interhelm:diagnostic-maturation — Diagnostic Server Maturation Guide

When to Use

Use when:

• You've scaffolded a diagnostic server (via interhelm:runtime-diagnostics or manually) and it only returns basic boolean health
• You want to build CLI operator tools that call the server's diagnostic endpoints
• You need to audit an existing diagnostic server's maturity level
• You're evolving beyond the scaffold toward production-grade operator tooling

Do NOT use when:

• You don't have a diagnostic server yet (use interhelm:runtime-diagnostics first to scaffold endpoints)
• You're building monitoring/alerting infrastructure (this is operator tooling, not ops)
• The app is a web service with existing APM (Datadog, New Relic, etc.)

Relationship to runtime-diagnostics: That skill scaffolds the server-side endpoints (health, diff, assert, smoke-test HTTP routes). This skill guides building the CLI-side operator tools that call those endpoints — the client layer, formatting, interactive modes, and the domain modeling that makes health checks meaningful.

Mode Selection

Infer the mode from context:

• If the prompt mentions "audit", "assess", "evaluate", or "what level" → Audit mode (§ Conformance Audit)
• If the prompt mentions "improve", "evolve", "mature", "next level", or "add" → Maturation mode (§ Maturation Ladder)
• If ambiguous, default to Audit mode first (assess current state, then guide improvement)

---

Maturation Ladder

Six levels from skeleton to production-grade. Each level is independently useful. Levels are sequential by default — each builds on the previous — but agents can skip levels if the app doesn't need them (e.g., skip L5 Diff Engine if the app has no snapshot-comparison use case).

Domain translation: The examples below use Shadow Work (a geopolitical simulation) as the reference. To apply to your domain, map "subsystem" to your app's major independent functional areas. Examples:

• File sync tool: watcher, transfer queue, conflict resolver, network layer
• CLI dev tool: parser, linter, formatter, cache
• Game engine: renderer, physics, audio, input, networking
• API server: auth, data layer, job queue, external integrations

Level 1: Domain Health Modeling

The hardest and most important level. Most agents stop at "return true for healthy" — the real work is deciding WHAT to check.

Process:

1. Identify subsystems. List the 3-8 major subsystems of your application. Examples from Shadow Work (6 subsystems): simulation engine, economy, finance, emergence/AI, countries/entities, error tracking.

   Ask yourself: "If this subsystem is broken, would the app behave incorrectly?" If yes, it's a subsystem.

2. Define states per subsystem. Each subsystem has three states:

   • Healthy — operating within normal parameters
   • Degraded — functional but outside ideal range (e.g., tick time >100ms but <500ms)
   • Unhealthy — broken or producing incorrect results

3. Wire detail fields. Each subsystem's health response should include domain-specific metrics, not just a boolean. Examples:

   Simulation: { status, tick_count, avg_tick_ms, last_tick_ms, speed }
   Economy:    { status, gdp_total, trade_volume, active_markets }
   Finance:    { status, total_treasury, tax_revenue_rate }
   

4. Set meaningful thresholds. Define concrete numbers for state transitions:

   Simulation healthy:  avg_tick_ms < 100
   Simulation degraded: avg_tick_ms >= 100 && avg_tick_ms < 500
   Simulation unhealthy: avg_tick_ms >= 500 OR tick_count == 0
   

5. Implement the aggregated health endpoint. Return per-subsystem status with detail fields:

   {
     "overall": "degraded",
     "subsystems": {
       "simulation": { "status": "healthy", "tick_count": 4200, "avg_tick_ms": 42 },
       "economy":    { "status": "degraded", "gdp_total": 0, "active_markets": 12 }
     }
   }
   

You know you're done when:

• Structural check: curl /diag/health | python3 -c "import json,sys; d=json.load(sys.stdin); assert 'subsystems' in d; assert len(d['subsystems'])>=3; assert all('status' in v for v in d['subsystems'].values()); print('L1 PASS')" prints L1 PASS
• Design check (human judgment): Each subsystem has at least one metric beyond status, and thresholds for healthy/degraded/unhealthy are documented in code comments or a config file

---

Level 2: Smoke Test Suite

Build 5-15 end-to-end assertions that verify the app is working correctly. These are not unit tests — they exercise the running app through its diagnostic endpoints.

