setup
Skill136 starsChanged 19 days agoMITView on GitHub
Set up, inspect, or repair repository infrastructure for an Autoform Lean project, including the Lean/Mathlib shell, an in-repository Obsidian-compatible blueprint vault, ignore rules, MkDocs, GitHub Pages, and verification CI, with optional Zulip community synchronization. Use for new repositories, environment repair, publication setup, infrastructure checks, or an explicitly requested Zulip project sync; do not choose mathematical scope or build the roadmap and theorem DAG.
191 other projects use a version of this.
Adopted by
Most recently taken up by
- hydra-db/hydradb-claude-codetheir own version, first seen Nov 8, 2026
- WALKERKILLER/Look-Tongji-Notestheir own version, first seen Nov 8, 2026
- steerlabs/opensteertheir own version, first seen Nov 8, 2026
- ansvisor/ansvisortheir own version, first seen Nov 8, 2026
- dsifry/metaswarmtheir own version, first seen Nov 8, 2026
- anombyte93/prd-taskmastertheir own version, first seen Nov 8, 2026
Used with
Projects with it also use add-voice-transcription, add-discord and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
juancruzmunozalbeloSkilladd-voice-transcription
Add voice message transcription to NanoClaw using OpenAI's Whisper API. Automatically transcribes WhatsApp voice notes so the agent can read and respond to them.
Also in 5 projects that use this
MicrosoftSkilladd-discord
Add Discord bot channel integration via Chat SDK.
Also in 11 projects that use this
MicrosoftSkilladd-slack
Add Slack channel integration via Chat SDK.
Also in 11 projects that use this
juancruzmunozalbeloSkilladd-telegram-swarm
Add Agent Swarm (Teams) support to Telegram. Each subagent gets its own bot identity in the group. Requires Telegram channel to be set up first (use /add-telegram). Triggers on "agent swarm", "agent teams telegram", "telegram swarm", "bot pool".
Also in 7 projects that use this
- +8 more from microsoft
Other versions191 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 191
- anthropics/knowledge-work-plugins★ 35ksame name
- garrytan/gbrain★ 32ksame name
- gavrielc/nanoclaw★ 31ksame name
- nanocoai/nanoclaw★ 31ksame name
- qwibitai/nanoclaw★ 31ksame name
- Q00/ouroboros★ 6.3ksame name
- pytorch/executorch★ 5.2ksame name
- agenticnotetaking/arscontexta★ 3.5ksame name
- aws/agent-toolkit-for-aws★ 3.4ksame name
- spotify/portal-ai-plugins★ 2.8ksame name
181 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository7 other files it gives its agents.
- AGENTS.mdAGENTS.md
- agent-reviewSkill
- debrief-orchestratorSkill
- develop-pluginSkill
- formalizeSkill
- human-reviewSkill
- roadmapSkill
The file
We split this skill into the rules it gives an agent
4 rules. 0 are common, 0 are distinctive, and 4 are very specific to this project.
0
Common
Most of the 170 edited versions we read keep these.
0
Distinctive
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 48
If the user names a donor (never search for one), <PKGS> is its .lake/packages, or the donor if it is a packages folder.
- Very specific: line 95
If neither tag exists in git ls-remote --tags https://github.com/leanprover-community/repl.git, ask the user instead of guessing.
- Very specific: line 117
After it runs, fill in what only a human or a source can supply: the project description in blueprint/README.md, the coverage contract, and …
- Very specific: line 156
Before reporting, follow Debrief Orchestrator with --skill setup.
“Common” is measured: kept by at least 6 in 10 of the 170 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Set up an Autoform repository
Inspect before writing and preserve existing Lean, Markdown, workflow, and
ignore files. Resolve AUTOFORM_PLUGIN_ROOT from the loaded plugin and inspect
an existing project read-only:
AUTOFORM_PLUGIN_ROOT="<loaded-plugin-root>"
PROJECT="<existing-project>"
uv run --project "$AUTOFORM_PLUGIN_ROOT" autoform project inspect "$PROJECT" --jsonInfer safe local defaults from the request and repository. If a material choice is missing, ask once for the run type (new, repair, or inspect), UpperCamelCase package name, target directory, and whether publication is wanted. Without explicit publication approval, make no remote changes.
The Cabannes thesis project shows a set-up project with a deliberately partial DAG. Read it, but do not copy it.
For a new repository, create the project in an absent target directory:
AUTOFORM_PLUGIN_ROOT="<loaded-plugin-root>"
TARGET="<absent-target>"
PACKAGE="<UpperCamelCaseName>"
uv run --project "$AUTOFORM_PLUGIN_ROOT" autoform project new "$TARGET" --package "$PACKAGE"Project setup gives
project new's version flags, what it writes, and its error codes. Keep the
recommended release from autoform project versions unless the user needs
another Lean version (at least v4.27.0). On project-parent-unsafe, choose
another parent or, with the user's agreement, run chmod g-w,o-w on it.
