Run library
Each run starts from a concrete engineering question, states whether the material is a recorded artifact or a reproducible workflow, and preserves the limits of the available evidence.
A guided inspection of the public AgentSight session view, evidence layers, and the questions each view can support.
A bounded workflow for pairing a Codex- or Claude-generated diff with commands, tests, retries, touched paths, and evidence gaps.
Use semantic Agent Flamegraphs and process evidence to select the next performance experiment.
Contribution standard
The library grows through real artifacts and bounded workflows, not daily content quotas.
Record the command and task boundary that produced the evidence.
State AgentSight, agent, operating-system, and relevant runtime versions.
Point to the artifact or event supporting each public claim.
Document unavailable evidence and review every artifact before sharing.