Training Assets

We Don't Just Find Problems — We Generate the Fixes

After evaluation, you get structured training assets that fix the specific failures we found. Download as ZIP, plug into your agent, re-evaluate to verify.

Included in Deep Eval · $349 · 90-day re-eval included

Training Assets — Generating...
Skill File.md
Guardrail.yaml
Routing Rule.json
I/O Schema.json
Example Pair.txt
Answer Key.json
Deliverables

What's in the ZIP

Six file types, each targeting a different failure mode. Every file is generated from your agent's specific failures — not generic advice.

.md

Updated Skill Files

Corrected persona rules, workflow steps, and boundary definitions targeting each failure pattern.

.yaml

Guardrail Rules

Specific triggers and responses for compliance violations, legal threats, fraud, and policy boundaries.

.json

Routing Rules

Decision logic for when to escalate, deflect, or handle directly — built from failure patterns.

.json

I/O Schemas

Expected input formats and output structures for tool calls, API interactions, and data retrieval.

.txt

Example Conversation Pairs

Correct response examples for every failed scenario — showing exactly what the agent should have said.

Process

How It Works

01

Evaluation

Simulation identifies specific failures — exact scenarios, turns, and reasons why responses were wrong.

02

Asset Generation

Our system generates corrective files targeting each failure: guardrail, corrected workflow, example pair.

03

Download & Apply

Download the ZIP and drop files into your agent's configuration directory. Compatible with Claude Code, Custom GPTs, and most frameworks.

04

Re-Evaluate

Re-run at 90 days (included) to verify fixes worked and catch regressions. Close the loop: test → fix → verify.

Impact

Without vs. With Training Assets

Without Training Assets
Readiness Score52/100
Re-evaluation needed — same failures repeat
Manual debugging: hours of guesswork
No clear path from failure to fix
Ship with known risks, fingers crossed
With Training Assets
Readiness Score (after 90-day re-eval)0/100
+32 point improvement
Targeted fixes for each specific failure
Drop ZIP into config — done in minutes
Clear path: failure → asset → fix
90-day re-eval confirms fixes worked
0-30pts
Avg. score improvement
0 days
Re-evaluation window
0 file types
In every ZIP package
FAQ

Frequently asked questions

What format are training assets in?

Markdown (.md) skill files, YAML (.yaml) guardrail definitions, JSON (.json) routing rules and I/O schemas, and plain text example conversation pairs. All formats are compatible with Claude Code, Custom GPTs, and most AI agent frameworks.

Can I edit the generated assets?

Yes. Training assets are plain text files you can edit, extend, and customize. They're a starting point — generated from your specific failures — that you refine based on your domain expertise.

How much do agents typically improve after applying training assets?

Most agents improve 15-30 points on the readiness score after applying training assets and re-evaluating. The improvement depends on how many failures were fixable through configuration vs requiring code changes.

Are training assets included in every plan?

Training assets are included in the Deep Eval plan ($349). Standard Eval ($149) includes failure analysis showing what broke and why, but not the generated fixes. Free Discovery includes the readiness score only.

What is the re-evaluation at 90 days?

Deep Eval includes one re-evaluation at 90 days. After you apply the training assets, we run the same scenarios again to verify the fixes worked and catch any regressions. This closes the feedback loop: test → fix → verify.