowserCommand } from './../commands/genome/dataset-prepare/browser/GenomeDatasetPrepareBrowserCommand';
 import { GenomeDatasetSynthesizeBrowserCommand } from './../commands/genome/dataset-synthesize/browser/GenomeDatasetSynthesizeBrowserCommand';
+import { GenomeDemoRunBrowserCommand } from './../commands/genome/demo-run/browser/GenomeDemoRunBrowserCommand';
 import { GenomeJobCreateBrowserCommand } from './../commands/genome/job-create/browser/GenomeJobCreateBrowserCommand';
 import { GenomeJobStatusBrowserCommand } from './../commands/genome/job-status/browser/GenomeJobStatusBrowserCommand';
 import { GenomeTrainBrowserCommand } from './../commands/genome/train/browser/GenomeTrainBrowserCommand';
+import { GenomeTrainingExportBrowserCommand } from './../commands/genome/training-export/browser/GenomeTrainingExportBrowserCommand';
 import { GenomeTrainingPipelineBrowserCommand } from './../commands/genome/training-pipeline/browser/GenomeTrainingPipelineBrowserCommand';
 import { HelpBrowserCommand } from './../commands/help/browser/HelpBrowserCommand';
 import { IndicatorBrowserCommand } from './../commands/indicator/browser/IndicatorBrowserCommand';
@@ -181,6 +183,7 @@ import { PingBrowserCommand } from './../commands/ping/browser/PingBrowserComman
 import { PositronCursorBrowserCommand } from './../commands/positron/cursor/browser/PositronCursorBrowserCommand';
 import { ProcessRegistryBrowserCommand } from './../commands/process-registry/browser/ProcessRegistryBrowserCommand';
 import { RuntimeMetricsBrowserCommand } from './../commands/runtime/metrics/browser/RuntimeMetricsBrowserCommand';
+import { SentinelCodingAgentBrowserCommand } from './../commands/sentinel/coding-agent/browser/SentinelCodingAgentBrowserCommand';
 import { SentinelRunBrowserCommand } from './../commands/sentinel/run/browser/SentinelRunBrowserCommand';
 import { SentinelStatusBrowserCommand } from './../commands/sentinel/status/browser/SentinelStatusBrowserCommand';
 import { SessionCreateBrowserCommand } from './../commands/session/create/browser/SessionCreateBrowserCommand';
@@ -890,6 +893,11 @@ export const BROWSER_COMMANDS: CommandEntry[] = [
     className: 'GenomeDatasetSynthesizeBrowserCommand',
     commandClass: GenomeDatasetSynthesizeBrowserCommand
   },
+{
+    name: 'genome/demo-run',
+    className: 'GenomeDemoRunBrowserCommand',
+    commandClass: GenomeDemoRunBrowserCommand
+  },
 {
     name: 'genome/job-create',
     className: 'GenomeJobCreateBrowserCommand',
@@ -905,6 +913,11 @@ export const BROWSER_COMMANDS: CommandEntry[] = [
     className: 'GenomeTrainBrowserCommand',
     commandClass: GenomeTrainBrowserCommand
   },
+{
+    name: 'genome/training-export',
+    className: 'GenomeTrainingExportBrowserCommand',
+    commandClass: GenomeTrainingExportBrowserCommand
+  },
 {
     name: 'genome/training-pipeline',
     className: 'GenomeTrainingPipelineBrowserCommand',
@@ -1150,6 +1163,11 @@ export const BROWSER_COMMANDS: CommandEntry[] = [
     className: 'RuntimeMetricsBrowserCommand',
     commandClass: RuntimeMetricsBrowserCommand
   },
+{
+    name: 'sentinel/coding-agent',
+    className: 'SentinelCodingAgentBrowserCommand',
+    commandClass: SentinelCodingAgentBrowserCommand
+  },
 {
     name: 'sentinel/run',
     className: 'SentinelRunBrowserCommand',

