菜单
Audit a past pruning or approval decision by creating a counterfactual branch from a saved snapshot without mutating the live graph. Use when the user asks what would have happened under another decision, disputes a paused branch, or wants to inspect an earlier checkpoint.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Rank competing hypotheses or proof skeletons from the complete comparison ledger using a joint batch MAP Bradley-Terry fit and approximate posterior intervals. Use whenever 3 or more candidates compete for the next experiment, or when the user asks which branch currently leads. Do not present the intervals as calibrated confidence bounds.
Diagnose errors, failed tests, crashes, hangs, regressions, suspicious outputs, and unexpected experimental results through reproducible hypothesis-driven debugging. Use whenever a script or system behaves incorrectly, even if the user only says it is broken or pastes an error. Respect diagnosis-only requests; implement a fix only when authorized.
Lock a confirmatory falsification target and its fixed multiple-comparison family before observing the confirmatory result. Use before promoting an exploratory finding to a main claim or whenever several related hypotheses need Bonferroni control. Records metric, threshold, family id/size, correction, and seed budget.
Build and audit a statistical or mathematical proof from proposition capture through skeleton selection, diagnosis, correction, optional empirical checking, and optional Lean reinsurance. Use when the user asks to prove or rigorously derive a proposition, a graph proposition lacks a verified proof, or a reviewer requires a theorem-side gate.
Run an end-to-end, auditable research workflow from question framing through literature, competing hypotheses, experiment selection, implementation, verification, and claim handoff. Use whenever the user asks to investigate, compare, test, validate, or establish an empirical research claim, including when they do not explicitly call it research. Do not use for a simple factual lookup or a narrowly scoped code change with no research claim.
Produce or revise research reports, result-bearing Markdown, paper sections, and manuscripts without overstating evidence. Use whenever writing text that reports experimental metrics, statistical conclusions, hypothesis rankings, theorem claims, or research findings. Do not trigger for ordinary README edits, changelogs, or prose with no research-result claim.
| name | replay |
| description | Audit a past pruning or approval decision by creating a counterfactual branch from a saved snapshot without mutating the live graph. Use when the user asks what would have happened under another decision, disputes a paused branch, or wants to inspect an earlier checkpoint. |
Replay is an inspection over frozen state. It does not execute new experiments and must not modify live graph nodes, ratings, judgments, or preregistrations.
mcp__memory__list_snapshots(limit=20). Show label, timestamp, and counts;
do not create a new snapshot merely to replay the past.Approve hyp_07 instead of pausing it.mcp__memory__replay_counterfactual(snapshot_id, counterfactual) and
retain the returned branch_id.seed_perturb, or the sequestered-data query tool.resume_branch; replay itself never flips status.Return:
snapshot_id: <snap_...>
branch_id: <replay_...>
counterfactual: <one sentence>
live_graph_mutated: false
evidence: <recorded facts used>
finding: <what changes under the counterfactual>
recommendation: <none or explicit user action>
limitations: <what replay cannot establish without a new experiment>
For several what-ifs, create one replay branch per counterfactual and keep each
output independent. Use mcp__memory__list_replay_branches(limit=20) to audit
existing branches before adding duplicates.
Replay is complete when the source snapshot and new branch are identifiable, the live graph is unchanged, every finding cites recorded snapshot evidence, and any proposed live change is clearly left for separate user approval.