# Gitlab Kiosk > Use `glk` as the default for GitLab reads: issues, merge requests, pipelines, CI/CD status, files, releases, and project search. Works with any GitLab instance. Replace `curl`, `fetch`, and `glab` with `glk` for GitLab access to public projects. - Skill: `radiusmethod/gitlab-kiosk` (Agent Skill, multi-file: 3 files) - Install (CLI): `npx skillmds@latest add radiusmethod/gitlab-kiosk` - Raw SKILL.md: https://api.skillmd.com/api/skills/radiusmethod/gitlab-kiosk/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: DevOps & Infra - Author: radiusmethod (https://skillmd.com/u/radiusmethod) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/radiusmethod/gitlab-kiosk --- # glk — read-only GitLab CLI Use `glk` as the default for GitLab reads: issues, merge requests, pipelines, CI/CD status, files, releases, and project search. Works with any GitLab instance. Replace `curl`, `fetch`, and `glab` with `glk` for GitLab access to public projects. **When to use:** Any time a git remote points to the configured GitLab host, use `glk` first for GitLab metadata and one-off reads. ## Choose the right tool - Use `glk` for project metadata, issues, merge requests, labels, pipelines, releases, and one-off file reads. - Use `grec` when the task becomes code exploration: multiple file reads, content search, tracing code across directories, or general repository spelunking. - Use `glk file` for a single file read. Use `grec locate` or `grec checkout` for repository inspection. ## Command guessing rules When guessing commands: - singular noun = one item: `glk issue`, `glk mr`, `glk project`, `glk pipeline` - plural noun = list: `glk issues`, `glk mrs`, `glk pipelines`, `glk releases` - do not try `list`, `view`, or `show` - pass project as a positional argument, not `--project` ## Project auto-detection Inside a git checkout whose remote points to the configured GitLab host, omit `[project]` unless you need a different project. `glk` infers it from `git remote -v` and prints the detected value to stderr. Prefer auto-detection over typing the project path manually. ## Commands glk artifact [project] [--output OUTPUT] # Download a single artifact file from a job glk artifacts [project] [--match MATCH] [--path PATH] [--recursive] # List artifact files in a job glk artifacts download [project] [--archive ARCHIVE] [--dir DIR] # Download a job artifact archive glk code-search [project] [--filename FILENAME] [--group] # Search code in a public GitLab project or group glk doctor # Check local binary and skill health glk epic # Get a single epic by IID glk epic issues [--labels LABELS] [--state STATE] # List issues in an epic glk epic notes [--system] # List comments on an epic glk epics [--labels LABELS] [--search SEARCH] [--state STATE] # List epics for a public GitLab group glk file [project] [--ref REF] # Read file contents from a public GitLab repository glk init [--dry-run] [--overwrite] [--replace] # Create a starter glk config safely glk install-hooks [--dry-run] [--host HOST] # Install glk PreToolUse hooks for Claude and Codex glk install-skills [--dry-run] [--host HOST] [--yes] # Install skill files and shell completions glk issue [project] # Get a single issue by IID glk issue notes [project] [--system] # List comments on an issue glk issues [project] [--labels LABELS] [--state STATE] # List issues for a public GitLab project glk job-trace [project] [--output OUTPUT] [--tail N] # Show job trace output glk labels [project-or-group] [--group] [--scope SCOPE] [--search SEARCH] # List labels for a public GitLab project or group glk mr [project] # Get a single merge request by IID glk mr diffs [project] [--full] # List changed files in a merge request glk mr discussions [project] [--system] # List threaded discussions on a merge request glk mr notes [project] [--system] # List comments on a merge request glk mrs [project] [--author AUTHOR] [--labels LABELS] [--source SOURCE] [--target TARGET] [--state STATE] # List merge requests for a public GitLab project glk pipeline [project] [--interval DURATION] [--timeout DURATION] [--wait] # Get details of a specific CI/CD pipeline glk