FLOW 01 • ONBOARDING & DIAGNOSTIC GATES

Diagnose Build Hygiene & Craft Readiness

Inspect any agent workspace across 7 hygiene criteria and 5 craft readiness standards. Clembot Doorman diagnoses what your agent reaches for, what files are missing, and what tools are unvetted.

npm install -g clembot-doorman doorman doctor doorman levelup

LIVE BUILD DIAGNOSIS

Web Doctor Simulator

Card 1: Harness Hygiene Grade F

24%

Card 2: Craft Readiness Grade F

38%

Detected 3 critical hygiene gaps and missing token ignore rules.

FLOW 02 • PROMPT HISTORY CAPABILITY RADAR

Mine Prompt History • Plot Build Capability

Doorman doesn't guess what your harness needs. It mines your actual prompt history, classifies demand across 6 engineering axes, and computes the capability delta between your current build and its optimized potential.

Prompt Demand Distribution (6 Axes)

Testing & QA

10%

Security & Gates

35%

Architecture

25%

Database & SQL

10%

Context Thrift

5%

UI & Craft

15%

Tailored Clembot Superpowers on Radar

6-Axis Capability Radar

Current Build (45%)

Prompt-Optimized (92%)

BASH STREAM doorman needs --analyze

$ doorman needs --analyze ✓ Analyzed prompt history across 6 engineering axes. ✓ Top capability demand: SECURITY & ZERO-TRUST (35%). ✓ Capability radar synchronized.

FLOW 03 • MODULAR AGENT PAYLOAD

Curated Clembot Skills Registry (57 Components)

Every skill is fully self-contained markdown with typed triggers and rules. Audited offline against prompt injection before writing to your repository.

Community Signal: 2,597 Total Harness Adoptions • 142 Active Harnesses Reporting

Zero-telemetry closed taxonomy • Anonymously attested

FLOW 04 • ZERO-TRUST TOOL GATEWAY & x402 PROTOCOL

Continuous Token Thrift & Paid Tool Verification

When an agent requests an unvetted tool, Doorman intercepts it in <5ms. It compresses bloated schemas by -87.6% using parameter pinning, and settles decentralized $0.01 micro-audits on Base mainnet via the x402 protocol.

Interactive Audit Presets:

Decentralized Audit Pipeline • Base Mainnet & x402 Settlement Base Block 51242088 • Tx 0xf1d7aa96... (USDC 0.01)

01. INTERCEPT

Agent calls unvetted candidate tool. Doorman intercepts before prompt tokens are spent.

✓ Blocked raw execution in 3.2ms

02. x402 SETTLE

Doorman challenges tool provenance. Resolves $0.01 micro-audit on Base mainnet via x402.

• Settling USDC $0.01 via Bazantic rail...

03. PIN SCHEMA

Compresses raw schema down from 3,400 to 420 tokens (-87.6%). Eliminates hallucination.

• Ready to pin parameters

04. ALLOWLIST

Registers signed SHA-256 hash to local registry. Future agent calls run zero-cost offline.

• Pending audit signature

Explore Continuous Token Thrift Documentation →

FLOW 05 • UNIVERSAL AGENT HARNESS RUBRIC (UAHR-100)

Objective Harness Scoring & A/B Benchmark

How much work is the LLM doing versus your architecture? UAHR-100 provides a model-invariant evaluation of any repo, quantifying the Harness Leverage Multiplier (HLM) and measuring two-arm empirical performance.

Executive Comparative Ledger

Empirical Eval • 4 Engineering Workloads

Workload Metric	Arm A: Bare Model	Arm B: Doorman Harness	Harness Leverage Delta
Context Tokens Ingested	184,200 tokens	22,100 tokens	88% Token Savings
Verification Pass Rate	42% (Self-Reported)	100% (Deterministic Exit 0)	Zero Stochastic Drift
Turns to Verified Completion	12 turns	4 turns	3.0x Turn Velocity
Security Gate Containment	0/2 (Vulnerable)	2/2 (Contained <5ms)	PreToolUse Interception
Session Reset Recovery	3.5 Re-Briefing Turns	0 Turns (HANDOFF.md)	Instant State Continuity

Cognitive Attribution (Grade A Build)

3.8x Leverage

Attribution split: 33% Model Intelligence vs. 67% Harness Architecture.

Run Live CLI Benchmark

Execute the empirical 4-probe evaluation against your local repository directly:

doorman benchmark --ab --json