Files
akushonkamen 87d8687f3f fix(ruflo-agentdb): stop the secret screen treating a hyphen-named memory key as a credential field (#4202) (#4211)
* fix(ruflo-agentdb): stop the secret screen treating a hyphen-named memory key as a credential field

assignmentSecret() ran the KEY_NAME key-assignment check on every match, so
any entry whose name starts with auth (auth-flow-notes) plus up to 40 name
characters looked like a config field, and keyedSecret()/plausibleSecret()
then judged the entry's value as a credential: a write of
{ key: 'auth-flow-notes', value: 'getUser-not-getSession' } was refused
(#4202). A key that is itself a multi-word hyphenated name is an entry's
name, not a field, so the keyed check now skips it; the secret shapes still
screen the whole text, and single-word credential keys (auth, api_key,
token) keep blocking keyed literals.

Synced to the 39 plugin copies with scripts/sync-mod-screen.mjs.

Fixes #4202
Assisted-by: AI (automated contributor session, disclosed on the issue)

* chore(ci): exclude plugins/ruflo-agentdb/tests/guard-keys.test.ts from the root vitest run

The new test imports claude-code/testing like its six siblings, so the root
ratchet run (scripts/ci-test-ratchet.mjs) cannot collect it outside Claude
Code; list it in scripts/ci-test-excluded.txt per the documented convention
so ci.yml stays green until 'claude plugin test plugins/ruflo-agentdb' runs
it in all-plugins-smoke.

Follow-up to #4202
Assisted-by: AI (automated contributor session, disclosed on the issue)
2026-10-17 14:58:08 +01:00
..

ruflo-iot-cognitum

IoT device lifecycle, telemetry anomaly detection, fleet management, and witness chain verification for Cognitum Seed hardware.

Hardware

This plugin requires a Cognitum Seed device. Get one at https://cognitum.one — the Seed is an edge appliance with on-device vector store, Ed25519 identity, OTA firmware, mesh networking, and a witness chain. Default address when attached via USB-C is http://169.254.42.1 (link-local, no auth) or https://169.254.42.1:8443 (LAN, bearer auth required for state-mutating operations).

Overview

Treats every Cognitum Seed device as a Ruflo agent with hardware capabilities. Devices progress through a 5-tier trust model, emit telemetry vectors for anomaly detection, participate in mesh networks, and maintain Ed25519 witness chains for provenance.

Backed by @claude-flow/plugin-iot-cognitum (239 tests, 39 source files).

Installation

claude --plugin-dir plugins/ruflo-iot-cognitum

Agents

Agent Model Role
device-coordinator sonnet Device lifecycle, 5-tier trust scoring, mesh coordination
telemetry-analyzer sonnet Z-score anomaly detection, SONA learning, AgentDB persistence
fleet-manager sonnet Fleet CRUD, firmware rollout state machine, fleet policies
witness-auditor haiku Witness chain epoch verification, gap detection

Skills

Skill Usage Description
iot-register /iot-register <endpoint> Register a Seed device
iot-fleet /iot-fleet <create|list|add|remove|delete> Fleet management
iot-anomalies /iot-anomalies <device-id> Detect telemetry anomalies
iot-firmware /iot-firmware <deploy|advance|rollback|status|list> Firmware rollouts
iot-witness-verify /iot-witness-verify <device-id> Verify witness chain integrity

Commands (25 subcommands)

# Device lifecycle
# `endpoint` defaults to http://169.254.42.1/ (the Seed link-local USB Ethernet address)
iot register [endpoint] [--token TOKEN]
iot list
iot status <device-id>
iot pair <device-id>
iot unpair <device-id>
iot remove <device-id>

# Telemetry
iot ingest <device-id>
iot baseline <device-id> [--compute]
iot anomalies <device-id>
iot query <device-id> --vector "[1,2,3]" --k 10

# Fleet management
iot fleet create --name "my-fleet"
iot fleet list
iot fleet add <fleet-id> <device-id>
iot fleet remove <fleet-id> <device-id>
iot fleet delete <fleet-id>

# Firmware rollouts
iot firmware deploy <fleet-id> --version "2.0.0"
iot firmware advance <rollout-id>
iot firmware rollback <rollout-id>
iot firmware status <rollout-id>
iot firmware list

# Mesh & witness
iot mesh <device-id>
iot witness <device-id>
iot witness verify <device-id>
iot health <device-id>
iot trust <device-id>

Trust Model (5 Tiers)

Level Name Score Range Capabilities
0 UNKNOWN 0.0–0.19 Discovery only
1 REGISTERED 0.2–0.39 Status, identity queries
2 PROVISIONED 0.4–0.59 Telemetry ingest, vector store
3 CERTIFIED 0.6–0.79 Mesh participation, firmware deploy
4 FLEET_TRUSTED 0.8–1.0 Full fleet operations, witness signing

