Iterate on a mission using evaluator-driven pass/fail JSON until max-runtime ends.
npx skills add https://github.com/xdkp/oh-my-claudecode --skill autoresearch-xdkp
Please help me install this Agent Skill. Skill: autoresearch Source: https://github.com/xdkp/oh-my-claudecode/tree/main/skills/autoresearch Command: npx skills add https://github.com/xdkp/oh-my-claudecode --skill autoresearch-xdkp
Autoresearch solves the problem of getting reliable, step-by-step improvement for one mission without losing control over evaluation quality or run duration.
pass (and optional numeric score), and persists iteration-by-iteration logs and decisions under .omc/autoresearch/<mission-slug>/.max-runtime, while also supporting explicit terminal conditions and cancellation, so runs don’t drift indefinitely./deep-interview --autoresearch, use this skill to continuously refine outputs based on pass/fail evaluation and keep a human-readable decision log.In your Claude Code session, run /deep-interview --autoresearch first, then activate autoresearch with your mission so it begins iterating and writing evaluation JSON plus markdown decision logs.
High-intent search queries and answers about installing and using this skill.
To run a stateful improvement loop safely, you need a bounded workflow with strict stop rules. This skill executes a single-mission loop that stops only on explicit terminal conditions like max-runtime, preventing indefinite drift while persisting iteration logs.
The evaluator output must be structured JSON containing a required boolean `pass` field and an optional numeric `score`. This strict contract allows the loop to programmatically determine pass/fail conditions and persist machine-readable evaluations for later analysis.
You must first run the `/deep-interview --autoresearch` command to prepare your mission and evaluator setup. Once configured, you can activate the improvement loop to continuously refine outputs based on the pass/fail evaluation contract.
Iteration logs and evaluation decisions are persisted as durable artifacts under the `.omc/autoresearch/<mission-slug>/` directory. This stores both machine-readable evaluation JSON and human-readable markdown decision logs for later analysis.
No, this workflow runs one mission at a time to ensure reliable, step-by-step improvement without losing control over evaluation quality. It applies experiment cycles to a single mission until an explicit stop condition is met.
Strict stop rules and max-runtime bounding prevent automated research loops from drifting indefinitely. Bounded runtime ensures predictable stopping while still supporting explicit terminal conditions and cancellation when needed.
xz1220Run iterative improvement loops with JSON evaluation and decision logs.
jimmi2051Run evaluator-driven mission improvement loops with JSON logs and Markdown decision records.
solitude6060Iterate a single mission through evaluator-gated experiments until terminal conditions.
Leap0920Iteratively improves a task until a bounded stop condition using JSON evaluator feedback.