Agent skill

Vero Source Lean

by sunblaze-ucb in sunblaze-ucb/vero

Load BEFORE curating an existing Lean 4 source repo into the benchmark format.

Apache-2.0Auto-check passedWriting & Content

Install Vero Source Lean

skills CLI
$ npx skills add sunblaze-ucb/vero --skill vero-source-lean -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install sunblaze-ucb/vero vero-source-lean --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/sunblaze-ucb/vero.git skills-src && mkdir -p .claude/skills && cp -r skills-src/.claude/skills/vero-source-lean .claude/skills/vero-source-lean && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
vero-source-lean
GitHub stars
107
Token cost
~2k tokens
SKILL.md length
929 words
Files
1
Skills in repo
16
Repo updated
First seen
Licence
Apache-2.0

At a glance

Load BEFORE curating an existing Lean 4 source repo into the benchmark format.

  • Works in 3 steps: APIs (Impl/.lean): the source's… → Specs (Spec/.lean): the source's… → Bundle / Harness / Test: standard,…
  • Tasks that involve Translation
  • SKILL.md covers What you're doing, Classification (vero-discover), Extracting specs from theorems and Naming, plus 4 more sections
  • Instructions only: no scripts, shell commands, URLs or credentials in SKILL.md

What it does

Vero Source Lean is an agent skill from sunblaze-ucb/vero. Load BEFORE curating an existing Lean 4 source repo into the benchmark format. Lean→Lean is mostly extraction, not translation — the source defs become the canonical Impl, the source theorems become the spec obligations. Pair with vero-discover, vero-plan, vero-translate, vero-spec-write, and vero-lean-pitfalls.

Its SKILL.md is about 2k tokens, which your agent loads only when the skill is triggered. It is a single SKILL.md file with no bundled scripts.

It sits in Writing & Content, covering Translation. The licence is Apache-2.0.

When your agent uses it

  • Tasks that involve Translation

Example prompts

  • “/vero-source-lean”

Workflow steps

3 steps, taken from the first numbered list in SKILL.md.

  1. APIs (Impl/.lean): the source's executable defs become def . : := wrapped in !benchmark code def= markers. The body is the real Lean…
  2. Specs (Spec/.lean): the source's theorems / lemmas become def spec_ (impl : RepoImpl) : Prop := definitions. The theorems themselves are…
  3. Bundle / Harness / Test: standard, mirrors BankLedger.

What it can do on your machine

Read from SKILL.md and the folder at commit 0a7325d. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    No scripts in the folder and no shell commands in SKILL.md (its code samples are lean).

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Vero Source Lean loads about 2k tokens when it runs. Until then it costs about 83 tokens; SKILL.md has 929 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~83
When it runs · the whole SKILL.md, loaded when a task matches
~2k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from sunblaze-ucb/vero at commit 0a7325d, republished under its Apache-2.0 licence (© sunblaze-ucb). 929 words, ~2,000 tokens.

Download SKILL.mdSave it as .claude/skills/vero-source-lean/SKILL.md (or your agent's skills folder).
name
vero-source-lean
description
Load BEFORE curating an existing Lean 4 source repo into the benchmark format. Lean→Lean is mostly extraction, not translation — the source defs become the canonical Impl, the source theorems become the spec obligations. Pair with vero-discover, vero-plan, vero-translate, vero-spec-write, and vero-lean-pitfalls.

VCG Source — Lean

Lean source is the special case: the upstream is already in the target language. The curation work is not translation but extraction + reshaping to fit the ratified benchmark paradigm.

Reference the canonical shape at reference/BankLedger/. The end product looks identical to a benchmark curated from Dafny / Verus / Coq — the difference is the path to get there.

What you're doing

Given a Lean 4 source repo (no lakefile.toml at the curation-output path, just .lean files), produce a benchmark Lean project with:

  1. APIs (Impl/<Module>.lean): the source's executable defs become def <Project>.<name> : <Sig> := <body> wrapped in !benchmark code def=<name> markers. The body is the real Lean source body — pre-agent-gen will replace it with sorry before the LLM sees the benchmark.
  2. Specs (Spec/<Module>.lean): the source's theorems / lemmas become def spec_<name> (impl : RepoImpl) : Prop := <statement> definitions. The theorems themselves are dropped — proofs live in Proof/ (materialized later for gen).
  3. Bundle / Harness / Test: standard, mirrors BankLedger.

