addit / harness

Getting started

Install (Claude Code)

No clone, no shell script — install the plugin from inside Claude Code:

/plugin marketplace add addit-digital/addit-harness
/plugin install addit-harness@addit
/addit-harness:setup
  • The plugin (agents/, skills/) tracks the repo via git automatically — no re-run needed to pick up updates.
  • /addit-harness:setup places the parts a plugin can’t carry natively (CLAUDE.md, AGENTS.md, rules/, references/, settings.json) — run it once after installing, and again after an update to re-sync. Add --scope project to confine it to the current project instead of ~/.claude (default), or --link to symlink instead of copy.
  • The plugin itself can also be scoped: /plugin install addit-harness@addit --scope local keeps it to just the current repo; --scope project shares it with collaborators via that repo’s .claude/settings.json; default --scope user is global.

addit-harness supports Claude Code only. If you used the old install.sh, see Migrating from install.sh below.

First run

The first time you start an interactive terminal session after installing, the plugin shows one welcome line. If setup has not run yet, it says to run /addit-harness:setup; once setup has run, it points at /addit-harness:dev-flow <what to build>. Both versions mention /addit-harness:tips.

  • It appears once. The session-start hook creates an empty marker file, onboarding/welcome-v1, in the plugin’s data folder before printing, and stays silent whenever that file exists. If the folder cannot be written, it prints nothing.
  • It appears only at a fresh startup in an interactive terminal session (the hook checks that CLAUDE_CODE_ENTRYPOINT is cli or unset): not under claude -p, and not on resume, /clear or compaction. If your first session is headless, the welcome waits for the first interactive one. The filter is deliberately narrow: other entrypoints (the desktop app, IDE integrations) are not shown the welcome.
  • It is shown to you as a hook message and is not added to the model’s context. Nothing is sent anywhere (see Privacy).

When setup finishes it prints a next-steps line: /addit-harness:dev-flow for non-trivial work, /addit-harness:tips for the short guide, and a reminder that local telemetry stays off unless you turn it on in /config.

/addit-harness:tips prints five short tips: when to use /dev-flow, that you approve the plan before any code is written, where the documents land, how to call the specialist agents (and the Go and React convention commands), and how the local telemetry option works. It is user-invoked only; the model does not run it on its own.

What setup does to your settings.json

Setup merges the plugin’s settings.json into yours instead of replacing it, and backs up your previous file first (~/.claude/.install-backups/<timestamp>/, or .claude/.install-backups/<timestamp>/ for --scope project):

  • Keys the template does not set (hooks, statusLine, env and so on) are kept as they are.
  • permissions lists (allow, deny, ask) are unioned: your own rules stay, first, and the template’s are added after them.
  • enabledPlugins, extraKnownMarketplaces and env are merged key by key with your values winning, so a plugin you disabled stays disabled and an environment variable you set is never overwritten.
  • env gains CLAUDE_CODE_ENABLE_TODO_TOOLS=1 unless you already set that variable. It turns on Claude Code’s task tools, which are off by default on newer models, so /dev-flow can keep a lifecycle task list (see Watching a run). The extra per-turn context this costs has not been measured. To opt out, set your own value for the variable; setup keeps it.
  • model is set to the template’s value (opusplan). Change it back with /model if you prefer another default.
  • With --link, settings.json is symlinked instead of merged.

Re-running is idempotent. On a re-sync, setup also removes reference files that an earlier release shipped and a later one replaced (references/java/java-best-practices.md, references/go/app-erp-conventions.md), but only when the file is unchanged or is a symlink into the plugin. A file you edited is copied to the backup folder and left in place, and setup tells you.

--plugins (opt-in). /addit-harness:setup --plugins also runs claude plugin marketplace add and claude plugin install for every marketplace and every plugin set to true in the placed settings.json (language servers, pr-review-toolkit, figma and others). It needs the claude CLI on PATH. If an install fails from inside a session, run the printed claude plugin ... command in a normal shell.

Migrating from install.sh

install.sh, sync_tools.py and tools.config.json were removed; addit-harness supports Claude Code only. To move to the plugin:

/plugin marketplace add addit-digital/addit-harness
/plugin install addit-harness@addit
/addit-harness:setup [--plugins]

With --scope global, setup retires the unprefixed copies the old script put in ~/.claude/agents and ~/.claude/skills. A copy at least 50% similar (by lines) to the plugin’s version is backed up and then removed; a less similar file with the same name is left alone and reported. Config already synced into another tool keeps working but no longer updates; to keep using the old installer, pin tag addit-harness--v0.3.0.

The @addit-harness:feature-investigator agent was renamed to @addit-harness:product-owner; the old name no longer resolves.

The @addit-harness: namespacing gotcha

Plugin-provided agents are namespaced — invoke them as @addit-harness:code-reviewer, not bare @code-reviewer. This is the most common “why isn’t this working” moment for new installs. See Subagents for the full list.

Verify it worked

Run /agents — you should see @addit-harness:code-reviewer and the other fourteen subagents listed. If they’re missing, re-run /addit-harness:setup and check the plugin installed without errors.

All docs

The sidebar has the full list; here’s the same thing with what each page actually covers, so you don’t have to click through page by page to find it:

Start

Reference

  • Subagents — what each of the 15 subagents does, and its tools
  • Skills & commands — every /slash-command this plugin adds
  • Model & cost — which subagent runs on which model, and how to cut spend
  • Enabling MCP — connecting Jira, databases, and other optional integrations
  • Local telemetry — the opt-in, local-only usage log and its export

Guides

  • Use cases — concrete workflows showing which configs fire together
  • Extending — adding your own conventions, agents, or skills

Project

  • Roadmap — planned work and how to contribute
  • Changelog — unreleased changes and every release
  • Privacy — what the plugin does and does not do with data

Edit this page on GitHub ↗