Use Emendant with Claude Code

A coding agent that upgrades an SDK from memory can write calls that the new release changed or removed. The skill makes Claude Code run Emendant and show you what it reports.

IntegrationsPublished 2026-11-13

The skill is in a public repository, Emendant/emendant-skill. It makes no network request of its own. It runs the same npx emendant commands that you can run yourself.

Install

/plugin marketplace add Emendant/emendant-skill
/plugin install emendant@emendant

Run it with /emendant:emendant, or ask Claude to upgrade a covered SDK. To get /emendant in one repository, copy the skill in:

mkdir -p .claude/skills/emendant
curl -fsSL https://raw.githubusercontent.com/Emendant/emendant-skill/main/skills/emendant/SKILL.md -o .claude/skills/emendant/SKILL.md

What the agent does

  1. It runs emendant scan and shows the report as Emendant printed it.
  2. For each change Emendant does not patch, it runs emendant explain, edits the reported places, and says that Emendant did not verify the edit.
  3. It runs emendant fix. Emendant proves its patches in a copy of your repository and gives each a grade: test verified, typechecked or structurally checked.
  4. With your agreement, it upgrades the dependency and applies the patches.
  5. It runs emendant scan --adopted again, then your typecheck and tests.
  6. It ends with a table of each change, its grade, who made the edit, and the change's page on SDK Changes.

The agent does not add --allow-host-run, --allow-registry-credential, --model-provider or --no-verify unless you ask.

Coverage. Emendant reports the changes in its feed. A clean scan does not show that an upgrade is complete. See supported projects and known limitations.

The full guide is Use Emendant from a coding agent.

Related