pipeline-jobs [project] [--exclude-children] [--fail-fast] [--interval DURATION] [--match MATCH] [--timeout DURATION] [--wait] # List jobs for a CI/CD pipeline glk pipelines [project] [--status STATUS] [--ref REF] # List CI/CD pipelines for a public GitLab project glk project [project] # Get metadata for a public GitLab project glk releases [project] # List releases for a public GitLab project glk search # Search public GitLab projects glk tree [project] [path] [--recursive] [--ref REF] # Browse repository directory tree glk uninstall-hooks [--dry-run] [--host HOST] # Remove glk-managed PreToolUse hooks for Claude and Codex glk user-events [username] [--action ACTION] [--days N] [--target-type TARGET-TYPE] # List a user's recent contribution events (defaults to authenticated user) glk users # Search GitLab users by name, username, or email glk whoami # Show the authenticated user's profile glk work-items [--search SEARCH] [--state STATE] # Search epics and issues in a public GitLab group This is the curated agent-facing command reference. For the full runnable command and output inventory, run `glk help-all`. All list commands accept `--limit N` and `--page N`. ## Deriving from a git remote Given a remote like `https://repo1.dso.mil/big-bang/product/packages/kyverno-policies.git`, the project path is `big-bang/product/packages/kyverno-policies`. Strip the host and `.git` suffix. ## Output format Use TSV first. Reach for `--json` only when: - TSV does not include a field you need - you will actually parse nested structure TSV is preferred in agent sessions because it costs fewer tokens and keeps the next action obvious. Example: glk issues --limit 5 glk mrs --state opened --limit 10 artifact: raw artifact file content artifacts: TYPE PATH SIZE code-search: PROJECT_PATH FILE_PATH LINE doctor: GROUP CHECK STATUS DETAIL epic issues: IID STATE TITLE AUTHOR LABELS UPDATED epic notes: ID AUTHOR CREATED BODY epics: IID STATE TITLE AUTHOR LABELS UPDATED file: raw file content issue notes: ID AUTHOR CREATED BODY issues: IID STATE TITLE AUTHOR LABELS UPDATED job-trace: raw job trace content labels: NAME DESCRIPTION COLOR mr diffs: STATUS OLD_PATH NEW_PATH mr discussions: DISCUSSION ID RESOLVED AUTHOR CREATED BODY mr notes: ID AUTHOR CREATED BODY mrs: IID STATE TITLE AUTHOR SOURCE_BRANCH TARGET_BRANCH UPDATED pipeline-jobs: PIPELINE_ID ID KIND NAME STAGE STATUS DURATION DOWNSTREAM_PIPELINE_ID pipelines: ID STATUS REF CREATED releases: TAG DATE TITLE search: PATH DESCRIPTION STARS UPDATED tree: TYPE NAME SIZE user-events: DATE ACTION TARGET_TYPE TITLE WEB_URL users: USERNAME NAME STATE WEB_URL work-items: KIND REF STATE TITLE SCOPE WEB_URL UPDATED Detail commands output key-value pairs. Treat stdout as machine-readable output. Treat stderr as guidance: detected project, pagination notes, and `btw:` next-command suggestions. When `btw:` appears, prefer that command shape over inventing a new shell wrapper. ## Pipeline monitoring workflow When diagnosing a failed pipeline: 1. `glk pipeline --wait` 2. `glk pipeline-jobs ` to find the failed job 3. `glk job-trace --tail 50` for the immediate error For deeper analysis, save the full log locally and use Grep/Read: glk job-trace --tail 0 --output /tmp/job.log If the failure evidence is in artifacts, prefer downloading the full job archive once: glk artifacts download By default, `glk` extracts to a temp directory and prints the path. Use `--dir DIR` only when you need a stable destination. Use `glk artifacts --recursive --match REGEX` only to inspect what exists, and `glk artifact --output FILE` only when you need one known file. `glk artifacts --path` filters the remote artifact tree; it is not a local output directory. ## Recovery rules If a `glk` call fails, check these before reaching for a different tool: - Are you inside a checkout where project auto-detection should work? - Did you guess a noun+verb command that should be a bare noun? - Has the task become code exploration, where `grec` is the better tool? ## Exit codes 0 Success 1 General error (network, parse, config) 2 Safety refusal (private repo, non-public visibility) 3 Not found (404)