Produces executed Jupyter notebook reports analyzing experiment runs, datasets, and system behavior.
npx skills add https://github.com/coollx/stable-harness --skill analysis-coollx
Please help me install this Agent Skill. Skill: analysis Source: https://github.com/coollx/stable-harness/tree/main/.claude/skills/analysis Command: npx skills add https://github.com/coollx/stable-harness --skill analysis-coollx
*_plan.md file when invoked with a run number. - Interview-first planning: For open-ended questions, it derives an analysis plan through batched questions and requires explicit researcher confirmation before writing code. - Executed notebook reports: Produces committed Jupyter notebooks with outputs (sources, question, method, findings, verdict) so researchers read rendered results without running cells. - Use Case: After run 183 finishes an ablation, invoke "/analysis 183" to execute its pre-registered plan and receive a committed notebook report with the verdict stated first. ## Quick Start Ask the assistant to analyze run 183 or pose an analysis question about your experiment data and confirm the proposed analysis plan.High-intent search queries and answers about installing and using this skill.
Invoke the analysis with the run number, such as "/analysis 183". The Skill reads that run's plan file and follows its pre-registered analysis steps, producing an executed notebook report under analyses/ named with the run's number and slug.
Cross-run analyses use a descriptive slug instead of a run number. The Skill first interviews you with batched questions to derive an analysis plan, then proceeds only after your explicit confirmation of the plan.
No. Analysis of literature is explicitly out of scope and belongs under the refs/ directory handled by the separate /ref skill. This Skill covers only runs, datasets, and system behavior from your own work.
The Skill stops and tells you rather than reconstructing numbers from memory or conversation. It enforces a no-pasted-numbers rule, so every reported figure must come from persisted data read by the report.
Notebooks are executed headlessly with jupyter nbconvert and committed with their outputs included. The researcher reads the rendered results directly and never needs to run cells manually.
K-Dense-AIEnforce timestamped preregistration of hypotheses and analysis plans before results.
daddiaEvaluate backlog.md for strategic alignment, acceptance criteria, dependencies, and sprint feasibility.
Tyler-R-KendrickCoordinates research loops linking training pipelines with curated documentation and Linear tracking.
NaKMiersRun report-only QA testing on web applications and generate structured defect reports.