Skip to main content
Version: 0.8.3

Grok Build Integration (Native Claude Skills Spec)

Uses the native Claude skills specification. Grok Build (xAI's terminal coding agent) reads Claude Code marketplaces, plugins, skills, MCP servers, agents, hooks, and instruction files (CLAUDE.md, CLAUDE.local.md, .claude/rules/) with zero configuration — same mechanism family as Kilo and Cursor. There is no .agent/ tree and no per-skill adaptation file to maintain.

Prerequisites​

  • A SuperGrok or X Premium+ subscription (Grok Build requirement)
  • Grok Build installed — curl -fsSL https://x.ai/cli/install.sh | bash (macOS/Linux/WSL) or irm https://x.ai/cli/install.ps1 | iex (Windows PowerShell); the command is grok, first launch signs in via browser or XAI_API_KEY
  • Jira auth configured — per-request Basic auth from .jst/jira-sdlc-tools.local.env, no login step (see jira-api-reference.md §9)
  • gh (GitHub CLI) authenticated
  • .jst/jira-sdlc-tools.env and .jst/jira-sdlc-tools.local.env in your project root — see project-config.md

Install / Wire-up Steps​

Structural rule preserved by every method: the three skill folders and _shared must stay siblings. The SKILL.md files reach _shared by relative path, so moving or renaming it breaks them — and never double-nest (cp -r plugins/jira-sdlc .grok/skills/ produces .grok/skills/jira-sdlc/skills/…, which Grok does not discover). Copy the contents of skills/, not the plugin root.

Pick one method. Method A leads — it's the least work.

Method A — Already installed as a Claude Code plugin (recommended, nothing to do)​

If this plugin is installed in Claude Code (it lives under ~/.claude/plugins/…), Grok Build reads it automatically. Run grok in your project and confirm with grok inspect. Invocation keeps the plugin namespace (/jira-sdlc:…, below).

Method B — Point Grok at the repo, no copying​

Add the skills folder to ~/.grok/config.toml (Windows: %USERPROFILE%\.grok\config.toml):

[skills]
paths = ["</PATH>/plugins/jira-sdlc/skills"]

Replace </PATH> with the absolute path where this plugin lives. _shared sits alongside the three skills (no SKILL.md, so it isn't loaded as a skill). No duplication, references intact.

Method C / D — Copy into a project (C) or globally (D)​

Copy the contents of skills/ (trailing /.) so the skill folders land directly under the skills root:

mkdir -p .grok/skills   && cp -r plugins/jira-sdlc/skills/. .grok/skills/    # C: per-repo
mkdir -p ~/.grok/skills && cp -r plugins/jira-sdlc/skills/. ~/.grok/skills/  # D: every project

Invoking the Three Skills​

Open the extensions modal with /skills to confirm the three are listed (grok inspect prints the same discovery from the shell), then invoke:

  • /jira-sdlc:jira-task-assigner — break down a task into Jira issues with branches + worktrees
  • /jira-sdlc:jira-task-executor — implement an issue end-to-end from its worktree
  • /jira-sdlc:jira-task-reviewer — review sub-task PRs from the parent worktree

The /jira-sdlc:… namespace applies to Method A (installed plugin). With a copy-in / config-path install (Methods B–D) the skills load by folder name, so invocation is the bare form (/jira-task-assigner, etc.), and the /jira-sdlc:… cross-references inside the skill bodies must be read as their bare equivalents — same drop-in caveat documented for Cursor. Reopen /skills (or restart the session) after adding skills so they're picked up.

Platform-Specific Caveats​

  • disable-model-invocation: true — expected honoured, verify locally (Unverified on a live Grok Build run: not tested this session). All three skills set this deliberately (explicit invocation only — see root CLAUDE.md). xAI's docs confirm Grok reads Claude skills zero-config and honour user-invocable (→ /<skill-name> slash command) but do not document disable-model-invocation. Because Grok Build targets Anthropic-skill-format compatibility, the expected behaviour is that it honours the flag the same way the other native-spec platforms do (Kilo and Cursor both honour it — verified for those). The native spec has no per-skill override file (unlike Codex's agents/openai.yml), so there is nothing to add: the flag in the shipped frontmatter is the gate. Known gap to check: a cross-platform bug (anthropics/claude-code#38969, mirrored as openai/codex-plugin-cc#211) can make disable-model-invocation: true hide the skill from the command list entirely, blocking even user-initiated invocation. Confirm via grok inspect / the /skills modal that all three skills appear; if one is missing, that's the hide-bug — and since native spec offers no alternative mechanism to reproduce the gate, a Grok version that ignores the flag while needing the gate is a genuine gap to file rather than something this doc can work around.
  • Windows — Grok installs via the PowerShell one-liner above; config lives at %USERPROFILE%\.grok\config.toml. Discovery rules are identical; the skills' own OS dispatch runs the _shared/scripts/win/*.ps1 PowerShell twins instead of the posix/*.sh scripts (see windows-scripts.md).
  • Environment files are read by the plugin's own scripts (e.g. ensure_local_env.sh), not by Grok Build — no Grok-specific env support to configure; keep them where the scripts expect them.
  • .grokignore / .grok/AGENTS.md are not documented Grok features — use .gitignore for VCS and keep project rules in the AGENTS.md/CLAUDE.md families Grok walks from the working directory up to the repo root.