laxmi-narayan-87/harness-engineering
Practical workflows for designing, auditing, and evaluating reliable AI agent harnesses, with attention to state, verification, safety, and long-running coding tasks.
Use when designing repeatable evaluations for agent harnesses or skills, comparing prompt-only and harnessed workflows, assessing traces, or creating quality gates and regression tests.
Use when building bounded autonomous coding-agent loops, maker-checker workflows, multi-step orchestration, retries, checkpoints, or human-in-the-loop approval flows.
Use when reviewing an existing repository or agent workspace for harness quality, missing reliability controls, weak verification, unsafe autonomy, poor session continuity, or unsubstantiated completion claims.
Use when designing or improving a harness around an AI coding or research agent, including repository instructions, environment setup, persistent state, scope control, verification, observability, and clean session handoff.