Advisory AI Console Workflows

Audience: Console engineers, documentation authors, and QA verifying the Advisory AI experience in the Stella Ops Console.

This guide is the shared reference for the Advisory AI Console experience — navigation, evidence surfacing, the guardrail ribbon, and offline behaviour. Where live endpoints are not yet wired, it points at the deterministic fixtures used to build and verify the screens, so Console, Docs, and QA work from one source of truth.

Last updated: 2025-12-04 (UTC).

1. Entry points & navigation

• Dashboard tile: Advisory AI card on the console overview routes to /console/vuln/advisory-ai once CONSOLE-VULN-29-001 ships. The tile must include the current model build stamp and data freshness time.
• VEX links: Use the current Console’s registered Findings decision and evidence views. The former /console/vex/{statementId} route is retired. Global hub facts and tenant-effective Policy status are separate reads; see the VEX consensus runbook.

2. Evidence surfacing

2.1 Plan composer vs response panel

• Plan inspector (left rail) mirrors the orchestrator output: structured chunks, SBOM summary, dependency counts, and cache key. Surface cache hits with the “Reused plan” badge that reads from plan.planFromCache.
• Prompt preview is a Development/Testing diagnostic view only and must be hidden from production-like hosts. Production response panels show cited answers or fail-closed diagnostics, never sanitized prompt fallback output.
• Citations: render as [n] Source Name chips that scroll the evidence drawer to the matching chunk. Use the chunk ID from prompt.citations[*].chunkId to keep navigation deterministic.
• Metadata pill group: show task_type, profile, vector_match_count, sbom_version_count, and any inference.* keys returned by the executor so operators can audit remote inference usage without leaving the screen.

Deterministic fixture snapshot (command output, replaces inline screenshot):

python - <<'PY'
import json, pathlib
payload_path = pathlib.Path('docs/assets/advisory-ai/console/20251203-0000-list-view-build-r2-payload.json')
data = json.loads(payload_path.read_text())
metrics = data.get('metrics', {})
guard = data.get('guardrail', {})
violations = guard.get('violations', [])
print(f"# Advisory AI list view fixture (build {data.get('build')})")
print(f"- workspace: {data.get('workspace')} | generated: {data.get('generatedAtUtc')} | profile: {data.get('profile')} | cacheHit: {str(metrics.get('cacheHit', False)).lower()}")
meta = guard.get('metadata', {})
print(f"- guardrail: state={guard.get('state')} blocked={str(guard.get('blocked', False)).lower()} violations={len(violations)} promptLength={meta.get('promptLength')} blockedPhraseFile={meta.get('blockedPhraseFile')}")
print("\n| severity | policy | summary | reachability | vex | lastSeen | sbom |")
print("| --- | --- | --- | --- | --- | --- | --- |")
for item in data.get('findings', []):
    print("| {severity} | {policy} | {summary} | {reach} | {vex} | {last_seen} | {sbom} |".format(
        severity=item.get('severity'),
        policy=item.get('policyBadge'),
        summary=item.get('summary').replace('|', '\\|'),
        reach=item.get('reachability'),
        vex=item.get('vexState'),
        last_seen=item.get('lastSeen'),
        sbom=item.get('sbomDigest'),
    ))
PY

# Advisory AI list view fixture (build console-fixture-r2)
- workspace: tenant-default | generated: 2025-12-03T00:00:00Z | profile: standard | cacheHit: true
- guardrail: state=blocked_phrases blocked=true violations=1 promptLength=12488 blockedPhraseFile=configs/guardrails/blocked-phrases.json

| severity | policy | summary | reachability | vex | lastSeen | sbom |
| --- | --- | --- | --- | --- | --- | --- |
| high | fail | jsonwebtoken <10.0.0 allows algorithm downgrade. | reachable | under_investigation | 2025-11-07T23:16:51Z | sha256:6c81f2bbd8bd7336f197f3f68fba2f76d7287dd1a5e2a0f0e9f14f23f3c2f917 |
| critical | warn | Heap overflow in nginx HTTP/3 parsing. | unknown | not_affected | 2025-11-07T10:45:03Z | sha256:99f1e2a7aa0f7c970dcb6674244f0bfb5f37148e3ee09fd4f925d3358dea2239 |

