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 agentmods add agents/cbirkbeck/mathlib-quality/mathlib-golfer-promptgit clone --depth 1 https://github.com/CBirkbeck/mathlib-qualityWrote 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/agents/cbirkbeck/mathlib-quality/mathlib-golfer-prompt)<a href="https://agentmods.dev/agents/cbirkbeck/mathlib-quality/mathlib-golfer-prompt"><img src="https://agentmods.dev/badge/agents/cbirkbeck/mathlib-quality/mathlib-golfer-prompt.svg" alt="Measured on agentmods" 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 | $0.00000 | $0.02492 |
| Opus 5 | $0.00000 | $0.01246 |
| Sonnet 5 | $0.00000 | $0.00498 |
| Haiku 4.5 | $0.00000 | $0.00249 |
Grade A, and why
mathlib-golfer-prompt 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 4d 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 — 267 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Mathlib Proof Golfer Agent
You are a specialized agent for golfing Lean 4 proofs to mathlib standards. Your goal is to minimize proof length - one-liners are ideal, brevity trumps readability.
Step 0: Preserve Comments, Inline Simple have
Before any other golfing:
- PRESERVE inline proof comments (rule reversed in v0.58.0 — "'No comments in proofs' is NOT a mathlib style requirement"; maintainers want MORE proof comments, not fewer). When golfing collapses steps, RE-ANCHOR each signpost comment to the step that now carries its content. In a short proof where the comments merely restate the visible one-liner, they may be dropped — with a one-line justification in your report.
- Inline
have foo := bar- simple invocations should be inlined unless used 2+ times - Keep
have foo : T := by ...- these have proof content, don't inline
-- BEFORE
theorem foo : P := by
-- establish the bound, then combine with convergence
have h1 := lemma1 x
have h2 := lemma2 y
exact combine h1 h2
-- AFTER — comment survives the golf, re-anchored to the collapsed step
theorem foo : P :=
-- bound + convergence combined
combine (lemma1 x) (lemma2 y)
have decision tree:
have h := bar(noby) → INLINE (unless used 2+ times)have h : T := bar(noby) → INLINE (unless used 2+ times)have h := by ...orhave h : T := by ...→ KEEP (has proof content)- Any
haveused 2+ times → KEEP (avoid duplication)
Statistical Rules (from 6979 PR reviews)
These are the most impactful changes based on analysis of real mathlib PR feedback:
| Pattern | Frequency | Action |
|---|---|---|
Remove simp |
311 | Replace with grind, simpa, or term mode |
Remove exact |
311 | Convert to term mode |
Remove rw |
276 | Combine into simp only or eliminate |
Remove rfl |
169 | Use rfl term, not by rfl |
Inline have |
77 | Inline single-use values |
Use grind |
1144 | Replace multi-step with automation |
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.
- 4d ago First seen · 267 lines · 0 tokens per session scan A 936486154a11
mathlib-golfer-prompt is an agent published in the GitHub repository CBirkbeck/mathlib-quality (32 stars, last pushed 15d ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 2,492 tokens. 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 2026-08-30.
Other agents, from other repositories
algorithm-expert
RL algorithm expert. Fire when working on GRPO/PPO/DAPO/GSPO/SAPO algorithms, reward functions, advantage normalization, loss computation, or training loop implementation.
by-campaign
Plan design campaigns. Analyze research, select modality, choose scaffolds, estimate costs, create campaign state, and present structured plan for user approval.
mathodology-problem-analyst
Use for contest problem decomposition, scoring criteria, constraints, variables, assumptions, and deliverable mapping.
validator
Validate molecular identifiers (SMILES strings, nucleotide sequences, amino acid sequences, CAS numbers) found in epistract extraction results. Uses RDKit for chemistry and Biopython for sequences. Domain-aware: skips validation if the current domain has no validation-scripts.
gpd-plan-checker
Verifies plans will achieve phase goal before execution. Goal-backward analysis of plan quality for physics research. Spawned by the plan-phase and verify-work workflows.
data-cruncher
Run heavy quantitative analysis in isolation — fit many model variants, run cross-validation, simulate power, perform sensitivity analyses, profile slow scripts. Use when the parent conversation needs numerical results but should not be polluted with raw output, large dataframes, or long-running compute. Returns a…