upgrade-plan
Skill5 starsChanged 11 days agoMITView on GitHub
Plan a library or framework upgrade from the installed version to the latest. Finds what breaks, which local patches and workarounds can go, and what to adopt, sorted into must do, should do, and do later.
In this repository57 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- architecture-satisfactionSkill
- autopilotSkill
- browser-testSkill
- chatter-scoutSkill
- closeoutSkill
- codebase-designSkill
- code-reviewSkill
- code-smells-auditSkill
- commitSkill
- dependency-auditSkill
- diagnosing-bugsSkill
- domain-modelingSkill
- fix-ci-failuresSkill
- fix-github-issuesSkill
- fix-nightwatch-issuesSkill
- fix-posthog-issuesSkill
- grillingSkill
- grill-meSkill
- grill-with-docsSkill
- handoffSkill
- harness-toolsSkill
- housekeeperSkill
- implementSkill
- implement-specSkill
- improve-codebase-architectureSkill
- laravel-herd-worktreesSkill
- panelSkill
- prototypeSkill
- prSkill
- prune-testsSkill
- releaseSkill
- repository-cleanupSkill
- researchSkill
- resolve-review-commentsSkill
- resolving-merge-conflictsSkill
- retroSkill
- run-afterSkill
- scratch-cleanupSkill
- scratch-statusSkill
- setupSkill
- sweepSkill
- tddSkill
- teachSkill
- to-questionnaireSkill
- to-specSkill
- to-ticketsSkill
- triage-github-prSkill
- triageSkill
- ui-designSkill
- verifySkill
- wait-whatSkill
- wayfinderSkill
- which-skillSkill
- wizardSkill
- writing-for-agentsSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 32
Cascade: Use the script output to find what else has to move:
- Distinctive: line 43
Add each package that must move to the run, with its own target, and run step 1 on it.
- Very specific: line 67
Triage: Put every item with a verdict other than none in exactly one bucket.
- Distinctive: line 68
Use the first rule that fits:
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
Upgrade Plan
Plan the upgrade of one or more packages from the installed version to a target. The output is one plan file. The run is read-only: it never edits a manifest or runs an install, update, or require. Upgrading is the user's next move, with the plan in hand.
The lead (this skill) resolves versions, writes briefs, judges, and writes the plan. Reading changelogs fans out to cheap collectors. Reading the codebase fans out to scouts. The lead re-reads the official upgrade guide for every major version crossed, because a breaking change the collector misses is the costliest miss.
Steps
- Resolve. Split
$ARGUMENTSinto packages, each with an optional@target. Map a loose name to the installed package (laravel→laravel/framework); if several match, ask which one. For each package, runbash <skill-dir>/scripts/versions.sh <package> [target]from the project root. It prints the installed version, the constraint, the latest version, the stable releases crossed, what the target requires, and the installed packages that depend on this one. In another ecosystem (pip, Cargo, Bundler, Go), read the same facts from the lockfile and the registry. A package already at its target leaves the run. Done when every package in the run has installed, target, releases crossed, requires, and dependents. Cascade. Use the script output to find what else has to move:
- A required or peer package whose installed version is outside the target's constraint. It must move.
- A dependent whose constraint excludes the target. It must move to a release that
allows the target. If no such release exists, it is a blocker: name the highest
target it allows. For Composer, confirm with
composer why-not <package> <target>(it also seesreplaced packages that the script misses). - A companion that moves in lockstep (
react+react-dom+@types/react, a framework's first-party packages, one scope's plugins). - A raised runtime (PHP, Node, an extension). This is a requirement item, not a package.
Add each package that must move to the run, with its own target, and run step 1 on it. Repeat until nothing new appears. Past 8 packages, list the rest and ask which to include. Done when the run set is closed, and every blocker and runtime requirement is named.
- Collect and inventory. Split each package's releases into spans: one per major
line, and no more than 20 releases per span. Dispatch all of these in parallel:
one collector per span, and one inventory scout per package. Each brief carries the
run facts and the matching section of BRIEFS.md (Collector or
Inventory). In Claude Code with this plugin installed, use
subagent_type=leek-skills:release-notes-collectorandleek-skills:upgrade-impact-scout. In plain Claude Code, usegeneral-purposewithmodel: haikufor collectors andmodel: sonnetfor scouts. In other harnesses, dispatch in the background per harness-tools, on the cheapest model for collectors and a mid-tier one for scouts, or run them inline one after another. While they run, fetch the official upgrade guide for each major version crossed, and runcomposer auditornpm auditfor advisories against the installed versions. Done when every span has returned (send eachuncoveredlist to one more collector), every package has an inventory, and everybreakingchange in each upgrade guide is an item, added by you if the collectors missed it. - Match. Split each package's items into batches of about 15. Dispatch one match
scout per batch in parallel (same agent and fallbacks as step 3). Each brief
carries the package's inventory and the Match section of BRIEFS.md. Then open
every site behind a
hits,retires, oradoptverdict yourself, and settle everyunsure. Done when every item has a final verdict that you checked. Triage. Put every item with a verdict other than
nonein exactly one bucket. Use the first rule that fits:- Must do: a blocker; a runtime or package requirement the project does not meet;
a
breakinghit; asecurityitem that affects the installed version. - Should do: a
deprecationhit; aretires(a patch, pin, override, or workaround that upstream made redundant: remove it and its comment); anadoptthat deletes code or removes a known bug class; afixfor a bug the code works around. - Do later: any other
adopt, such as a feature that improves the code or the user experience but needs design work, or a refactor whose payoff is small.
Give each item a size: S (under an hour), M (a day), or L (more). Done when every non-
noneitem has a bucket and a size, and thenoneitems are counted.- Must do: a blocker; a runtime or package requirement the project does not meet;
a
- Write the plan in the format in PLAN-FORMAT.md, at
.scratch/upgrades/<package-slug>-<from>-to-<to>.md(the first package names the file). Fill its upgrade checklist with only the lines that apply to this project, with the exact commands. Done when the file exists and every bucket item cites a site and a source URL. - Report in 5–8 lines: the packages and their version jumps, any blockers, the count in each bucket, the three most important Must do items, and the plan path. Start each blocker line with ⚠️; no other marks.
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.