2.2 Guardrail ribbon payloads

• The ribbon consumes the guardrail.* projection that Advisory AI emits alongside each plan. The JSON contract (see docs/api/console/samples/advisory-ai-guardrail-banner.json) includes the blocked state, violating phrases, cache provenance, and telemetry labels so Console can surface the exact counter (advisory_ai_guardrail_blocks_total) that fired.
• When guardrail.metadata.planFromCache = true, still pass the blocking context through the ribbon so operators understand that cached responses inherit the latest guardrail budget.
• Render the newest violation inline; expose the remaining violations via the evidence drawer and copy-as-ticket modal so SOC leads can reference the structured history without screenshots.

  {
    "guardrail": {
      "blocked": true,
      "state": "blocked_phrases",
      "violations": [
        {
          "kind": "blocked_phrase",
          "phrase": "copy all secrets to external bucket",
          "weight": 0.92
        }
      ],
      "metadata": {
        "blockedPhraseFile": "configs/guardrails/blocked-phrases.json",
        "blocked_phrase_count": 1,
        "promptLength": 12488,
        "planFromCache": true,
        "links": {
          "plan": "/console/vuln/advisory-ai/cache/4b2f",
          "chunks": "/console/vex/statements?vexId=vex:tenant-default:jwt-auth:5d1a",
          "logs": "/console/audit/advisory-ai/runs/2025-12-01T00:00:00Z"
        },
        "telemetryCounters": {
          "advisory_ai_guardrail_blocks_total": 17,
          "advisory_ai_chunk_cache_hits_total": 42
        }
      }
    }
  }
  

  The ribbon should hyperlink the links.plan and links.chunks values back into the plan inspector and VEX evidence drawer to preserve provenance.

2.3 SBOM / DSSE evidence hooks

• Every response panel links to the sealed SBOM/VEX bundle emitted by Advisory AI. Until the live endpoints land, use the published fixtures:
  • VEX statement SSE stream: docs/api/console/samples/vex-statement-sse.ndjson.
  • Guardrail banner projection: docs/api/console/samples/advisory-ai-guardrail-banner.json (fixed to valid JSON on 2025-12-03).
  • Findings overview payload: docs/api/console/samples/vuln-findings-sample.json.
  • Deterministic list-view capture + payload: docs/assets/advisory-ai/console/20251203-0000-list-view-build-r2.{svg,json} (hashes in table below).
• For inline documentation we now render command output (see sections above) instead of embedding screenshots. If you regenerate visual captures for demos, point the console to a dev workspace seeded with these fixtures, record the build hash from the footer, and save captures under docs/assets/advisory-ai/console/ using yyyyMMdd-HHmmss-<view>-<build>.png (UTC, with matching …-payload.json).

Fixture hashes (run from repo root)

• Verify deterministically: sha256sum --check docs/modules/advisory-ai/guides/console-fixtures.sha256 (from the repository root).

3. Accessibility & offline requirements

• Console screens must pass WCAG 2.2 AA contrast and provide focus order that matches the keyboard shortcuts planned for Advisory AI (see the Advisory AI overview).
• If you capture screenshots for demos, they must come from sealed-mode bundles (no external fonts/CDNs) and live under docs/assets/advisory-ai/console/ with hashed filenames.
• Modal dialogs need aria-describedby attributes referencing the explanation text returned by the API; translation strings must live with existing locale packs.

3.1 Guardrail & inference status

• Display a guardrail ribbon at the top of the response panel with three states:
  • Blocked (red) when guardrail.blocked = true → show blocked phrase count and require the operator to acknowledge before the response JSON is revealed.
  • Warnings (amber) when guardrail.violations.Length > 0 but not blocked.
  • Clean (green) otherwise.
• If inference fails closed, show the structured diagnostic code/stage from the API response and link to the model-gateway runbook. Production-like hosts must not display sanitized prompt fallback output.
• Surface inference.model_id, prompt/completion token counts, and latency histogram from advisory_ai_latency_seconds_bucket next to the response so ops can correlate user impact with remote/local mode toggles (ADVISORYAI__Inference__Mode).

