Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add gyf9712/stat-theory-skills --skill proof-writergit clone --depth 1 https://github.com/gyf9712/stat-theory-skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/gyf9712/stat-theory-skills/proof-writer)<a href="https://agentmods.dev/skills/gyf9712/stat-theory-skills/proof-writer"><img src="https://agentmods.dev/badge/skills/gyf9712/stat-theory-skills/proof-writer/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/gyf9712/stat-theory-skills/proof-writer"><img src="https://agentmods.dev/badge/skills/gyf9712/stat-theory-skills/proof-writer.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00073 | $0.04022 |
| Opus 5 | $0.00036 | $0.02011 |
| Sonnet 5 | $0.00015 | $0.00804 |
| Haiku 4.5 | $0.00007 | $0.00402 |
Grade A, and why
proof-writer scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 12d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 320 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Proof Write: Rigorous Theorem / Lemma Drafting
🔬 Run this skill on Claude Opus. Writing rigorous proofs needs deep reasoning; run
/model opusfirst if your session is not on Opus.
Write a mathematically honest proof package, not a polished fake proof.
The unit of completion is the closed obligation, not prose. A proof is finished when every nontrivial obligation it raises is discharged into a typed closure object, not when it reads smoothly. This single idea drives the whole skill: it is what stops the model from quitting early (an open obligation is visible and costly) and from faking (you cannot prose your way into a typed closure).
Constants
- DEFAULT_PROOF_DOC =
PROOF_PACKAGE.mdin project root - STATUS =
PROVABLE AS STATED | PROVABLE AFTER WEAKENING / EXTRA ASSUMPTION | NOT CURRENTLY JUSTIFIED - VERIFICATION =
Verified | Conditionally verified | Gap found(defined in../stat-shared-references/proof-closure-machinery.md, single source of truth) - SCRIPT =
../stat-shared-references/scripts/proof_gap_scan.py - ASSUMPTION_LOCK =
assumptions.lock.md— the framework's shared assumption store (schema + discipline in../stat-shared-references/assumptions-lock-protocol.md). Theorems invoke assumptions by ID from here; they do not declare their own.
Context: $ARGUMENTS
Goal
Produce exactly one of three terminal outputs:
- a complete proof of the original claim, every obligation closed;
- a corrected (usually weaker) claim plus a complete proof of it;
- a blockage record stating precisely why the claim is not currently justified.
There is no fourth option. A sketch, a partial expansion, or a proof with a residual "open risk" footnote is not a terminal output.
Workflow
1. Gather context
Determine the target proof file by priority: (1) a path the user specified, (2) a
proof draft referenced in local notes or theorem files, (3) PROOF_PACKAGE.md.
Read the target if it exists, plus theorem notes and any files the user names.
Extract the exact claim, assumptions, notation, any user sketch, and nearby lemmas
the draft depends on. If notation or assumptions are ambiguous, state the exact
interpretation you will use before proving anything.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 12d ago First seen · 320 lines · 73 tokens per session scan A 3c4228c341a5
proof-writer is a skill published in the GitHub repository gyf9712/stat-theory-skills (9 stars, last pushed 1mo ago), licensed MIT. It adds 73 tokens to every session and 4,022 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
did
Design, estimate, validate, and write up a difference-in-differences analysis, using the post-2018 heterogeneity-robust toolkit with an explicit statement of which parallel-trends assumption is imposed. TRIGGER on "difference-in-differences", "DiD", "TWFE", "event study", "staggered adoption", "parallel trends"…
iv
Design, estimate, validate, and write up an instrumental-variables analysis: single-instrument LATE designs, weak-instrument-robust inference, shift-share instruments, formula instruments that need recentering, and leniency (judge and examiner) designs estimated by UJIVE. TRIGGER on "instrumental variable", "IV"…
conjoint
Design, analyze, and write up conjoint experiments in both traditions: as randomized experiments identifying average marginal component effects, and as preference-measurement instruments (hierarchical Bayes partworths, WTP, choice-share simulation), with measurement-error correction, multiple-testing correction, and a…
research-talk
Author a Quarto reveal.js deck for a research talk: seminar, conference, job talk, brown bag. TRIGGER on "slides for my talk", "seminar deck", "job talk", "turn this paper into slides", "add an appendix slide", "export a handout". Classroom material is teaching-lecture; auditing an existing deck is slide-review.
field-experiment
Design, analyze, and write up randomized experiments (field experiments, A/B tests, RCTs): stratified and clustered designs, randomization inference, covariate adjustment done right, noncompliance, attrition and gated outcomes, treatment-effect heterogeneity, and interference. TRIGGER on "A/B test", "randomized…
preregister
Draft a registry-ready preregistration or preanalysis plan for AsPredicted, OSF, or the AEA RCT Registry. TRIGGER on "preregister", "draft a preregistration", "AsPredicted form", "OSF prereg", "AEA RCT registry", "PAP", "preanalysis plan", or before launching a vignette, conjoint, MTurk/Prolific, or field experiment.…