Trust Score Formula:

0.3×pairingIntegrity + 0.15×firmwareCurrency + 0.2×uptimeStability
+ 0.15×witnessIntegrity + 0.1×anomalyHistory + 0.1×meshParticipation

Anomaly Detection

Z-score composite scoring: min(1, meanZ/3)

Type Detection Rule Typical Cause
spike maxZ > 5 Sudden sensor failure
flatline all zero + low Z Sensor disconnected
drift 1-2 dimensions high Z Gradual calibration loss
oscillation alternating high/low Feedback loop
pattern-break moderate Z, multiple dims Environmental change
cluster-outlier >50% dimensions high Z Multi-sensor failure

Firmware Rollout State Machine

pending → canary → rolling → complete
                ↘ rolled-back ↙
  • canary: Deploy to ceil(deviceCount × canaryPercentage/100) devices
  • rolling: If canary anomaly score < rollback threshold, deploy to remaining
  • rolled-back: Force rollback triggered by anomaly threshold breach

Background Workers

Worker Interval Event
HealthProbeWorker 30s iot:device-offline
TelemetryIngestWorker 60s —
AnomalyScanWorker 120s iot:anomaly-detected
MeshSyncWorker 120s iot:mesh-partition
FirmwareWatchWorker 300s iot:firmware-mismatch
WitnessAuditWorker 600s iot:witness-gap

Integrations

  • AgentDB HNSW: Telemetry vectors stored in iot-telemetry namespace with HNSW indexing (M=16, efConstruction=200)
  • SONA Neural: Anomaly patterns fed to SONA for cross-device correlation and predictive maintenance
  • Cognitum SDK: @cognitum-one/sdk/seed SeedClient with 12 typed endpoints

Compatibility

  • CLI: pinned to @claude-flow/cli v3.6 major+minor.
  • Hardware: requires Cognitum Seed device. SDK: @cognitum-one/sdk/seed.
  • Verification: bash plugins/ruflo-iot-cognitum/scripts/smoke.sh is the contract.

Namespace coordination

This plugin owns five AgentDB namespaces, all compliant with the ruflo-agentdb ADR-0001 §"Namespace convention" (<plugin-stem>-<intent> kebab-case):

Namespace Purpose
iot-devices Device trust history per Cognitum Seed
iot-telemetry Telemetry vectors (HNSW: M=16, efConstruction=200)
iot-telemetry-anomalies Detected anomalies tagged by type + remedial action
iot-anomalies Skill-level anomaly index (alias of above)
iot-audit Witness-chain gap records

Reserved namespaces (pattern, claude-memories, default) MUST NOT be shadowed.

As a mod (0.3.2)

A function-hook mod ships beside the skills (ADR-445 pattern). Needs a Claude Code with mods (2.1.287+); older builds ignore it. No network, no process spawning: it only tightens calls to this plugin's own tools and reads through tools already connected.

Piece Default What it does
Write guard on Refuses a secret in an iot-* memory record, or on a cognitum-iot command line.
Destructive confirm on Refuses cognitum-iot fleet delete and device delete/remove/revoke/decommission/deregister unless the command has --confirm/--yes or the COGNITUM_IOT_CONFIRM=1 prefix (rollbacks and lists are untouched).
/iot-mod — status, scan <text>, devices (reads the iot-devices namespace through the connected memory tool); answered locally, no model call.
Status file — .claude-flow/iot-mod/status.json (version, updatedMs, mode flags and counters); written at session start and when a counter changes.

Options (userConfig): guard on|off, confirmDestructive on|off. Refusals never echo the value they matched.

claude plugin test plugins/ruflo-iot-cognitum   # 10 tests

Trust model parallel with federation

This plugin's 5-tier device trust model (UNKNOWN → REGISTERED → PROVISIONED → CERTIFIED → FLEET_TRUSTED) follows the same shape as the ruflo-federation 5-tier trust model (UNTRUSTED → VERIFIED → ATTESTED → TRUSTED → PRIVILEGED). Different surface (IoT devices vs federation peers) and distinct naming, but the score-driven progression and capability-gating principle are the same.

Verification

bash plugins/ruflo-iot-cognitum/scripts/smoke.sh
# Expected: "12 passed, 0 failed"

Architecture Decisions

  • ruflo-agentdb — HNSW-indexed telemetry storage backend; namespace convention owner
  • ruflo-federation — 5-tier trust model parallel (different surface, distinct naming, same shape)
  • ruflo-intelligence — SONA neural pattern learning
  • ruflo-observability — Telemetry correlation and tracing

License

MIT