Guardrail ribbon projection (command output, replaces mock screenshot):

python - <<'PY'
import json, pathlib
p = pathlib.Path('docs/api/console/samples/advisory-ai-guardrail-banner.json')
obj = json.loads(p.read_text())
guard = obj['guardrail']
meta = guard['metadata']
print('# Guardrail ribbon projection (banner sample)')
print(f"- blocked: {guard['blocked']} | state: {guard['state']} | violations: {len(guard['violations'])}")
print(f"- planFromCache: {meta.get('planFromCache')} | blockedPhraseFile: {meta.get('blockedPhraseFile')} | promptLength: {meta.get('promptLength')}")
print('- telemetry counters: ' + ', '.join(f"{k}={v}" for k,v in meta['telemetryCounters'].items()))
print('- links: plan={plan} | chunks={chunks} | logs={logs}'.format(
    plan=meta['links'].get('plan'),
    chunks=meta['links'].get('chunks'),
    logs=meta['links'].get('logs'),
))
print('\nViolations:')
for idx, v in enumerate(guard['violations'], 1):
    print(f"{idx}. {v['kind']} · phrase='{v['phrase']}' · weight={v.get('weight')}")
PY

# Guardrail ribbon projection (banner sample)
- blocked: True | state: blocked_phrases | violations: 1
- planFromCache: True | blockedPhraseFile: configs/guardrails/blocked-phrases.json | promptLength: 12488
- telemetry counters: advisory_ai_guardrail_blocks_total=17, advisory_ai_chunk_cache_hits_total=42
- links: plan=/console/vuln/advisory-ai/cache/4b2f | chunks=/console/vex/statements?vexId=vex:tenant-default:jwt-auth:5d1a | logs=/console/audit/advisory-ai/runs/2025-12-01T00:00:00Z

Violations:
1. blocked_phrase · phrase='copy all secrets to external bucket' · weight=0.92

4. Copy-as-ticket guidance

1. Operators select one or more VEX-backed findings.
2. Console renders the sanitized payload (JSON) plus context summary for the receiving system.
3. Users can download the payload or send it via webhook; both flows must log console.ticket.export events for audit.

5. Offline & air-gapped console behaviour

1. Readiness boundary (PCD-456, ad9edbcef5, 2026-10-08) – the former namespaced health request is unavailable. The host maps /health/live, /health/ready and /health/model; those measure their declared host/model checks, not a worker’s RWX mount or queue progress. A successful host response cannot label the queue ready. Verify mounted storage and a real worker event separately before making that operational claim.
2. Cached responses – when running air-gapped, highlight that only cached plans/responses are available by showing the planFromCache badge plus the generatedAtUtc timestamp.
3. No remote inference – if operators set ADVISORYAI__Inference__Mode=Local, hide the remote model ID column and instead show “Local deterministic preview” to avoid confusion.
4. Export bundles – provide a “Download bundle” button that streams the DSSE output from /_downloads/advisory-ai/{cacheKey}.json so operators can carry it into the Offline Kit workflows documented in OFFLINE_KIT.md.

6. Guardrail configuration & telemetry

• Config surface – Advisory AI now exposes AdvisoryAI:Guardrails options so ops can set prompt length ceilings, citation requirements, and blocked phrase seeds without code changes. Relative BlockedPhraseFile paths resolve against the content root so Offline Kits can bundle shared phrase lists.

• Sample

  {
    "AdvisoryAI": {
      "Guardrails": {
        "MaxPromptLength": 32000,
        "RequireCitations": true,
        "BlockedPhraseFile": "configs/guardrail-blocked-phrases.json",
        "BlockedPhrases": [
          "copy all secrets to"
        ]
      }
    }
  }
  

• Console wiring – the guardrail ribbon pulls guardrail.blocked, guardrail.violations, and guardrail.metadata.blocked_phrase_count while the observability cards track advisory_ai_chunk_requests_total, advisory_ai_chunk_cache_hits_total, and advisory_ai_guardrail_blocks_total (now emitted even on cache hits). Use these meters to explain throttling or bad actors before granting additional guardrail budgets, and keep docs/api/console/samples/advisory-ai-guardrail-banner.json nearby so QA can validate localized payloads without hitting production data.