Process:

1. Design test categories. Group tests by subsystem. Each test has a name, assertion, and expected result. Example from Shadow Work (12-point suite):

   connectivity:    Can reach the debug server
   simulation:      Sim is running (not paused, tick count > 0)
   economy:         GDP is positive
   finance:         Treasury is non-negative
   countries:       Country count > 0
   emergence:       Agent count matches expected
   trade:           Trade volume > 0
   error-rate:      Error count below threshold
   

2. Implement pass/fail/skip semantics:

   • Pass — assertion succeeded
   • Fail — assertion failed (the thing we're checking is broken)
   • Skip — can't evaluate (endpoint unreachable, data not yet available)

   Skip is critical — without it, agents can't distinguish "feature is broken" from "feature hasn't loaded yet."

3. Run tests sequentially and collect results. Return structured output:

   {
     "passed": 10, "failed": 1, "skipped": 1, "total": 12,
     "tests": [
       { "name": "connectivity", "status": "pass", "ms": 12 },
       { "name": "economy-gdp", "status": "fail", "detail": "GDP is 0" },
       { "name": "emergence-agents", "status": "skip", "detail": "emergence not initialized" }
     ]
   }
   

4. Wire a CLI command: <app>-agent smoke that runs the suite and prints colored output.

You know you're done when: Running <app>-agent smoke prints a pass/fail/skip summary with per-test results, and at least 5 tests cover different subsystems.

---

Level 3: Assert Language

Build a lightweight expression evaluator for scripted verification. This lets agents (and humans) run ad-hoc assertions without modifying code.

Process:

1. Define expression syntax. Keep it simple — compound assertions joined by &&:

   health.simulation.status == healthy && health.simulation.avg_tick_ms < 100
   

2. Build a state context. Flatten the diagnostic server's state into a queryable namespace:

   health.simulation.status → "healthy"
   health.simulation.tick_count → 4200
   health.economy.gdp_total → 1500000
   

3. Evaluate clauses. For each clause (split on &&):

   • Parse left-hand side (dotted path into state context)
   • Parse operator (==, !=, <, >, <=, >=, contains)
   • Parse right-hand side (literal value)
   • Resolve LHS against state context
   • Compare and return pass/fail with explanation

4. Report results per clause with colored output:

   ✓ health.simulation.status == healthy
   ✗ health.economy.gdp_total > 1000000 (actual: 0)
   RESULT: FAIL (1/2 passed)
   

5. Wire a CLI command: <app>-agent assert "<expression>"

You know you're done when: <app>-agent assert "health.simulation.status == healthy && health.simulation.tick_count > 0" evaluates both clauses against live state and prints per-clause pass/fail.

---

Level 4: CLI Formatting

Build human-readable output with compact numbers, colored status, and aligned tables.

Process:

1. Compact number formatting. Large numbers should be human-readable:

   1234        → "1,234"
   1500000     → "1.5M"
   2300000000  → "2.3B"
   4500000000000 → "4.5T"
   

   Thresholds: K (≥1,000), M (≥1,000,000), B (≥1,000,000,000), T (≥1,000,000,000,000).

2. Percentage and currency helpers:

   pct(0.1534) → "15.3%"
   money(1500000) → "$1.5M"
   

3. Status coloring. Use ANSI colors consistently:

   healthy/pass  → green
   degraded/skip → yellow
   unhealthy/fail → red
   

4. Table padding. Right-align numbers, left-align text. Use consistent column widths:

   Country          GDP      Population   Status
   United States    $21.4T   331M         healthy
   China            $14.7T   1.4B         degraded
   

5. Apply formatting to all existing commands (status, smoke, assert, health).

You know you're done when: All CLI commands produce aligned, colored output with compact numbers. Large values like 1500000000 display as "1.5B", not raw integers.

---

Level 5: Diff Engine

Build snapshot-before/after comparison with domain-specific deltas.

Process:

1. Capture snapshots. Before an operation (e.g., advancing 100 simulation ticks), snapshot the full diagnostic state:

   snapshot_before = GET /diag/state
   

2. Run the operation. Execute the action being tested.

3. Capture after state:

   snapshot_after = GET /diag/state
   

4. Compute domain-specific deltas. Don't just diff JSON — compute meaningful changes:

   Simulation: +100 ticks, avg_tick_ms 42→38 (-9.5%)
   Economy:    GDP $1.5M→$1.8M (+20.0%), 3 new markets
   Countries:  2 countries changed status
   

5. Wire a CLI command: <app>-agent diff <command> that snapshots, runs, snapshots, and shows deltas.

You know you're done when: <app>-agent diff "step 100" shows before/after comparison with domain-specific deltas (not raw JSON diff), formatted with compact numbers and colors.

