Evidence-Backed Research Reproduction

复制 Prompt

/goal
GOAL:
Complete Evidence-Backed Research Reproduction for a research task: Reproduce a paper or research result as far as local materials allow while separating confirmed findings, approximate reconstructions, blocked claims, and remaining uncertainty.

CONTEXT:
- Before editing, read the nearest AGENTS.md/CLAUDE.md, current issue or PLAN.md, and any failing logs already in the repo.
- Inspect source lists, citation notes, evidence files, and acceptance criteria.
- Establish a baseline by running or locating evidence for: `report with confirmed findings, approximate reconstructions, blocked claims, remaining uncertainty, and inspected outputs`.

CONSTRAINTS:
- Keep the scope limited to this goal; do not expand into unrelated cleanup.
- Do not weaken tests, delete assertions, or mask errors to make verification pass.
- Respect the repository's AGENTS.md/CLAUDE.md instructions and existing patterns.
- Do not present unsourced claims as facts.
- Keep direct quotes short and attach a public URL for every external claim.

DONE WHEN:
- The implementation or documentation directly satisfies: Reproduce a paper or research result as far as local materials allow while separating confirmed findings, approximate reconstructions, blocked claims, and remaining uncertainty.
- The verification command or evidence path succeeds: `report with confirmed findings, approximate reconstructions, blocked claims, remaining uncertainty, and inspected outputs`.
- The final diff is scoped to the relevant files and has no unrelated formatting churn.

VERIFY:
- Run `report with confirmed findings, approximate reconstructions, blocked claims, remaining uncertainty, and inspected outputs` or the closest repo-local equivalent if the exact command is not available.
- Capture before/after evidence for the behavior, metric, report, or artifact involved.
- If verification cannot run locally, stop and report the missing dependency instead of guessing success.

OUTPUT:
- Summarize changed files, key decisions, verification output, and remaining risks.
- Include any follow-up that is required for production rollout or human review.

STOP RULES:
- Pause if secrets, production access, stakeholder decisions, or destructive data operations are required.
- Pause after three failed fix attempts on the same symptom and challenge the root-cause hypothesis.
- Do not mark the goal complete until the current repository state has been audited against DONE WHEN.