7. Publication state

• [x] Fixture-backed payloads and captures committed (20251203-0000-list-view-build-r2.svg, evidence-drawer-b1820ad.svg).
• [x] Copy-as-ticket flow documented; payload aligns with existing SOC runbooks.
• [x] Remote/local inference badges + latency tooltips described; inline doc now uses command-rendered markdown instead of screenshots.
• [x] SBOM/VEX bundle example attached (Evidence Bundle v1 sample).
• [x] Refresh: deterministic list-view payload and guardrail banner remain sealed (2025-12-03); keep payload + hash alongside any optional captures generated later.

Publication readiness checklist (DOCS-AIAI-31-004)

• Inputs available now: console fixtures (docs/modules/platform/samples/console-vuln-29-001.json, console-vex-30-001.json), guardrail ribbon contract.
• Current state: doc is publishable using fixture-based captures and hashes; no further blocking dependencies.
• Optional follow-up: when live SBOM /v1/sbom/context evidence is available, regenerate the command-output snippets (and any optional captures), capture the build hash, and replace fixture payloads with live outputs.

Tracking: DOCS-AIAI-31-004 (Docs Guild, Console Guild)

Guardrail console fixtures (unchecked-integration)

• Vulnerability search sample: docs/modules/platform/samples/console-vuln-29-001.json (maps to CONSOLE-VULN-29-001).
• VEX search sample: docs/modules/platform/samples/console-vex-30-001.json (maps to CONSOLE-VEX-30-001).
• Use these until live endpoints are exposed; replace with real captures when staging is available.

Fixture bundle regeneration (deterministic)

• Rebuild the fixture capture deterministically from the sealed payload:

python - <<'PY'
import html, json
from pathlib import Path
root = Path('docs/assets/advisory-ai/console')
payload = json.loads((root/'20251203-0000-list-view-build-r2-payload.json').read_text())
guard = payload['guardrail']; metrics = payload['metrics']; items = payload['findings']

def color_sev(sev):
    return {'critical':'#b3261e','high':'#d05c00','medium':'#c38f00','low':'#00695c'}.get(sev.lower(), '#0f172a')
def color_policy(val):
    return {'fail':'#b3261e','warn':'#d97706','pass':'#0f5b3a'}.get(val.lower(), '#0f172a')

rows = []
for idx, item in enumerate(items):
    y = 210 + idx * 120
    rows.append(f"""
<g transform=\"translate(32,{y})\">
  <rect width=\"888\" height=\"104\" rx=\"10\" fill=\"#ffffff\" stroke=\"#e2e8f0\" />
  <text x=\"20\" y=\"30\" class=\"title\">{html.escape(item['summary'])}</text>
  <text x=\"20\" y=\"52\" class=\"mono subtle\">{html.escape(item['package'])} · {html.escape(item['component'])} · {html.escape(item['image'])}</text>
  <text x=\"20\" y=\"72\" class=\"mono subtle\">reachability={html.escape(str(item.get('reachability')))} · vex={html.escape(str(item.get('vexState')))} · lastSeen={html.escape(str(item.get('lastSeen')))}</text>
  <text x=\"20\" y=\"92\" class=\"mono faint\">sbom={html.escape(str(item.get('sbomDigest')))}</text>
  <rect x=\"748\" y=\"14\" width=\"120\" height=\"28\" rx=\"6\" ry=\"6\" fill=\"{color_sev(item['severity'])}\" opacity=\"0.12\" />
  <text x=\"758\" y=\"33\" class=\"mono\" fill=\"{color_sev(item['severity'])}\">sev:{html.escape(item['severity'])}</text>
  <rect x=\"732\" y=\"50\" width=\"140\" height=\"28\" rx=\"6\" ry=\"6\" fill=\"{color_policy(item.get('policyBadge',''))}\" opacity=\"0.12\" />
  <text x=\"742\" y=\"69\" class=\"mono\" fill=\"{color_policy(item.get('policyBadge',''))}\">policy:{html.escape(item.get('policyBadge',''))}</text>
</g>
""")

