Plans and executes research experiment runs with numbered run folders and launch commands.
npx skills add https://github.com/coollx/stable-harness --skill experiment-coollx
Please help me install this Agent Skill. Skill: experiment Source: https://github.com/coollx/stable-harness/tree/main/.claude/skills/experiment Command: npx skills add https://github.com/coollx/stable-harness --skill experiment-coollx
High-intent search queries and answers about installing and using this skill.
Invoke the plan mode, which interviews you in batches of three questions with recommended answers, allocates the next global run number, writes run.yaml with overrides only, and produces a plan document ending with the exact launch command.
For runs expected to exceed about one hour, execute mode runs a smoke pass first — the full workflow on fewer than three instances into outputs/smoke/. A failed smoke means fix the pipeline before launching the real run.
No. The skill hands the researcher the launch command and never auto-launches long jobs. Short runs may be run directly, and once the researcher launches, the skill has standing authorization to recover transient failures.
The skill monitors outputs/run.log with failure-signature searches, since silence is not success. Transient failures are recovered, each incident is recorded as a dated paragraph in the run log, and long pipelines resume via their volume argument.
On completion the skill confirms summary.json landed and commits it with the run's spine line. A discarded run commits nothing, keeping the git history clean of abandoned experiments.
K-Dense-AIConverts vague research ideas into precise, falsifiable framing documents with hypotheses and success criteria.
methodic-researchEvaluates Chronicle experiment runs against pre-registered hypotheses using real metrics.
SilentEchoeCommit and push code changes to a Git repository with conventional commit messages.
dominodatalabTrack MLflow experiments in Domino Data Lab with auto-logging.
