Adversarial Verification
An automatic second look at every deliverable, before it ships.
When an agent produces a result -- code, a document, an analysis -- how do you ensure its quality? Manual review is expensive and often skipped. The /verify skill automates this critical step.
By invoking /verify, an independent verification agent reviews the deliverable with an adversarial eye. It actively looks for errors, omissions, inconsistencies, and edge cases. The result is a structured verdict: PASS, FAIL, or PARTIAL.
This process doesn't replace human validation, but it prepares it. When a deliverable reaches a human's desk, obvious problems have already been caught and fixed.
Invocation
The primary agent completes its work and launches /verify on the result, or the skill is automatically triggered via a post-execution hook.
Isolation
A separate verification agent is instantiated. It has read-only access to the deliverable and its context.
Adversarial analysis
The verification agent actively looks for: logical errors, unhandled edge cases, inconsistencies with requirements, security issues.
Verdict
A structured report is produced: PASS / FAIL / PARTIAL, with detailed justification for each point.
Correction
On FAIL or PARTIAL, the primary agent receives the report and can fix issues before delivery.
Code generation
An agent writes a complex function. /verify catches an unhandled edge case (division by zero, empty list) before the code is committed.
Content writing
An agent produces a business document. /verify identifies an inconsistency between two quoted figures and an inappropriate tone in a paragraph.
System configuration
An agent generates a config file. /verify spots an already-used port and a missing environment variable.
featurePages.verify.config
- Available for all EasyClaw v2 agents
- Usable on demand (/verify) or in automatic mode
- Recommended in combination with the Verification Agent for critical tasks