Because Lean→Lean is extraction, every selected theorem/spec must keep source traceability. Emit or preserve a source map that records the source declaration, target Lean spec/API, and whether the item was extracted, compressed, trusted/opaque, or intentionally skipped. Do not assume one source file maps to exactly one benchmark file.

Classification (vero-discover)

Source constructOutput classification
def f : T := body (top-level executable)API (kind=api) — body becomes code marker interior
def _f / private def f / nested def inside a non-APIAPI helper (kind=api_helper)
theorem t : P x := proofSpec candidate — P x is the spec body
lemma t : P x := proofSpec candidate (same as theorem)
def P : T → Prop := … (predicate)Spec helper (spec_helpers[]) — used in spec bodies
structure Foo where ...Type — preserved verbatim
inductive Foo := ...Type
abbrev FooSig := T → USig abbrev — passed through as-is
instance : Hashable Foo := …Type-class instance — preserved in global_aux if unrelated to spec bodies
axiom A : PTrusted axiom — added to manifest.json::trusted_axioms
opaque f : T → UOpaque — preserved in global_aux
import …Import — preserved at file head
#guard … / example …Test → Test.lean
#eval … (debug)Skip; flag with @review human

Extracting specs from theorems

The single hard part. A typical Lean theorem looks like:

lean
theorem ledger_create_zero (ledger : Ledger) (id : AccountId) :
    ¬ accountExists id ledger →
    getBalance id (createAccount id ledger) = some 0 := by
  intro h
  …

The benchmark spec is the proposition shape, not the proof. Strip the proof (:= by …) and reshape into a RepoImpl parameter:

lean
def spec_ledger_create_zero (impl : RepoImpl) : Prop :=
  ∀ (ledger : Ledger) (id : AccountId),
    ¬ impl.bankLedger.accountExists id ledger →
    impl.bankLedger.getBalance id (impl.bankLedger.createAccount id ledger) = some 0

Rules:

  1. Every API reference goes through impl.<repo_impl_field>.<fn>. The source uses bare names (createAccount); the spec uses bundle-qualified names (impl.bankLedger.createAccount).
  2. Theorem : becomes Prop (it already was).
  3. Bound parameters become ∀-quantified at the spec level (the source theorem's parameters become explicit ∀ in the spec body).
  4. The proof is dropped. It's regenerated by the LLM during gen.
  5. Spec helpers: if the theorem statement uses a curator-given predicate (e.g., def isWellFormed : Ledger → Prop), preserve that predicate as a spec_helper_* in the same Spec/<Module>.lean file. Reference it from the spec body.

Naming

  • Theorems prefixed <module>_<intent> map to specs spec_<module>_<intent>.
  • Theorems with verb-style names (e.g., createAccountZeroBalance) map to snake_case (spec_create_account_zero_balance).
  • When the source uses Mathlib-style naming (createAccount_zero_balance), preserve and just prefix spec_.

Extraction workflow (translate stage)

For each module:

  1. Read <source>/<Module>.lean.
  2. Identify executable defs (potential APIs) and theorems (potential spec sources).
  3. Group APIs into <Project>/Impl/<Module>.lean:
    • copy imports + structures + helpers verbatim into global_aux and outside-marker context
    • emit abbrev <Name>Sig := <type> for each API
    • emit def <Project>.<name> : <Project>.<Name>Sig := <original-body> inside code marker pair
    • emit code_aux marker pair (empty, room for the LLM to add internal helpers)
  4. For each chosen theorem (per select.json):
    • shape into def spec_<name> (impl : RepoImpl) : Prop := <body> in <Project>/Spec/<Module>.lean
    • swap bare API references → impl.<repo_impl_field>.<fn>
  5. Update manifest.json::packages[].modules[]:
    • apis[] lists the executables; each entry { name, sig, type, kind: "api" }
    • specs[] lists the spec names (bare strings)
    • spec_helpers[] lists predicate helpers if any
  6. Bundle.lean + Harness.lean + Test.lean standard (one bundle field per API; canonical wires through to <Project>.<api>; #guards from source #guard/example).
Show full SKILL.md (306 more words)Show less

No spec synthesis for Lean source

Lean→Lean curation is honest extraction of what the upstream source already proves — it must NOT synthesize new specs. The lean_spec workflow therefore deliberately omits the spec_write stage; the only specs in the curated benchmark are the ones the translate stage extracts from existing source theorem / lemma / example declarations.

