gardener
Skill2 starsChanged 36 days agoMITView on GitHub
Keep rat-stack current and clean. Update exact dependency versions, learn from repos on the same versions, and add lint rules before cleanup.
- Installs packages
2 other projects use a version of this.
Adopted by
Most recently taken up by
- jdiazromeral/tarstheir own version, first seen Oct 3, 2026
- khou/gardenkittheir own version, first seen May 14, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- khou/gardenkit★ 4same name
- jdiazromeral/tars★ 1same name
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 repository15 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- add-a-capabilitySkill
- add-a-lifecycle-machineSkill
- add-a-storeSkill
- investigate-a-failureSkill
- keep-or-cutSkill
- learn-alchemySkill
- profile-a-workerSkill
- query-designSkill
- rat-stack-modeSkill
- shipSkill
- test-behaviourSkill
- uncomplectSkill
- write-a-wiki-pageSkill
The file
We split this skill into the rules it gives an agent
12 rules. About 0 are common, 5 are distinctive, and 7 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
7
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 24
Read AGENTS.md first. Work in a clean clone off origin/main, never in a checkout with other people's uncommitted files.
- Very specific: line 45
Check peer ranges before installing: pnpm view alchemy@<pin> peerDependencies.
- Very specific: line 48
Read what changed upstream: git -C .agent_sources/github.com/<owner>/<repo> log --oneline <old-tag>..<new-tag> when the old tag is …
- Very specific: line 49
Update the version text in AGENTS.md and README.md.
- Very specific: line 50
Run pnpm turbo run check test build --force and check its exit code.
- Distinctive: line 52
Commit each dependency family separately, so a regression can be traced to one library.
- Very specific: line 60
Run pnpm --filter @rat-stack/site generate, read .brain/data/unlinked-mentions.generated.json, and turn useful unlinked mentions into …
- Distinctive: line 62
Look for technical debt and patterns that should not spread:
- Very specific: line 72
Write the lint rule or test that fails on it first.
- Distinctive: line 74
If some findings have to stay for now, record them as a baseline that can only shrink.
- Distinctive: line 75
Delete code that nothing uses.
- Distinctive: line 77
When a pattern cannot be caught by lint, fix it in the codebase so the wrong version is hard to write.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Maintain rat-stack
Rat-stack runs on Effect 4 stable, Alchemy 2 betas, and XState 6 alphas. Effect describes work with typed failures and required dependencies. Alchemy declares and deploys cloud resources through Effect programs. Keep copied code compatible with the current dependency versions. Simplify it as these dependencies change.
Follow Lauren Tan's gardener loop from her Dune talk: remove technical debt, keep one supported implementation pattern, and reject prohibited patterns with lint. In her words: "whenever you see tech debt or bad patterns, your instinct should be, I need to write a lint rule against it." She ranks where to put corrections by enforcement strength: codebase, static analysis, rules, skills, style guide. Use the strongest applicable option for each finding.
Read AGENTS.md first. Work in a clean clone off origin/main, never in a checkout with other people's uncommitted files.
1. Check the pins
The core pins live in the package.json files and in minimumReleaseAgeExclude in pnpm-workspace.yaml. Compare each with the currently published versions on npm. Run registry queries outside the repo so its development-engine check does not block them:
cd /tmp
for p in effect @effect/platform-node @effect/vitest @effect/tsgo alchemy xstate @xstate/effect; do
printf '%-24s ' "$p"; pnpm view "$p" dist-tags --json | tr -d '\n '; echo
doneRead the returned versions, not only the tag names. Effect 4 and its adapters now use latest; rc points to the older release candidate. Alchemy 2 and @effect/tsgo use latest. XState 6 uses alpha; its latest still points to XState 5. For @xstate/effect, compare both alpha and latest across alpha versions: latest can be newer. Check the bridge's peer ranges before choosing the matching XState pin.
2. Update one dependency family at a time
For each dependency family that is behind:
- Replace the pin everywhere it appears. Effect runtime packages and adapters move together;
rg -l -F '"<old-pin>"' --glob '**/package.json'finds both stable and prerelease pins. Keep each Effect-family update in its own commit, separate from other dependency updates and cleanup. XState and its Effect bridge move together. - Respect the release-age policy. Replace an existing
minimumReleaseAgeExcludeentry with the new exact package version; never widen it to a package name or range. Before adding a necessary exact-version exception, inspect the registry integrity, provenance, tarball file list, package scripts, and dependency changes. Record what you checked. - Check peer ranges before installing:
pnpm view alchemy@<pin> peerDependencies. Alchemy pins exact Drizzle versions and a minimum Effect; follow them. - Let
pnpm installgenerate the lockfile; never edit it by hand. Before gating, prove freshness withpnpm --config.optimistic-repeat-install=false install --frozen-lockfile --force, thenpnpm peers check; compare warnings withorigin/main. After any dependency change, runpnpm install --frozen-lockfilebefore each gate. KeepverifyDepsBeforeRun: error: pnpm otherwise auto-installs during scripts. Even--forcecan skip broken YAML through the optimistic cache;lockfile:checkdisables that cache and never repairs the lock. - Refresh the source mirrors with
./scripts/vendor-agent-sources.sh --refreshand update the Ref column in.agent_sources/README.md. Agents read the mirrors as truth, so a stale mirror teaches old APIs. - Read what changed upstream:
git -C .agent_sources/github.com/<owner>/<repo> log --oneline <old-tag>..<new-tag>when the old tag is fetchable, or the package's CHANGELOG. Grep our code for every API those commits touch. - Update the version text in
AGENTS.mdandREADME.md.packages/core/test/docs-pins.test.tsfails until they match the manifests. - Run
pnpm turbo run check test build --forceand check its exit code.
Commit each dependency family separately, so a regression can be traced to one library.
3. Peers refresh
Check the dataset's oldest peer checked date on every gardener pass. If it is fourteen days old, refresh it before continuing; a shared dependency update always triggers a refresh. This repo-native pass owns the cadence. Put the rubric, full ranking, tier changes, and unresolved versions at the top of the report for feedback after shipping; no separate approval round is required.
4. Remove prohibited patterns
Run pnpm --filter @rat-stack/site generate, read .brain/data/unlinked-mentions.generated.json, and turn useful unlinked mentions into explicit source links after checking their context; leave incidental matches alone.
Look for technical debt and patterns that should not spread:
- Duplicated logic.
- A second implementation where a supported pattern already exists.
- Unchecked
unknownvalues passed to application code. - Unused exports.
- Tests that assert nothing.
For each one:
- Write the lint rule or test that fails on it first. Put a real-
oxlintfixture test beside the other rule tests inpackages/core/test/. Break the rule once to watch the test fail. - Then fix every existing instance, so the rule starts at zero findings.
- If some findings have to stay for now, record them as a baseline that can only shrink. Never raise a baseline.
- Delete code that nothing uses. Git keeps it.
When a pattern cannot be caught by lint, fix it in the codebase so the wrong version is hard to write. Only fall back to prose in AGENTS.md or a skill when neither works.
Report
Report:
- each pin moved, from and to;
- upstream changes that affected us;
- peers studied, and what was adopted or rejected;
- lint rules added, and the findings each one removed;
- anything left undone.
Name the changes required in apps that copy rat-stack.
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.