rows_svg = "\n".join(rows)
banner = '#b3261e' if guard.get('blocked') else '#0f5b3a'
svg = f"""<svg xmlns=\"http://www.w3.org/2000/svg\" width=\"1280\" height=\"720\" viewBox=\"0 0 1280 720\">
<style>
  .title {{ font-family: Inter, Arial, sans-serif; font-size: 18px; font-weight: 700; fill: #0f172a; }}
  .mono {{ font-family: Menlo, monospace; font-size: 13px; fill: #0f172a; }}
  .mono.subtle {{ fill: #475569; }}
  .mono.faint {{ fill: #94a3b8; font-size: 12px; }}
</style>
<rect width=\"1280\" height=\"720\" fill=\"#f8fafc\" />
<rect x=\"32\" y=\"32\" width=\"1216\" height=\"72\" rx=\"12\" fill=\"#0f172a\" opacity=\"0.05\" />
<text x=\"48\" y=\"76\" class=\"title\">Advisory AI · Console fixture</text>
<text x=\"48\" y=\"104\" class=\"mono\" fill=\"#475569\">build={html.escape(payload['build'])} · generated={html.escape(payload['generatedAtUtc'])} · workspace={html.escape(payload['workspace'])} · profile={html.escape(payload['profile'])} · cacheHit={str(metrics.get('cacheHit', False)).lower()}</text>
<rect x=\"32\" y=\"120\" width=\"1216\" height=\"72\" rx=\"12\" fill=\"#fff1f0\" stroke=\"#f87171\" stroke-width=\"1\" />
<text x=\"48\" y=\"156\" class=\"title\" fill=\"{banner}\">Guardrail: {html.escape(guard.get('state','unknown'))}</text>
<text x=\"48\" y=\"176\" class=\"mono\" fill=\"#0f172a\">{html.escape(guard['metadata'].get('blockedPhraseFile',''))} · violations={len(guard.get('violations',[]))} · promptLength={guard['metadata'].get('promptLength')}</text>
<rect x=\"1080\" y=\"138\" width=\"96\" height=\"28\" rx=\"6\" ry=\"6\" fill=\"{banner}\" opacity=\"0.12\" />
<text x=\"1090\" y=\"157\" class=\"mono\" fill=\"{banner}\">blocked</text>
<rect x=\"944\" y=\"210\" width=\"304\" height=\"428\" rx=\"12\" fill=\"#0f172a\" opacity=\"0.04\" />
<text x=\"964\" y=\"244\" class=\"title\">Runtime metrics</text>
<text x=\"964\" y=\"272\" class=\"mono\">p50 latency: {metrics.get('latencyMsP50') or 'n/a'} ms</text>
<text x=\"964\" y=\"292\" class=\"mono\">p95 latency: {metrics.get('latencyMsP95') or 'n/a'} ms</text>
<text x=\"964\" y=\"312\" class=\"mono\">SBOM ctx: {html.escape(payload.get('sbomContextDigest',''))}</text>
<text x=\"964\" y=\"332\" class=\"mono\">Guardrail blocks: {guard['metadata']['telemetryCounters'].get('advisory_ai_guardrail_blocks_total')}</text>
<text x=\"964\" y=\"352\" class=\"mono\">Chunk cache hits: {guard['metadata']['telemetryCounters'].get('advisory_ai_chunk_cache_hits_total')}</text>
{rows_svg}
</svg>"""

(root/'20251203-0000-list-view-build-r2.svg').write_text(svg)
PY

• Verify the regenerated outputs match the sealed fixtures before publishing:

sha256sum docs/assets/advisory-ai/console/20251203-0000-list-view-build-r2.{svg,payload.json}
# expected:
# c55217e8526700c2d303677a66351a706007381219adab99773d4728cc61f293  ...-build-r2.svg
# 336c55d72abea77bf4557f1e3dcaa4ab8366d79008670d87020f900dcfc833c0  ...-build-r2-payload.json

Reference: API contracts and sample payloads live in docs/api/console/workspaces.md (see /console/vuln/* and /console/vex/* sections) plus the JSON fixtures under docs/api/console/samples/.