---

Level 6: Interactive REPL + Watch Mode

Add interactive exploration and continuous monitoring.

Process:

1. REPL mode. Interactive prompt that accepts any CLI command without the binary prefix:

   > status
   (shows status)
   > assert "health.simulation.status == healthy"
   ✓ health.simulation.status == healthy
   > smoke
   12/12 passed
   > help
   (shows available commands)
   

   Key features:

   • Command history (readline/rustyline)
   • Tab completion for commands
   • .exit or Ctrl+D to quit

2. Watch mode. Poll a command at an interval and redisplay:

   <app>-agent watch status 5    # refresh status every 5 seconds
   <app>-agent watch smoke 30    # run smoke tests every 30 seconds
   

   Key features:

   • Clear screen between updates (if TTY detected; print separator lines otherwise)
   • Show timestamp of last update
   • Ctrl+C to stop
   • Highlight changes since last update
   • Graceful degradation in non-TTY environments (agent shells, piped output): skip ANSI colors, use text separators instead of screen clears

3. Wire CLI commands: <app>-agent repl and <app>-agent watch <cmd> [interval]

You know you're done when: <app>-agent repl gives an interactive prompt with history, and <app>-agent watch status 5 continuously refreshes the display.

---

Conformance Audit

Evaluate an existing diagnostic server against the maturation ladder.

Process

1. Discover the diagnostic server. Find the diagnostic endpoint configuration:

   # Look for diagnostic server code
   grep -r "/diag/" src/ src-tauri/ --include="*.rs" --include="*.ts" --include="*.py" -l 2>/dev/null
   # Look for CLI tools
   find tools/ bin/ -name "*agent*" -o -name "*diag*" 2>/dev/null
   

2. Check each level. For each maturation level, look for evidence of implementation:

   Level	Check for	Evidence
   L1: Domain Health	Per-subsystem health with detail fields	Health endpoint returns JSON with >1 subsystem, each having status + metrics
   L2: Smoke Tests	Structured test suite with pass/fail/skip	File with test definitions, CLI command that runs them, results with counts
   L3: Assert Language	Expression evaluator	CLI command accepting assertion strings, clause-by-clause evaluation
   L4: CLI Formatting	Compact numbers, colored output	Formatter module with compactNum/pct/money helpers, ANSI color usage
   L5: Diff Engine	Snapshot comparison	Before/after capture, domain-specific delta computation
   L6: REPL + Watch	Interactive and continuous modes	Readline/prompt loop, polling with interval

3. Report findings. For each level, report:

   • Implemented — evidence found, list what exists
   • Partial — some elements present, list gaps
   • Missing — no evidence found

4. Provide next steps. For the first missing or partial level, provide actionable guidance:

   Current Level: 3/6 (Assert Language)
   
   Implemented:
     L1 Domain Health: ✓ 6 subsystems with thresholds
     L2 Smoke Tests:   ✓ 12-point suite with pass/fail/skip
     L3 Assert:        ✓ Expression evaluator with && support
   
   Missing:
     L4 CLI Formatting: No compact number helpers found. Start with:
       - Create a formatter module with compactNum(), pct(), money()
       - Apply to status and smoke test output
     L5 Diff Engine:   No snapshot comparison found
     L6 REPL + Watch:  No interactive mode found
   
   Recommended next: Level 4 (CLI Formatting)
   

Shadow Work Reference

Shadow Work's sw-agent is a Level 6 reference implementation:

• L1: 6 subsystems (simulation, economy, finance, emergence, countries, errors) with healthy/degraded/unhealthy per subsystem
• L2: 12-point smoke test suite with SmokeTest class (pass/fail/skip semantics, timing)
• L3: runAssert(client, expr) — splits on &&, evaluates against flattened state context
• L4: compactNum() (T/B/M/K), pct(), money(), pad()/padLeft() helpers
• L5: Snapshot-before/after with per-subsystem deltas
• L6: Interactive REPL with readline + watch command with configurable interval

Use this as a reference for what "done" looks like at each level, adapted to your app's domain.