src/commands/genome/demo-run/README.md

@@ -0,0 +1,72 @@
+# genome/demo-run
+
+Run a sentinel demo pipeline where Claude Code builds real software from a project spec, capturing all interactions for LoRA training.
+
+## Usage
+
+```bash
+# Run task-tracker demo (3 milestones, ~$3-7, ~20 minutes)
+./jtag genome/demo-run --project=task-tracker --personaId=<uuid>
+
+# With custom settings
+./jtag genome/demo-run \
+  --project=task-tracker \
+  --personaId=<uuid> \
+  --personaName="Helper AI" \
+  --maxRetries=3 \
+  --maxBudget=10 \
+  --maxTurns=40
+```
+
+## Parameters
+
+| Parameter | Required | Default | Description |
+|-----------|----------|---------|-------------|
+| project | Yes | - | Project name (maps to `src/projects/<name>/`) |
+| personaId | Yes | - | Target persona UUID for training |
+| personaName | No | 'demo-persona' | Persona display name |
+| baseModel | No | LOCAL_MODELS.DEFAULT | Base model for LoRA training |
+| maxRetries | No | 2 | Max CodingAgent retries per milestone |
+| maxBudget | No | 5.0 | Max USD per milestone |
+| maxTurns | No | 30 | Max CodingAgent turns per milestone |
+| provider | No | 'claude-code' | CodingAgent provider |
+| epochs | No | 3 | LoRA training epochs |
+| rank | No | 32 | LoRA rank |
+
+## Pipeline Flow
+
+```
+Step 0: Shell — create temp dir, copy scaffold, npm install
+Step 1: Loop (milestones):
+  loop.0: Shell — read milestone spec
+  loop.1: CodingAgent — Claude Code builds it (captureTraining=true)
+  loop.2: Shell — run deterministic tests
+  loop.3: Condition — passed?
+    then: emit milestone:passed
+    else: CodingAgent retry → re-run tests (up to maxRetries)
+  loop.4: Emit milestone:complete
+Step 2: Command — genome/train (captured interactions → LoRA adapter)
+Step 3: Command — genome/phenotype-validate (before vs after)
+Step 4: Emit — demo:complete
+```
+
+## Available Projects
+
+| Project | Milestones | Difficulty | Est. Cost |
+|---------|-----------|------------|-----------|
+| task-tracker | 3 | Intermediate | $3-7 |
+| ecommerce-api | 6 | Advanced | $10-20 |
+| url-shortener | 3 | Beginner | $2-5 |
+
+## Monitoring
+
+```bash
+# Check pipeline status
+./jtag sentinel/status --handle=<returned-handle>
+
+# Tail pipeline logs
+./jtag sentinel/logs/tail --handle=<returned-handle>
+
+# Check training data
+./jtag data/list --collection=genome_layers --limit=5
+```

src/commands/genome/demo-run/browser/GenomeDemoRunBrowserCommand.ts

@@ -0,0 +1,20 @@
+/**
+ * Genome Demo Run Command - Browser Implementation
+ *
+ * Delegates to server for demo pipeline orchestration.
+ */
+
+import { CommandBase, type ICommandDaemon } from '@daemons/command-daemon/shared/CommandBase';
+import type { JTAGContext } from '@system/core/types/JTAGTypes';
+import type { GenomeDemoRunParams, GenomeDemoRunResult } from '../shared/GenomeDemoRunTypes';
+
+export class GenomeDemoRunBrowserCommand extends CommandBase<GenomeDemoRunParams, GenomeDemoRunResult> {
+
+  constructor(context: JTAGContext, subpath: string, commander: ICommandDaemon) {
+    super('genome/demo-run', context, subpath, commander);
+  }
+
+  async execute(params: GenomeDemoRunParams): Promise<GenomeDemoRunResult> {
+    return await this.remoteExecute(params);
+  }
+}

