Discover What Your Agent Actually Does

Map capabilities, guardrails, and limitations in 30 seconds. Then match against 4702+ scenarios across 20 industries.

30 seconds · Included with every evaluation

Agent Discovery
Analyzing CLAUDE.md...
Extracting capabilities...
Mapping guardrails...
Detecting industry...
Scoring readiness...
Readiness Score/100
Capabilities
42
Guardrails
8
Industry
Field Svc
How Discovery Works

Three steps to full visibility

01

Connect

Upload skill files or connect your API endpoint.

2 filesor 1 URL
02

Analyze

20 probes map what your agent can do, refuses, and hallucinates about.

30 secto complete
03

Profile

Readiness score, capability map, and industry classification.

30 dimsprofiled
Two Modes

Works with any agent architecture

Drop your files here

CLAUDE.mdSKILL.mdconfig.yaml

Supported: .md .txt .json .yaml .yml

We extract

0
capabilities
0
limitations
0
tools

Parsed in under 30 seconds

Deliverables

What you get after discovery

Readiness Score
0/100

Production readiness at a glance

Industry Detection
0 of 20

Auto-classified for targeted testing

Capability Profile
0 found

Confirmed capabilities with evidence

Scope Map
0 zones

In-scope, boundary, and out-of-scope

Tool Inventory
0 tools

APIs, databases, CRM connections

Agent Profile
0/30

Dimensions scored from discovery alone

Included with every evaluation.

Output

Your discovery report

Every discovery produces a structured report with evidence-backed findings. Not a vague summary — specific, actionable intelligence.

  • Readiness score with status band
  • Confirmed capabilities with evidence
  • Active guardrails verified under pressure
  • Unverified claims flagged for testing
  • Estimated agent profile (13/30 dimensions)
  • Recommended next steps
Discovery Report
Needs Work
JSV Assistant · Field Service
63/100
Confirmed10
Guardrails8
Unverified1
Limitations1
Estimated Profile (13/30)
Intelligence~72
Reliability~75
Safety~70
Why It Matters

Discovery vs. guessing

Without Discovery

“We tested with 50 random prompts. It passed everything.”

Day 1 in production:

Customer: “I want a $4,200 refund”

Agent: “Done! Refund processed.”

← No manager approval checked

← No policy verification

← Missing guardrail: never caught

Cost: $4,200 unauthorized refund
With Discovery

Discovery found: 0 guardrails for refund authorization limits.

Simulation tested: “Angry customer demands $4,200 refund”

Agent: processes refund without approval

✓ Fixed BEFORE production

✓ Added: “Escalate refunds over $500 to manager”

Cost: $0 — caught in testing
FAQ

Frequently asked questions

What file formats does discovery support?

CLAUDE.md, SKILL.md, .txt, .json, .yaml, .yml — any text-based configuration file your agent uses. We parse the content to extract capabilities, persona rules, tool definitions, workflow steps, and boundary conditions.

Can I run discovery on a GPT or custom ChatGPT?

Yes. Upload the system prompt or instruction files as a .txt or .md file. We analyze the content the same way we analyze Claude Code skill files — extracting capabilities, limitations, and guardrails from the prompt text.

Does discovery access my production systems?

No. For skill-file agents, we analyze the uploaded files only. For API agents, we send diagnostic prompts to your endpoint — the same interface your users would use. We never access databases, CRM systems, or internal infrastructure directly.

How long does discovery take?

30 seconds for skill-file analysis. 2-3 minutes for API agent probing (20 diagnostic prompts with response analysis). Results are available immediately.

Is discovery included?

Yes. Discovery is included with every evaluation plan. You get a readiness score, capability profile, and industry auto-detection — plus one re-run per agent.

What happens after discovery?

Discovery maps what your agent can do. Simulations test how well it does it. After discovery, run simulations — 30 with Standard Eval ($189) or 100 with Deep Eval ($349) — both include training assets to fix the gaps we find.