Goose

Works with the CLI alone. No gander.md account.

Goose is Block's open source AI agent. It can install, execute, edit, and test with any LLM, not only suggest code.

Block's Goose loads skills from ~/.agents/skills and .agents/skills. It still discovers .goose/skills, ~/.config/goose/skills, and Claude dests. gander skill already links ~/.agents/skills/gander. MCP is an extension in ~/.config/goose/config.yaml.

1. Install the CLI

curl -fsSL https://release.gander.md/install.sh | bash
gander --version

After Homebrew or a source build, run gander skill.

2. Confirm the skill landed

ls -l ~/.agents/skills/gander/SKILL.md

Optional Goose-native dest:

mkdir -p ~/.config/goose/skills
ln -sfn ~/.gander/skill ~/.config/goose/skills/gander

Start a new Goose session. Ask which skills are available.

3. Sign up and add MCP

gander signup --email [email protected]

gander mcp install does not write Goose config. In ~/.config/goose/config.yaml:

extensions:
  gander:
    enabled: true
    type: stdio
    name: gander
    cmd: gander
    args: ["mcp"]
    timeout: 300

Or run goose configure, pick Extensions, add a stdio extension named gander. Restart Goose.

4. Use it

1. Ask Goose for a plan as markdown, then gander this file.
2. gander watch plan.md after signup.
3. Send the hosted URL.
4. @agent comments come through the Gander extension. Goose edits the local file.

Fallback: gander comments.

5. Verify

gander list
gander comments

Goose should list Gander as a skill and expose gander_list_comments, gander_reply_comment, gander_resolve_thread, gander_unresolve_thread.

More: Skill, MCP, Hosted watch.