Profiling an agent harness with jq: latency per turn and byte-level request composition
arthurcolle · x · 2026-09-27
The author shares a set of jq commands to analyze a harness audit report (profile.), apparently for a model dubbed gpt-daybreak-blue-latest:
- Sort turns by residual (request span minus transport seconds) and list the 10 slowest, with fields like requestspanseconds, transportseconds, requestbytes, totalinputtokens and cachereadtokens;
- Compute min/max/median/p95 of per-turn request bytes;
- Parse lastrequestcomposition with --arg and break down the input byte composition as percentages of wire bytes: functioncalloutput, instructions, reasoning, functioncall, and tool schemas.
The takeaway: once agent harness request profiles are dumped as JSON, plain jq is enough to attribute latency — distinguishing slow transport from bloated request bodies (e.g., oversized tool outputs or reasoning payloads).
More from coding & agent
- Why AI agents may not go mainstream this decade: most lives aren't complex enough — menhguin · 2026-09-27
- A practical AGI test: replace one human validating 150,000 document conversions — burkov · 2026-09-27
- Talk on self-improving agents with reflective optimization: GEPA, gskill, fast-slow training — CShorten30 · 2026-09-27
- Microsoft president Jeff Teper shares how he uses Autopilot for work routines — heyneighbor · 2026-09-27
- Same prompt weekly for a quarter: how a prod agent's policy compliance drifted until it broke — IsomuraArganee_95 · 2026-09-27
- Diplomacy comes to Multi-Agent Arena: test your social strategy against frontier LLM agents — ycombinator · 2026-09-27