Skip to content

INSTALLATION.md

Latest commit

 

History

History
138 lines (94 loc) · 10.6 KB

INSTALLATION.md

File metadata and controls

138 lines (94 loc) · 10.6 KB
Raw

Installation

coderabbit-threads is built as a Claude Code plugin. Claude Code is the only fully verified runtime, but the skill is structured to work on most agent hosts that load SKILL.md files. This document covers every supported install path.

Requirements

  • gh (GitHub CLI), authenticated: gh auth status must succeed
  • jq on PATH
  • bash 3.2+ (macOS system bash is fine; the script avoids bash-4-only features)
  • An open PR on the current branch with at least one CodeRabbit review thread

No tokens, no extra config. cr uses whatever gh auth login configured.

Claude Code — plugin marketplace (recommended)

/plugin marketplace add tkoehlerlg/coderabbit-threads
/plugin install coderabbit-threads@coderabbit-threads
/reload-plugins

The repo ships a single-plugin .claude-plugin/marketplace.json so /plugin marketplace add points at this repo directly. After install, Claude Code discovers the skill on next invocation, and the plugin loader puts bin/cr on $PATH automatically.

The coderabbit-threads@coderabbit-threads ID isn't a typo. Claude Code plugin IDs are <plugin-name>@<marketplace-name>, and this is a single-plugin marketplace where both names happen to be the same.

Verify the install:

/skills

coderabbit-threads should appear in the list. Invoke it via natural language ("Go through the open CodeRabbit threads on this PR.") or the bundled slash command:

/coderabbit-threads                                       # current-branch PR
/coderabbit-threads 142                                   # explicit PR number on this repo
/coderabbit-threads https://github.com/owner/repo/pull/14 # explicit URL

If the current branch has no PR, the command lists recent open PRs and asks which one to review. It never silently guesses.

Claude Code — manual install (git clone)

For development on the skill itself, or to pin to a fork:

# 1. Clone the repo wherever you like
git clone https://github.com/tkoehlerlg/coderabbit-threads.git ~/code/coderabbit-threads

# 2. Symlink the skill dir into ~/.claude/skills/ so Claude Code discovers SKILL.md
mkdir -p ~/.claude/skills
ln -s ~/code/coderabbit-threads/skills/coderabbit-threads \
      ~/.claude/skills/coderabbit-threads

# 3. Make `cr` callable. Either symlink into a PATH directory:
mkdir -p ~/.local/bin
ln -s ~/code/coderabbit-threads/bin/cr ~/.local/bin/cr

# …or set CR_BIN in your shell rc (the skill honors it):
export CR_BIN=~/code/coderabbit-threads/bin/cr

Since v0.4.0 the binary lives at <plugin-root>/bin/cr — outside the skill directory. The plugin marketplace install auto-PATHs it; manual installs need either a symlink into a PATH dir or CR_BIN.

Verify:

ls ~/.claude/skills/coderabbit-threads/SKILL.md
cr --help 2>&1 | head        # if you symlinked into ~/.local/bin
"$CR_BIN" --help 2>&1 | head  # if you set CR_BIN

Other agent runtimes

The cr CLI is plain bash + gh + jq, so it runs anywhere. The skill runbook (SKILL.md) is platform-aware but degrades to portable shell-only behavior when host primitives aren't available.

Targets below are grouped by integration shape. Tier 1 hosts ship a native Agent Skills system that loads SKILL.md directly. Tier 2 hosts need a thin wrapper file that references the runbook. Tier 3 is phrase-driven only. Only Claude Code is end-to-end verified; everything else is a documented install path, and PRs adding verified-on-X badges are welcome.

For every Tier 1 host other than Claude Code, bin/cr is not auto-added to $PATH. Either symlink cr into a PATH directory at install time, or set CR_BIN=<plugin-root>/bin/cr in your shell rc. The skill already honors CR_BIN.

Tier 1 — native Agent Skills (drop-in)

Runtime Install path Activation Notes
Claude Code Plugin marketplace, or ~/.claude/skills/coderabbit-threads/ /coderabbit-threads, or natural language All four primitives (AskUserQuestion, ScheduleWakeup, sticky approvals, slash command) work natively. Only fully verified runtime.
Cursor 2.4+ ~/.cursor/skills/coderabbit-threads/, or repo .cursor/skills/coderabbit-threads/ /coderabbit-threads, or natural language Same SKILL.md shape. No in-IDE ScheduleWakeup; Step 7 falls back to "re-run in ~2 min", with an optional Cursor Automation cron as a power-user upgrade. Use Auto-Run mode Sandbox or Ask Every Time.
Copilot CLI ~/.claude/skills/coderabbit-threads/ (Copilot reads .claude/skills/ natively), or ~/.copilot/skills/coderabbit-threads/ /coderabbit-threads, or natural language Pre-approve shell calls with copilot --allow-tool='shell(cr),shell(gh),shell(jq)' to avoid per-command prompts.
Copilot Chat (VS Code agent mode) repo .claude/skills/coderabbit-threads/, or ~/.claude/skills/coderabbit-threads/ /coderabbit-threads, or natural language Same skill system as Copilot CLI; runs in the local workspace so gh auth status carries over.
Codex CLI ~/.codex/skills/coderabbit-threads/, or repo .agents/skills/coderabbit-threads/ $coderabbit-threads, /skills picker, or natural language Recommended posture is --sandbox workspace-write --ask-for-approval on-request.
Gemini CLI ~/.gemini/skills/coderabbit-threads/, or gemini extensions install … /skills, activate_skill, or an optional .gemini/commands/coderabbit-threads.toml slash binding Add cr / gh / git to tools.allowed in ~/.gemini/settings.json to skip per-call prompts.

