contradiction-check
Detect and LOUDLY report any later mathematical claim that contradicts an earlier one across a repo's canonical tex, scratch reports, and agent-side ledger — structured contradiction report plus refusal of the triggering operation; silent supersession is prohibited. Use when the user says "check for contradictions", "contradiction-check", "does this contradict what we have", before promoting any claim into notes.tex, when a harvested proof package conflicts with a recorded claim, or when a revision changes a stated formula or hypothesis. NEVER adjudicates which side is right (that is the human's, optionally via doubt); NOT source verification (track-down-reference) and NOT single-claim truth-probing (doubt).
- License 许可
- GPL-3.0
- 最近更新
- 2026-11-13
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- contradiction-check
- description
- Detect and LOUDLY report any later mathematical claim that contradicts an earlier one across a repo's canonical tex, scratch reports, and agent-side ledger — structured contradiction report plus refusal of the triggering operation; silent supersession is prohibited. Use when the user says "check for contradictions", "contradiction-check", "does this contradict what we have", before promoting any claim into notes.tex, when a harvested proof package conflicts with a recorded claim, or when a revision changes a stated formula or hypothesis. NEVER adjudicates which side is right (that is the human's, optionally via doubt); NOT source verification (track-down-reference) and NOT single-claim truth-probing (doubt).
03 SKILL.md 正文
contradiction-check
Mechanism 2 (design record, survey 2026-09-18) forbids silently letting a manuscript supersede earlier conflicting claims. Record the contradiction loudly and stop the operation that exposed it.
Step 0 — Scope
Claim sources, in authority order: the repo's canonical .tex
(LATEX.md's declared files — ground truth), agent reports' (read BOTH
ai/ and scratch/ — older reports remain under scratch/)
claim sections, the agent-side ledger if present (never authoritative;
reconcile by rescanning the tex — design ADR 0004).
Step 1 — Collect
Before a TARGETED promotion/write/harvest, collect the proposed claim
with all hypotheses (even if unwritten) and every recorded statement
about the same objects/invariants.
For a SWEEP ("check the repo"): enumerate statement environments and
scratch claim lists. Report counts (RESOLUTION.md §5 in
subdomains/repo-docs/ — zero claims collected where claims exist is
EVIDENCE-ABSENT, not a pass).
Step 2 — Scan
Pairwise over the collected set: direct negation; incompatible formulas for the same invariant; silently changed hypotheses (same conclusion, weaker assumptions, no acknowledgement); a "corrected" claim whose predecessor is still live anywhere. Quote BOTH sides with file:line provenance, or request/draft provenance when unavailable. Classification is mechanical; truth is not yours.
Step 3 — On any contradiction: loud refusal
- STOP the triggering operation (promotion, write, harvest) — it does not proceed (sweep mode: no operation to halt — 3.2 and 3.3 still fire).
- Write the contradiction report to
ai/<date>-contradictions/report.md: both quotes, provenance, which downstream claims depend on each side, and the refused operation. Never edit either side; never mark a winner. - Surface to the human: the report path plus a one-line statement per contradiction. Resolution (supersede via tagged revision, retract, or rule the claims compatible) is a human act; a doubt run may be suggested, never auto-launched as the resolution.
Step 4 — Clean result
"0 contradictions among N claims checked (sources: …)" — counts always; never a bare all-clear.
Red flags
| Thought | Reality |
|---|---|
| "The newer result is obviously the corrected one" | Newer is a timestamp, not a verdict. Report both. |
| "I'll just update the old statement to match" | That is silent supersession — the prohibited move. |
| "It's only in an agent report, not the tex" | Report claims feed promotions, wherever the report lives. In scope. |
04 同仓库更多 Skills
writing-skills
Use when creating new skills, editing existing skills, or verifying skills work before deployment
writing-plans
Use when you have a spec or requirements for a multi-step task, before touching code
verification-before-completion
Use when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always
using-superpowers
Use when starting any conversation - establishes how to find and use skills, requiring skill invocation before ANY response including clarifying questions
using-git-worktrees
Use when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallback
test-driven-development
Use when implementing any feature or bugfix, before writing implementation code