If a curator wants additional specs, they should extend the upstream source repo with new theorems first (so the new specs are anchored in real, type-checked claims about the API), then re-run curation. Do not edit the curated benchmark to add specs that don't exist upstream.

Do not turn a source theorem into a benchmark obligation if the extracted spec_* ignores RepoImpl, concludes True, or only proves a library helper fact unrelated to the implementation. Classify those as spec_helper / source context unless there is an explicit human decision to keep a theorem-only obligation.

Verify before declaring done

  1. lake build succeeds.
  2. Every apis[] entry has its abbrev <Sig> + def <Project>.<name> in the right Impl file.
  3. Every specs[] entry exists as def spec_<name> (impl : RepoImpl) : Prop := … in the right Spec file.
  4. Spec/<Module>.lean files have NO theorem / lemma / example declarations (the validator's spec_shape check enforces this).
  5. The source repo's axiom declarations are listed in manifest.json::trusted_axioms.
  6. Translated-source provenance is present in .vero/source_map.json, .vero/discover.json, or another validator-readable artifact.

What NOT to do

  • Don't preserve theorems in Spec/. They're not benchmark obligations; they're proofs that the source was correct.
  • Don't translate Mathlib references unless the curator explicitly authorizes Mathlib in the target benchmark.
  • Don't introduce new instance blocks for DecidableEq unless they're already in the source — the validator's anti-cheat instance check (gen/eval side) rejects suspicious additions.
  • Don't conflate def spec_helper_x (executable predicate, used in specs) with def spec_x (proof obligation). They live in the same Spec file but their manifest classification differs.

© sunblaze-ucb, Apache-2.0. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

Just SKILL.md in .claude/skills/vero-source-lean of sunblaze-ucb/vero.

Open the folder on GitHubat commit 0a7325d

Compare with similar skills

Vero Source Lean next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Vero Source Lean compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Vero Source Lean this skillsunblaze-ucb/vero107—~2kAutomated safety check: PassApache-2.0
Translation Diff ExportDevolutions/UniGetUI26k—~1.1kAutomated safety check: PassMIT
Sync Translationssymfony/symfony31k—~1.9kAutomated safety check: PassMIT
Translation Diff ImportDevolutions/UniGetUI26k—~750Automated safety check: PassMIT
Translation Diff TranslateDevolutions/UniGetUI26k—~934Automated safety check: PassMIT
Generate Translationspayloadcms/payload45k—~1.1kAutomated safety check: PassMIT

Similar skills

  • Translation Diff Export

    Devolutions/UniGetUI

    Compares UniGetUI JSON locale files against English, identifies untranslated or source-changed keys, and generates patch, reference, and handoff files for a target language.

    26k GitHub stars~1.1k tokensUpdated yesterday
    Writing & ContentAuto-check passed
  • Sync Translations

    symfony/symfony

    Synchronize translation catalogs across maintained Symfony branches: find messages that newer branches added to the English catalogs but that are still missing from the oldest maintained branch…

    31k GitHub stars~1.9k tokensUpdated yesterday
    Writing & ContentAuto-check passed
  • Translation Diff Import

    Devolutions/UniGetUI

    Merges translated key-value pairs from a UniGetUI JSON localization patch back into the full language file and validates the merged result.

    26k GitHub stars~750 tokensUpdated yesterday
    Writing & ContentAuto-check passed
  • Translation Diff Translate

    Devolutions/UniGetUI

    Translates a sparse UniGetUI JSON language patch, writes completed entries into the working copy, preserves placeholders and terminology, and prepares the patch for merge-back.

    26k GitHub stars~934 tokensUpdated yesterday
    Writing & ContentAuto-check passed
  • Generate Translations

    payloadcms/payload

    A skill your agent uses when new translation keys are added to packages to generate new translations strings

    45k GitHub stars~1.1k tokensUpdated today
    Writing & ContentAuto-check passed
  • Drives long-form fiction, scripts, storyboards, interactive films and long-document translation through InkOS, with every change made by a typed action.

