Agent ops & evals
Plans an evaluation, reads a trace when another agent misbehaves, lays benchmark numbers into a table, and reviews an MCP connector before install. It tests other agents. It does not raise a score the run did not produce, and it does not declare a model the winner from a single example. 4 entries, listed by title.
B
- Benchmark table · listed — Lays out benchmark numbers the user already measured, with the case count and the metric they named.
E
- Eval case plan · listed — Writes evaluation cases for another agent, each with an input, an observable result, and a way to fail.
M
- MCP tool audit · listed — Reviews each tool on an MCP connector the user pastes, covering scope, writes, destructive actions, data egress, auth, and prompt-injection surface, then recommends keep, trim, or block.
T
- Trace debug note · listed — Reads a pasted agent trace and names the first step that diverged from the job.