Tier 2 — workflow / rule-file adapter (clean fit, thin wrapper)

These runtimes have their own runbook format. One-liner installer vendors the runbook into <your-repo>/.coderabbit-threads/ and drops the matching wrapper file(s) at the host's expected path:

curl -fsSL https://raw.githubusercontent.com/tkoehlerlg/coderabbit-threads/main/scripts/install-adapter.sh \
  | bash -s -- --host=<windsurf|cline|kilo|continue|zed>

Or from a clone: ./scripts/install-adapter.sh --host=windsurf. Run it from your project root; pass --target=<path> for a different root, --ref=v0.4.1 to pin a version, --force to overwrite existing files.

Drop-in wrappers also ship at adapters/ for manual copy if you'd rather not run a script. Each wrapper is a thin pointer that instructs the host agent to read the vendored SKILL.md at runtime (or, for Continue.dev, transcludes it via Handlebars). See adapters/README.md for details.

Runtime Copy adapter To repo path Activation Notes
Windsurf adapters/windsurf/.windsurf/workflows/coderabbit-threads.md .windsurf/workflows/coderabbit-threads.md /coderabbit-threads Frontmatter auto_execution_mode: 3. Wrapper instructs Cascade to read_file .coderabbit-threads/SKILL.md at runtime (no static @-include for workflows). Add cr to cascadeCommandsAllowList.
Cline (rule) adapters/cline/.clinerules/10-coderabbit-threads.md .clinerules/10-coderabbit-threads.md (always-loaded) Numeric 10- prefix orders it after lower-numbered rules in Cline's alphanumeric concat.
Cline (workflow) adapters/cline/.clinerules/workflows/coderabbit-threads.md .clinerules/workflows/coderabbit-threads.md /coderabbit-threads.md Per-permission auto-approve (Edit / Execute) maps onto the skill's MODE=auto gate. No 60s scheduler.
Kilo Code (mode) adapters/kilo-code/.roomodes .roomodes Mode switch Roo Code's successor — Roo archives 2026-05-15. No user-defined slash commands; pick the mode from the selector, or trigger by phrase.
Kilo Code (rule) adapters/kilo-code/.roo/rules-coderabbit-threads/01-runbook.md .roo/rules-coderabbit-threads/01-runbook.md (auto with mode) Flat-concatenated into the mode body when the mode is active.
Continue.dev adapters/continue/coderabbit-threads.prompt .continue/prompts/coderabbit-threads.prompt (or ~/.continue/prompts/) /coderabbit-threads The only Tier-2 host with native static includes — wrapper uses Handlebars {{{ .coderabbit-threads/SKILL.md }}} to transclude the runbook directly. Set run_terminal_command to Automatic for sticky-style approval.
Zed Agent Panel adapters/zed/.rules .rules (repo root) @coderabbit-threads plus trigger phrase Zed has no rule-file include syntax, so the wrapper instructs the agent to read_file the vendored runbook. Use the Write profile (terminal tool enabled). agent.tool_permissions rule always_allow ^cr gives sticky approval for cr calls.

Tier 3 — phrase-driven (degraded UX)

Runtime Host file Activation Notes
Aider read: SKILL.md in .aider.conf.yml; shell-out via /run cr … Phrase only ("walk the CodeRabbit threads on this PR") No slash registry, no structured prompts, no scheduler. Aider's single-conversation edit-centric model fights the per-thread loop, but the path works.

Not currently supported

  • Copilot coding agent (cloud). Firewalled sandbox, copilot/*-branch-only commits, no interactive Q/A, no path for replying on a human's PR threads. The cloud surface is the wrong tool for this job; coderabbit:autofix covers that lane.
  • Roo Code. RooCodeInc/Roo-Code archives 2026-05-15. Migrate to Kilo Code (above), which inherits the same custom-modes + .roo/rules-{slug}/ model.

Bare cr use — always works

Most of the value lives in cr itself. cr threads, cr context, cr proposed-fix, cr reply, cr resolve are all callable from any shell once gh is authenticated. The SKILL.md runbook is the choreography on top, and you can also follow it by hand.

Primitive-fallback summary

What's missing on hosts outside Claude Code, and how the skill compensates:

  • AskUserQuestion → numbered list with stdin / chat prompt.
  • ScheduleWakeup 60s polling → print "re-run in ~2 min to check CodeRabbit's reactions" and exit.
  • /coderabbit-threads slash command → native on most Tier 1 / Tier 2 hosts; otherwise a natural-language trigger.

The triage logic, the MODE (together/auto) gate, the RESOLVE_POLICY gate, sticky approvals, fix-then-reply, cr proposed-fix, and --since <ref> all work the same on every runtime.