Operating manual
Tag: Agent Operations
A focused reading path for Agent Operations: related field notes, evidence trails, and operating questions from the archive.
- an-agent-timeout-is-an-unknown-outcome.md
An Agent Timeout Is an Unknown Outcome
A timed-out tool call may already have committed. Give agent retries durable intent IDs, reconcile ambiguous outcomes, and test crash boundaries.
open artifact → - an-agent-cache-miss-can-start-in-the-tool-list.md
An Agent Cache Miss Can Start in the Tool List
Before upgrading models or extending cache lifetimes, test whether tool definitions and changing prompt prefixes are defeating reuse.
open artifact → - scoring-definition-is-the-lever-in-agent-evals.md
Scoring Definition Is the Lever in Agent Evals
A 24.0%-57.0% consensus swing on identical held-out data is a property of the scoring definition, not the agents. Lock the definition before reading the number.
open artifact → - agent-depth-trades-yield-for-cheaper-context.md
Agent Depth Trades Yield for Cheaper Context
Depth is not automatically cheaper context. Treating finding yield, root-context exposure, and dollars per finding as three independent axes exposes the real trade, and a per-tier alignment check prevents promoting another tier on paper numbers alone.
open artifact → - benchmark-agent-decisions-not-just-final-scores.md
Benchmark Agent Decisions, Not Just Final Scores
A final score hides whether an agent used evidence well. Evaluate the decision trace: context, intervention, consequence, validity, and budget.
open artifact → - the-last-wrong-step-is-not-where-the-agent-failure-started.md
The Last Wrong Step Is Not Where Failure Started
A four-stage operator model for tracing long-horizon agent errors from their initiating deviation through propagation, resolution, and terminal impact.
open artifact →