If the user names a donor (never search for one), <PKGS> is its
.lake/packages, or the donor if it is a packages folder. Before any Lake
command runs in the new project, lake update repl included, copy each git
package the new lake-manifest.json locks that .lake/packages lacks and whose
<PKGS>/<NAME> is no symbolic link, prints that rev for git rev-parse HEAD,
and prints nothing for git --no-optional-locks status --porcelain: mkdir -p
.lake/packages && cp -Rc <PKGS>/<NAME> .lake/packages/ (cp -R --reflink=auto
on Linux). If pgrep -x lake finds a Lake process, copy nothing and say why.
Autoform's library catalog,
$AUTOFORM_PLUGIN_ROOT/autoform_cli/project/libraries.json, lists shared Lean
libraries that project new --library <NAME> adds, required and locked. Offer
only entries with offered: true, and only for a new project (never inspect,
repair, or init). Ask in the single round of questions and state
the cost: the project moves to the library's release, its first build
downloads the library and its dependencies (about 270 MB for ATLAS), and the
modules it imports build locally. A yes may come in the request itself; a
request naming a catalog library uses it even when it is not offered, unless it
names another repository or revision, which makes it a library outside the
catalog. Without a yes, add no library, including in a headless run.
On a yes, pass --library <NAME> with no version flags, validate as below,
and run autoform library list on the project, as the
CLI reference
shows: report the entry's git and pinned commit if it lists the library as
usable, otherwise its reason, without claiming the library works. If a fetch
fails, report the error and that the project builds once a retry from the failed
validation step succeeds. Never run a bare lake update in such a project.
For a library outside the catalog, pass --lean-toolchain and --mathlib-rev
set to what its lean-toolchain and lake-manifest.json lock at the revision
the user names, writing 4.34.1 as v4.34.1; a read-only fetch of both files
is fine. Ask when the revision or either value is unknown, and report the pair
and why. Do not add the require or fetch the library into the project.
project new writes the complete Autoform vault, site, ignore rules, and CI.
For an incomplete existing repository, run autoform init through the same
uv run --project "$AUTOFORM_PLUGIN_ROOT" prefix: it writes the whole vault and
keeps existing files. Do not hand-build any of the vault. When either omits CI,
run the init it names on the same target with the commit the plugin was
installed from, or report that CI was not configured. Never invent a ref or
workflow source.
Neither project new nor init adds the Lean REPL that the REPL tool runs.
Require leanprover-community/repl under the name repl at the commit of the
upstream tag naming the project's Lean version or, failing that, its vX.Y.0
release, as the bundled example pins the commit of tag v4.32.0 for v4.32.2.
If neither tag exists in
git ls-remote --tags https://github.com/leanprover-community/repl.git, ask
the user instead of guessing. Then run lake update repl so an existing
manifest locks it without moving Mathlib, and commit the manifest.
For an existing project whose lakefile lacks either of
enableArtifactCache = true and restoreAllArtifacts = true, offer to add the
missing one; a lakefile.lean takes them in its package declaration. Never
enable the cache without restoreAllArtifacts.
The CLI reference
says when Pages deploys, and open
statements
what the generated verify workflow audits. Where several Lake projects back the
roadmap, offer init --no-verify-workflow; Pages then publishes unverified
commits. init also writes .github/CODEOWNERS.autoform.example, which GitHub
does not read; the
template says
how to activate it. Do not guess owners; activate it only with names the user
gives, never silently. Report that new rules cannot protect their own activation
unless equivalent coverage already exists on the pull request's base branch.
After it runs, fill in what only a human or a source can supply: the project
description in blueprint/README.md, the coverage contract, and a verified
repo_url, the formalization project's own repository, never AutoformBot's.
Pass --repository-url to autoform init only once the remote exists. When a
deployed site exists, feature its verified canonical URL in the root
README.md, never an inferred or pending one.
Adding workflow files is a local repository edit. Creating a remote, pushing, or enabling Pages are separate outward-facing actions; perform them only when the user requests them. Pin third-party Actions and the Autoform CLI source to immutable commits.
Validate the prepared repository before reporting it ready. Build Lean first:
lake update # only when the project has no lake-manifest.json
lake update repl # after adding the REPL require when a manifest exists
lake exe cache get # skip only when the project has no Mathlib dependency
lake build
lake build @repl/replThen run the publication sequence in the
CLI reference,
keeping --require-declarations.
Publication is opt-in: files under blueprint/ become public site content, with
derived progress, graph pages, and a publication manifest that names blueprint
files, hidden ones included. Show that boundary, confirm the exact repository
and visibility, default to private, and warn that private Pages may require a
paid GitHub plan. When approved, prepare the commit, remote, Pages source, and
push; otherwise leave the workflow inert. If credentials, hosting, or repository
settings block publication, report the minimal owner action required.
Zulip synchronization is a separate opt-in outward-facing action. When the user asks to discover community context or announce and coordinate the project, follow the shared Zulip workflow. Do not infer consent to post from repository setup, roadmap work, or permission to search.
Before reporting, follow Debrief Orchestrator
with --skill setup. Report the Lean toolchain, the packages copied from a
donor and those fetched instead, with why, vault path, CI and Pages files,
validation results, the publication decision, and any one-time GitHub setting
the user must still apply. State that no sources were scoped, roadmap nodes
created, or proofs started, then hand the repository to Roadmap.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

