Topics · AI Reliability Engineering
Each hub collects the tools, papers, terms and posts on one problem, with a reading path that keeps your place.
Hubs
Memory
Agents forget between sessions, and when several agents share one store, a single bad writer can poison what all of them recall. Memory needs a source for every fact and a way to stop trusting writers whose facts fail.
Evaluation
An agent's output legitimately changes from run to run, so a golden-file diff fails on correct runs and passes broken ones. Testing has to judge the distribution of runs, at a token cost a team can pay.
Loops
In most agent frameworks a step ends when the agent says it is done. A harness should instead prove that it terminates, stays within budget, and only reports done after an independent check passes.
Security
An agent skill is code that runs with the agent's own authority, often installed from a registry with no review. Skills need a threat model, static analysis, provenance checks and a bill of materials.
Contracts
A prompt asks an agent to behave; nothing checks that it did. Contracts state preconditions, invariants, governance policies and recovery, check them at runtime, and give each decision a typed answer and a receipt.