Lean Public Release logo

Lean Public Release

Community
scottnarmstrong
lean-public-release

Prepare and verify a private Lean repository for public release. Use for source scrubbing, selective syncs, clean history, proof-status and license review, fresh-checkout builds, optional registry or comparator intake, independent review, and an explicit publication gate.

Overview

Publisherscottnarmstrong
RepositoryLeanAutoformalizationSkills
Skill namelean-public-release
Stars
66
Forks
7
Bundled files
1
LicenseCC-BY-4.0
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 1 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by scottnarmstrong on GitHub. Read the source before you install it.

Installation

Install the Lean Public Release AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/scottnarmstrong/LeanAutoformalizationSkills/tree/main/skills/lean-public-release
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/scottnarmstrong/LeanAutoformalizationSkills.git /tmp/LeanAutoformalizationSkills
mkdir -p .claude/skills
cp -r /tmp/LeanAutoformalizationSkills/skills/lean-public-release .claude/skills/lean-public-release
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Lean Public Release in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Lean Public Release on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Lean Public Release is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Lean public release

Prepare a reader-facing public repository without exposing private paths, internal history, credentials, campaign records, or unpublished source material. Read the repository's instructions and release destination rules before choosing a workflow. Preparing a release does not itself authorize a commit, push, visibility change, registry submission, or publication.

Freeze the release contract

Record before editing:

  • the exact source revision and release target;
  • which library modules, verification tools, documentation, metadata, and license files will ship;
  • the intended proof-status policy, including any authorized drafts;
  • third-party code, data, text, and licenses that require attribution;
  • whether the public repository is a one-time export or a maintained mirror;
  • optional destination requirements such as a registry, archive, or challenge format.

Preserve source-to-statement fidelity throughout. Scrubbing must not change binders, carriers, quantifiers, hypotheses, conclusions, constants, endpoint conditions, or unique-characterization semantics. Never convert a proof step into an assumption or replace a definition with hypotheses about an abstract object. Keep draft theorems quarantined and honestly labeled.

Selective sync

  • Use an allowlist of files and directories that may ship. Treat everything else as excluded until reviewed.
  • Run a forbidden-reference lint on the trial tree. Check private repository names, usernames, absolute paths, scratch locations, internal issue or packet labels, credentials, unpublished-source paths, and excluded tooling.
  • Keep a no-commit preview mode. Sync first to a disposable checkout and inspect the complete resulting tree and diff.
  • If a fresh public history is desired, create it only after approval. A one-commit history is an option, not a universal requirement.
  • Define how later public changes are synchronized. Do not assume rewriting published history is acceptable.

Source and documentation scrub

Replace internal generation suffixes, work-item names, unstable line-number citations, and private file references with stable public declaration names, paper theorem numbers, labels, or public URLs. Verify every renamed declaration and cross-reference mechanically where possible.

Do not erase provenance or required attribution. Preserve copyright notices, third-party licenses, paper citations, and authorship information. Separate release-facing explanations from private process history, but retain audit artifacts that the release contract says are part of the public verification story.

Lean integrity gates

Run checks against the exact proposed public tree:

  1. Search for unauthorized sorry, admit, custom axioms, placeholder definitions, theorem-shaped proposition packages, and disabled checks.
  2. Run the repository's axiom audit from probes or a dedicated audit target. Do not leave expensive #print axioms commands in ordinary source files.
  3. Confirm imports resolve only through declared dependencies at pinned revisions. Remove private path dependencies, source symlinks, copied build trees, and hidden search paths.
  4. Build in a fresh checkout with empty project artifacts and the documented dependency setup. This is the isolation test; a build in the development checkout is insufficient.
  5. Run every shipped checker exactly as documented for readers.
  6. Compare public theorem statements with their authoritative sources after every rename or module move.

Respect local cache rules. A first-time fresh-checkout build may need to create dependency artifacts; a guard designed for an already-populated development checkout must account for that setup explicitly.

Optional registry and challenge requirements

Apply this section only when the chosen destination requires it. Read the current registry policy at release time; size limits, module rules, metadata, and schemas can change.

If every Lean file must be a module, port project modules and shipped checkers before release. Publish and pin project-owned dependencies in dependency order. Validate module headers, imports, visibility modifiers, root names, and aggregator files in the synced tree.

If comparator-style challenges are required:

  • give each main result a challenge with the destination's required deliberate proof hole and a solution proving the byte-identical statement;
  • match the allowed dependency closure and avoid private declarations whose generated names differ by module;
  • define every source-defined notion needed by the statement inside the permitted challenge surface;
  • for an object defined by unique characterization, include its actual construction or audited defining interface and verify existence and uniqueness without extra assumptions;
  • use concise definitions to meet size limits; do not hide definitions in hypotheses or move them outside the reviewed closure;
  • invalidate only the exact challenge artifacts before verification.

Registry-specific helper scripts from companion skills may be used only when they exist in the installed skill set and match the destination's current rules.

Hosted CI

If the release ships continuous integration:

  • Use two workflows, each with a README badge under the title: one builds the project and runs the axiom audit for every advertised theorem plus the module and metadata checks; the other runs the comparator checks with the independent kernel replay. Keep them manual-dispatch drafts until release, then trigger on pushes and pull requests to the default branch. Give each workflow its own build-cache key; restores may fall back to the other's prefix.