src/commands/genome/demo-run/package.json

@@ -0,0 +1,27 @@
+{
+  "name": "@jtag-commands/genome/demo-run",
+  "version": "1.0.0",
+  "description": "Run a sentinel demo pipeline — Claude Code builds real software, interactions are captured for LoRA training",
+  "main": "server/GenomeDemoRunServerCommand.ts",
+  "types": "shared/GenomeDemoRunTypes.ts",
+  "scripts": {
+    "test": "echo 'Integration test: ./jtag genome/demo-run --project=task-tracker --personaId=<uuid>'",
+    "lint": "npx eslint **/*.ts",
+    "typecheck": "npx tsc --noEmit"
+  },
+  "peerDependencies": {
+    "@jtag/core": "*"
+  },
+  "files": [
+    "shared/**/*.ts",
+    "browser/**/*.ts",
+    "server/**/*.ts",
+    "README.md"
+  ],
+  "keywords": [
+    "jtag",
+    "command",
+    "genome/demo-run"
+  ],
+  "license": "MIT"
+}

src/commands/genome/demo-run/server/GenomeDemoRunServerCommand.ts

@@ -0,0 +1,156 @@
+/**
+ * Genome Demo Run Command - Server Implementation
+ *
+ * Loads a project spec, builds a demo pipeline via buildDemoPipeline(),
+ * and dispatches it to the Rust sentinel executor via sentinel/run.
+ *
+ * The pipeline has Claude Code build real software milestone by milestone,
+ * capturing every interaction for LoRA training. After all milestones:
+ * genome/train produces a LoRA adapter, genome/phenotype-validate proves improvement.
+ */
+
+import * as fs from 'fs';
+import * as path from 'path';
+import { CommandBase, type ICommandDaemon } from '@daemons/command-daemon/shared/CommandBase';
+import type { JTAGContext } from '@system/core/types/JTAGTypes';
+import { ValidationError } from '@system/core/types/ErrorTypes';
+import type { GenomeDemoRunParams, GenomeDemoRunResult } from '../shared/GenomeDemoRunTypes';
+import { createGenomeDemoRunResultFromParams } from '../shared/GenomeDemoRunTypes';
+import { Commands } from '@system/core/shared/Commands';
+import type { SentinelStep } from '@system/sentinel/SentinelDefinition';
+import type { PipelineSentinelParams, SentinelRunResult } from '@commands/sentinel/run/shared/SentinelRunTypes';
+import type { ProjectSpec } from '@system/genome/shared/AcademyTypes';
+import { buildDemoPipeline } from '@system/sentinel/pipelines/DemoPipeline';
+import { DEFAULT_DEMO_TRAINING_CONFIG, DEMO_DEFAULTS } from '@system/sentinel/pipelines/DemoTypes';
+import type { DemoPipelineConfig } from '@system/sentinel/pipelines/DemoTypes';
+import { LOCAL_MODELS } from '@system/shared/Constants';
+
+export class GenomeDemoRunServerCommand extends CommandBase<GenomeDemoRunParams, GenomeDemoRunResult> {
+
+  constructor(context: JTAGContext, subpath: string, commander: ICommandDaemon) {
+    super('genome/demo-run', context, subpath, commander);
+  }
+
+  async execute(params: GenomeDemoRunParams): Promise<GenomeDemoRunResult> {
+    const { project, personaId } = params;
+
+    if (!project) {
+      throw new ValidationError('project', 'Required. Project name (e.g., "task-tracker").');
+    }
+    if (!personaId) {
+      throw new ValidationError('personaId', 'Required. Target persona ID for training.');
+    }
+
+    // Resolve project directory
+    const srcRoot = path.resolve(__dirname, '../../../../');
+    const projectDir = path.join(srcRoot, 'projects', project);
+    const projectJsonPath = path.join(projectDir, 'project.json');
+
+    if (!fs.existsSync(projectJsonPath)) {
+      return createGenomeDemoRunResultFromParams(params, {
+        success: false,
+        handle: '',
+        projectName: project,
+        milestoneCount: 0,
+        error: `Project not found: ${projectJsonPath}. Available projects are in src/projects/.`,
+      });
+    }
+
+    // Parse project spec
+    let projectSpec: ProjectSpec;
