Skip to content

sidiangongyuan/codex-skills-library / skills/experiment-planner

experiment-planner

Verified

Use when exploring, continuing, or accepting a deep-learning or computer-science research idea. Converts claims into pilot-first experiment matrices covering diagnostics, failure analysis, authorized execution, evaluation, and paper-story viability.

FileKindSize
LICENSEother1.0 KB
SKILL.mdskill md12.5 KB
agents/openai.yamlconfig276 B
references/experiment-matrix.mdmarkdown9.2 KB
references/source-map.mdmarkdown3.9 KB

references/source-map.md

View on GitHub
# Experiment Planner Source Map

This skill is a local Codex adapter. Use these upstream projects as pinned
references for workflow ideas, not as installed dependencies or copied text.

## Pinned References

| Source | Pinned commit | Reference paths | Use in this skill | Not adopted in v1 |
|---|---:|---|---|---|
| modelscope/Awesome-Vibe-Research | `3b13292b7653b12cc6faa37a17fc69fdf1b992e2` | `README.md` | Project map for separating direction scanning, method design, experiment execution, and writing. | No direct install or automatic orchestration. |
| wanshuiyin/Auto-claude-code-research-in-sleep | `dd1ee32dad5fd9a772eea0c5d0553cd938edcbf7` | `skills/idea-discovery/SKILL.md`, `skills/experiment-plan/SKILL.md`, `skills/experiment-bridge/SKILL.md` | Main reference for literature survey, idea refinement, novelty check, claim-driven experiment planning, pilot-first execution, and bridging plan to code. | No full ARIS install, no default report files, no automatic overnight/GPU deployment. |
| Imbad0202/academic-research-skills-codex | `36cc610fca8e3935904c6c175e3a0a05c7d9b8c6` | `skills/academic-research-suite/ars/experiment-agent/WORKFLOW.md` | Safety and reproducibility boundaries: no automatic code changes, no automatic retries, cautious statistical interpretation. | No suite install and no replacement of local writing/review/rebuttal skills. |
| Just-Curieous/Curie | `db1b1f56159b591515f77e03c55bf473d5c1c201` | `README.md`, workflow descriptions | Hypothesis to implementation to execution to analysis to reflection loop. | No Docker/API/key-based agent framework integration. |
| karpathy/autoresearch | `228791fb499afffb54b46200aca536f79142f117` | `README.md` | Short-loop experimental discipline: fixed question, fixed metric, small validation before expansion. | No nanochat-specific assumptions or single-project runner adoption. |

## Adaptation Principles

- Prefer ARIS-style flow for research ideation: survey, refine, novelty check,
  experiment planning, then pilot validation.
- Prefer ARIS experiment-plan style for claim-driven roadmaps: define the claim,
  anchor result, novelty isolation, simplicity check, failure analysis, and run
  order before coding.
- Prefer ARIS experiment-bridge style only as a coordination idea: plan to code
  via explicit tasks, sanity checks, and controlled execution. Keep Codex main
  session in charge of acceptance.
- Prefer Academic Research Skills Codex safety rules when execution pressure is
  high: do not silently broaden user authorization, retry outside an authorized
  bounded same-config recovery, or overstate statistical meaning. Within a
  still-valid user-approved scope,
  ordinary implementation, local bug fixes, queueing in-scope runs, and
  evaluation may proceed; new families, datasets, seed policies, and paper-only
  TODOs remain explicit opt-ins.
- Prefer Curie/autoresearch only for the research loop and experiment discipline,
  not for their runtime frameworks.

## Local Adapter Rules

The pinned sources provide provenance and workflow ideas; they do not override
the local experiment-planner contract. For continuations, reuse settled claim,
baseline, metric, protocol, budget, seed, and literature decisions instead of
restarting the full exploration checklist. Use impact-based validation with
reusable compact pass receipts, avoid routine hash gates and endless CPU gates,
freeze acceptance before outcomes, and distinguish engineering, optimization,
prototype, and core-construct failures. Diagnostic gates localize evidence but
are not the whole scientific verdict.

## Upgrade Rule

When intentionally updating this skill from upstream projects, first compare the
pinned commit against the new upstream state. A local workflow-policy update
does not change the upstream pins or claim upstream support. Only change a pin
when provenance actually changes, summarize what changed, and keep local
behavior conservative unless the user approves heavier automation.
add-product-e2e-evalpaperclipai/paperclipAdd or extend a Paperclip full-stack runner E2E workflow, fixture, matcher, or report evidence path for local or Daytona execution.Stars: 115kUpdated Claude CodeCodexVerifiedInstall
add-runner-evalpaperclipai/paperclipAdd or extend a Paperclip Runner protocol evaluation definition, roster, assertion, or report fixture with provenance and narrow validation.Stars: 115kUpdated Claude CodeCodexVerifiedInstall
diagnose-why-work-stoppedpaperclipai/paperclipCompatibility entry point for failed, stalled, looping, or over-recovered Paperclip work. Use fix-failed-paperclip-run for the full diagnosis, regression, repair, real verification, and green-PR workflow; honor an explicitly requested diagnosis-only scope.Stars: 115kUpdated Claude CodeCodexVerifiedInstall
fix-failed-paperclip-runpaperclipai/paperclipDiagnose and fix failed, stuck, slow, or incorrectly blocked Paperclip tasks and runs. Use with a task/run URL or failure report to trace the root cause, prove a regression red then green, check second-order effects, verify the real workflow, and deliver a PR with passing checks.Stars: 115kUpdated Claude CodeCodexVerifiedInstall