  • Free disk space first in every workflow that builds the project or runs comparators. A hosted runner starts with roughly 14 GB free, and replaying an exported solution over the full Mathlib closure in an independent kernel exhausts it. The typical signature is that both builds and both exports succeed, then the kernel replay dies with no space left on device. That is an environment failure, not a verification failure, but it still fails the badge. Put a step like this right after checkout:

    yaml
        - name: Free runner disk space
          run: |
            set -Eeuo pipefail
            # Only remove preinstalled, unused SDKs on disposable hosted runners.
            if [[ "${RUNNER_ENVIRONMENT:-}" == github-hosted ]]; then
              sudo rm -rf /usr/share/dotnet /usr/local/lib/android /opt/ghc /opt/hostedtoolcache/CodeQL
              sudo docker image prune --all --force > /dev/null || true
            fi
            df -h .
            available_kib="$(df -Pk . | awk 'NR==2 {print $4}')"
            test "$available_kib" -ge 20971520
  • Read the failing step's log before diagnosing. A failing comparator job after green local runs is usually a runner resource limit, not a statement mismatch.

  • Let every workflow finish green on the exact pushed commit before the repository becomes public; badges show no status until a run completes.

Reader-facing files

The README should tell a new reader:

  • what is formalized and which source results correspond to it;
  • the exact Lean toolchain and pinned dependencies;
  • how to obtain caches or build dependencies, build the project, run audits, and reproduce optional challenges;
  • which results belong to upstream dependencies;
  • the honest status of drafts, gaps, axioms, and excluded material;
  • the license and citation requirements.

Add docs/ERRATA-AND-DEVIATIONS.md and a README "Errata and deviations" section with counts per source, verified independently against the pinned sources before publication. Follow the lean-orchestrator skill's references/errata-and-deviations.md. Erratum vocabulary is scrubbed from code comments, not from this record.

GitHub renders README math after HTML sanitizing. Inside math, write \lt and \gt instead of < or > directly before a letter or backslash (\sum_{i<j} breaks the formula). Escape literal $ outside code in generated Markdown such as logs or transcripts. GitHub's MathJax also rejects macros outside its allowlist ("The following macros are not allowed: operatorname"). Write \mathrm{Im} or \mathop{\mathrm{Res}}, never \operatorname, \DeclareMathOperator, \newcommand or \def. Grep every shipped .md file for these before each sync. Check the rendered page on GitHub; local renderers do not reproduce these failures.

A comparator CI workflow on GitHub-hosted ubuntu-24.04 runners must install the sandbox before lake comparator runs; otherwise it fails with "bwrap is required to run lake comparator". Ubuntu 24.04 also restricts unprivileged user namespaces through AppArmor. Add this step before the comparator step:

yaml
- name: Install bubblewrap for lake comparator's sandbox
  run: |
    sudo apt-get update -qq >/dev/null
    sudo apt-get install -y -qq bubblewrap >/dev/null
    sudo sysctl -w kernel.apparmor_restrict_unprivileged_userns=0

README lean blocks that quote a theorem statement must be complete and verbatim. Never cut the source at its first :=: statements can contain (by have := …) or let x := …. Cut at the proof-region marker or at the := that ends the signature line. Then check each block is a substring of its source file, followed only by := and the proof.

Check LICENSE, citation metadata, repository metadata, and dependency licenses together. A code license does not automatically cover manuscript text, figures, datasets, or vendored third-party material.

Independent review and release gate

Have a reviewer who did not perform the scrub inspect the exact proposed tree:

  • statement and definition fidelity against pinned sources;
  • proof status, axiom output, draft quarantine, and absence of proof smuggling;
  • dependency isolation and fresh-checkout reproducibility;
  • forbidden references, secrets, and private artifacts;
  • README accuracy, licenses, attribution, and optional registry compliance.

Resolve each finding or record a justified disposition. Present the complete release tree, README, verification results, unresolved limitations, and planned external actions for approval. Only then perform the separately authorized commit, push, visibility change, archive upload, or registry submission.

Suggested sequence

  1. Freeze scope, source revision, proof-status policy, licenses, and destination rules.
  2. Scrub names and references in the private source while preserving statement meaning.
  3. Prepare public documentation, metadata, licenses, and CI in draft form.
  4. Build an allowlisted no-commit sync and burn forbidden-reference findings down to zero.
  5. Verify the trial tree in a fresh checkout and run all shipped checks.
  6. Obtain independent review and resolve findings.
  7. Obtain explicit approval for the README and proposed external actions.
  8. Commit or publish only within that approval.

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Lean Public Release AI skill do?

Prepare and verify a private Lean repository for public release. Use for source scrubbing, selective syncs, clean history, proof-status and license review, fresh-checkout builds, optional registry or comparator intake, independent review, and an explicit publication gate.

Why use Lean Public Release on TypingMind?

Because you install it once and use it with any model. Lean Public Release is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Lean Public Release in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/scottnarmstrong/LeanAutoformalizationSkills/tree/main/skills/lean-public-release. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Lean Public Release?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Lean Public Release?

As many as you like. As long as a model supports skills, you can use Lean Public Release with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Lean Public Release AI skill free?

Yes. It is published on GitHub by scottnarmstrong under the CC-BY-4.0 license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