+    try {
+      projectSpec = JSON.parse(fs.readFileSync(projectJsonPath, 'utf8'));
+    } catch (err) {
+      return createGenomeDemoRunResultFromParams(params, {
+        success: false,
+        handle: '',
+        projectName: project,
+        milestoneCount: 0,
+        error: `Failed to parse project.json: ${err}`,
+      });
+    }
+
+    // Validate scaffold exists
+    const scaffoldDir = path.join(projectDir, 'scaffold');
+    if (!fs.existsSync(path.join(scaffoldDir, 'package.json'))) {
+      return createGenomeDemoRunResultFromParams(params, {
+        success: false,
+        handle: '',
+        projectName: projectSpec.name,
+        milestoneCount: projectSpec.milestones.length,
+        error: `Missing scaffold/package.json in ${projectDir}`,
+      });
+    }
+
+    // Validate test files exist
+    for (const milestone of projectSpec.milestones) {
+      const testPath = path.join(projectDir, milestone.testFile);
+      if (!fs.existsSync(testPath)) {
+        return createGenomeDemoRunResultFromParams(params, {
+          success: false,
+          handle: '',
+          projectName: projectSpec.name,
+          milestoneCount: projectSpec.milestones.length,
+          error: `Missing test file: ${milestone.testFile} (milestone ${milestone.index}: ${milestone.name})`,
+        });
+      }
+    }
+
+    const personaName = params.personaName ?? 'demo-persona';
+    const baseModel = params.baseModel ?? LOCAL_MODELS.DEFAULT;
+
+    console.log(`\u{1F680} DEMO RUN: project="${projectSpec.name}", persona="${personaName}", milestones=${projectSpec.milestones.length}`);
+
+    // Build demo pipeline config
+    const config: DemoPipelineConfig = {
+      projectDir,
+      project: projectSpec,
+      personaId,
+      personaName,
+      baseModel,
+      maxRetries: params.maxRetries ?? DEMO_DEFAULTS.maxRetries,
+      maxBudgetPerMilestone: params.maxBudget ?? DEMO_DEFAULTS.maxBudgetPerMilestone,
+      maxTurnsPerMilestone: params.maxTurns ?? DEMO_DEFAULTS.maxTurnsPerMilestone,
+      provider: params.provider ?? DEMO_DEFAULTS.provider,
+      training: {
+        ...DEFAULT_DEMO_TRAINING_CONFIG,
+        ...(params.epochs !== undefined && { epochs: params.epochs }),
+        ...(params.rank !== undefined && { rank: params.rank }),
+      },
+    };
+
+    // Build the pipeline
+    const pipeline = buildDemoPipeline(config);
+
+    console.log(`   Pipeline: ${pipeline.name}, ${pipeline.steps.length} top-level steps`);
+
+    // Dispatch to Rust sentinel executor
+    // PipelineStep[] (Rust bindings) → SentinelStep[] (TS definitions) — structurally compatible wire types
+    const pipelineSteps = pipeline.steps as unknown as SentinelStep[];
+
+    const sentinelResult = await Commands.execute<PipelineSentinelParams, SentinelRunResult>('sentinel/run', {
+      type: 'pipeline',
+      definition: {
+        type: 'pipeline',
+        name: pipeline.name ?? `demo-${project}`,
+        description: `Demo pipeline: Claude Code builds ${projectSpec.name} (${projectSpec.milestones.length} milestones), captures training data for ${personaName}`,
+        version: '1.0',
+        steps: pipelineSteps,
+        loop: { type: 'once' },
+        tags: ['demo', project, projectSpec.skill],
+      },
+      parentPersonaId: personaId,
+      sentinelName: pipeline.name ?? `demo-${project}`,
+    });
+
+    const handle = sentinelResult.handle ?? '';
+    console.log(`\u2705 DEMO RUN: Pipeline dispatched, handle=${handle}`);
+
+    return createGenomeDemoRunResultFromParams(params, {
+      success: true,
+      handle,
+      projectName: projectSpec.name,
+      milestoneCount: projectSpec.milestones.length,
+    });
+  }
+}