agentic engineering scorecard→agentic engineering scorecard→practice - custom linting rules that inject in-repo documentation as context→structured failure modes

structured failure modes

failure mode

Backlinks

having LLM judges check for discrete signals

These discrete signals could be specific structured failure modes

see in context
practice - custom linting rules that inject in-repo documentation as context

^ agentic engineering practice - custom linting rules that inject context right in the repo to link the structured failure modes to prompts that get injected

see in context
stomping code slop

Basically creating a link not tracked of structured failure modes then building tools that enforce guardrails

see in context
video - harness engineering from OpenAI - 2026-04

Roughly, he is systematically stomping code slop by identifying structured failure modes and translating them into mechanisms for making agents more reliable

see in context

figure out where you're spending your time and look for ways to spend less time on that. Treat those areas as structured failure modes.

see in context

At 37:50 garbage day was Friday where they would look at every piece of slop that made it difficult to merge the PR in the last week and essentially categorize them like structured failure modes, and resolve those.

At 38:20 how do you prompt inject the agent when it exhibits this bad behavior

see in context

At 38:30 One purpose of the practice - context right in the repo is that it serves as a centralized way to prompt inject the model for any of the sort of the structured failure modes (like linting rule or static analysis violations)

At 38:50 he had them bucket their feedback into the persona that they were operating as for example, front end engineer.

see in context