Run evaluator-driven iterative improvement cycles until a bounded stop condition.
npx skills add https://github.com/nichobbs/lyric-lang --skill autoresearch-nichobbs
Please help me install this Agent Skill. Skill: autoresearch Source: https://github.com/nichobbs/lyric-lang/tree/main/.claude/skills/autoresearch Command: npx skills add https://github.com/nichobbs/lyric-lang --skill autoresearch-nichobbs
Autoresearch eliminates the friction of running repeated, evaluator-driven improvements by keeping a single mission moving through successive non-passing attempts until a strict time bound or explicit terminal condition is reached.
.omc/autoresearch/ or .omc/logs/ paths.Start a mission and evaluator first, then run autoresearch with a mission directory and max runtime so it can iteratively evaluate and log each attempt.
High-intent search queries and answers about installing and using this skill.
Automating iterative improvement is done by running repeated change cycles until a bounded stop condition is met. The loop records per-iteration results and stops at a strict max-runtime ceiling or explicit terminal condition.
Evaluator outputs require a structured JSON contract with a boolean pass field and an optional numeric score. This stable JSON schema allows the improvement loop to reliably assess each attempt and decide whether to continue iterating.
Bounded optimization persists durable state under canonical paths like `.omc/autoresearch/` or `.omc/logs/`. It saves per-iteration evaluation JSON alongside a human-readable markdown decision log to maintain state across each non-passing attempt.
Yes, you can use an automated refinement loop to tighten requirements, refine generated artifacts, or select better revisions. It iteratively executes one change cycle and evaluates the results against a required structured JSON contract until passing.
You should not use an iterative mission loop if your evaluator cannot produce stable structured JSON outputs with a boolean pass field. It also requires a strict max-runtime termination hook, making it unsuitable for unbounded or open-ended workflows.
jimmi2051Run evaluator-driven mission improvement loops with JSON logs and Markdown decision records.
xdkpIterate on a mission using evaluator-driven pass/fail JSON until max-runtime ends.
ZTE-AICloudExecute skills and append structured execution metrics to a JSON log.
EwallywIterate missions through evaluator contracts with JSON logging.