    10k GitHub starsUsed in 1 repo~1.1k tokens
    Writing & ContentAuto-check passed

More from sunblaze-ucb/vero

All 16 skills in this repo
  • Vero Discover

    sunblaze-ucb/vero

    A skill your agent uses when scanning a verified source repo (Dafny, Verus, or Coq) to classify every item and produce per-file discovery markdown for human curation.

    107 GitHub stars~3.9k tokensUpdated 1 mo ago
    Auto-check: notes
  • Vero Translate

    sunblaze-ucb/vero

    A skill your agent uses when translating selected verified items from Dafny/Verus/Coq into a compilable Lean 4 benchmark.

    107 GitHub stars~4.9k tokensUpdated 1 mo ago
    Auto-check: notes
  • Vero Coq Pitfalls

    sunblaze-ucb/vero

    Load BEFORE translating any Coq item to Lean 4 to avoid known Coq→Lean pitfalls.

    107 GitHub stars~1.5k tokensUpdated 1 mo ago
    Auto-check: notes
  • Vero Dafny Pitfalls

    sunblaze-ucb/vero

    Load BEFORE translating any Dafny item to Lean 4 to avoid known Dafny→Lean pitfalls.

    107 GitHub stars~1.2k tokensUpdated 1 mo ago
    Auto-check: notes
  • Vero Lean Pitfalls

    sunblaze-ucb/vero

    Load BEFORE writing any Lean 4 translation to avoid common Lean pitfalls (universes, coercions, type-class resolution, notation).

    107 GitHub stars~1.4k tokensUpdated 1 mo ago
    Auto-check: notes
  • Vero Plan

    sunblaze-ucb/vero

    Use after vero-select to write a detailed translation plan as .vero/plan.json — the authoritative contract the TRANSLATE stage executes.

    107 GitHub stars~3.6k tokensUpdated 1 mo ago
    Auto-check: notes

Questions about Vero Source Lean

What does Vero Source Lean do?

Load BEFORE curating an existing Lean 4 source repo into the benchmark format. Vero Source Lean is an agent skill from sunblaze-ucb/vero. Load BEFORE curating an existing Lean 4 source repo into the benchmark format.

When should I use Vero Source Lean?

Vero Source Lean fits situations like: tasks that involve Translation.

How do I install Vero Source Lean in Claude Code?

Run `npx skills add sunblaze-ucb/vero --skill vero-source-lean -a claude-code`. Or copy the skill folder (.claude/skills/vero-source-lean in sunblaze-ucb/vero) into .claude/skills/vero-source-lean in your project. Claude Code loads it when a task matches its description.

How do I install Vero Source Lean in Codex?

Run `npx skills add sunblaze-ucb/vero --skill vero-source-lean -a codex`. Or copy the skill folder (.claude/skills/vero-source-lean in sunblaze-ucb/vero) into .agents/skills/vero-source-lean in your project. Codex loads it when a task matches its description.

Can I use Vero Source Lean in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add sunblaze-ucb/vero --skill vero-source-lean -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/vero-source-lean, .gemini/skills/vero-source-lean, .github/skills/vero-source-lean and .opencode/skills/vero-source-lean in your project.

What does Vero Source Lean need to run?

SKILL.md names no scripts, command-line tools or credentials: Vero Source Lean is instructions for the agent only.

Does Vero Source Lean access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Vero Source Lean safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Vero Source Lean use?

Vero Source Lean is published under the Apache-2.0 licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Vero Source Lean use?

About 2k tokens (SKILL.md is roughly 8k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Vero Source Lean?

Skills that share tags, products or a category with Vero Source Lean: Translation Diff Export (Devolutions/UniGetUI, 26k stars), Sync Translations (symfony/symfony, 31k stars), Translation Diff Import (Devolutions/UniGetUI, 26k stars) and Translation Diff Translate (Devolutions/UniGetUI, 26k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Vero Source Lean?

sunblaze-ucb (a GitHub organization) maintains it in sunblaze-ucb/vero, which has 107 GitHub stars. The repository holds 16 skills in this directory. The repository was last updated on August 17, 2026.

Source: sunblaze-ucb/vero on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.