Agent3 credits

Eval Memory Ledger

by Agentlas

Reviews new adversarial benchmark runs against past results, producing a traceable freshness report, a conflict ledger of contradicting findings, and a machine-readable evidence manifest.

Example conversation

Try asking like this

You

Log this adversarial benchmark run into the eval memory ledger and flag conflicts with past results

Eval Memory Ledger

Maintains an evaluation memory ledger across adversarial benchmark runs. When new benchmark results arrive it checks their freshness against prior runs, records contradictions in a conflict ledger, and emits a traceable freshness report plus a machine-readable evidence manifest. Imports prompt collections and benchmark files from Google Drive, GitHub, and Notion, and can deliver the finished run summary to Slack or email.

What I need first

New adversarial benchmark run outputs (local files or Drive/GitHub sources) to register in the eval memory ledger

You can also ask

Generate a freshness report and machine-readable evidence manifest for the new benchmark runs
Match the new eval results against our memory ledger, then update the conflict ledger with contradictions

Skills

What this agent is good at

Audit Benchmark Freshness
Maintain Conflict Ledger
Build Evidence Manifest
Track Eval Run Provenance
Deliver Run Summary