agentsclimarketplace

Setup

Skill MisterRaindrop/gitlab-mr-guardian/skills/setup

Configure GitLab MR Guardian by persisting the hostname and safety options into the plugin data directory, then verify glab authentication and run a read-only status check. Use when the user asks to initialize, enable, or change automatic MR monitoring, rebasing, auto-merge, project filters, or the GitLab host.From its SKILL.md

Install
npx -y skills add MisterRaindrop/gitlab-mr-guardian --skill setup

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

5.2 KB, as published. Nobody here has run it

Set up GitLab MR Guardian

  1. Show the currently effective configuration and monitor switch:

    "${CLAUDE_PLUGIN_ROOT}/bin/gitlab-mr-guardian" \
      --plugin-data-dir "${CLAUDE_PLUGIN_DATA}" \
      configure
    

    The CONFIG_SHOW output includes config (effective values) and monitor_enabled (the real persistent background-polling switch).

  2. Determine the GitLab hostname: use $ARGUMENTS when supplied; otherwise keep the already-configured hostname when it is non-empty; otherwise infer it from git remote get-url origin; otherwise ask the user.

  3. Run glab auth status --hostname <hostname>. If it reports a token problem, do not stop there: confirm with a real read call such as glab api user --hostname <hostname>, because glab auth status can report an invalid token while API access still works. Never print or request an access token when glab API calls succeed. Warn the user that write operations (rebase, pipeline trigger, auto-merge) may still fail if the token lacks api scope.

  4. Explain the two mutation switches: auto_rebase is on by default and rewrites the source branch, but only when GitLab reports need_rebase and the MR is managed (offer --auto-rebase false to users who do not want it). Note that a rebase is allowed by default even when it clears existing approvals, because a managed MR has already been reviewed; mention --allow-rebase-that-resets-approvals false to anyone who needs those approvals preserved. auto_merge is off by default and can merge code immediately once all guarded checks pass; obtain confirmation before enabling it unless the user already explicitly requested it in the current turn.

  5. Persist the configuration into the plugin data directory. Pass only the options the user chose to set; omitted options keep their previous or default values:

    "${CLAUDE_PLUGIN_ROOT}/bin/gitlab-mr-guardian" \
      --plugin-data-dir "${CLAUDE_PLUGIN_DATA}" \
      configure \
      --hostname "<hostname>"
    

    This writes settings.json inside ${CLAUDE_PLUGIN_DATA}. The background monitor and all other commands read configuration from this file; native plugin userConfig substitution is not used. Never create or edit configuration inside the user's repository.

    Blocker-clearing rules are on by default, because a managed MR has already passed human review: max_pipeline_retries (retry a failed pipeline's jobs up to 10 times, one attempt per polling cycle), rebase_when_ci_failed (allow safe rebase for need_rebase MRs while CI is not yet successful — failed, running, missing, or skipped; a running pipeline on the stale base is superseded), allow_rebase_that_resets_approvals and allow_rebase_when_approval_policy_unknown (rebase even when it would clear approvals or the approval policy cannot be read). Turn any of them off with --<flag> false when the user wants a more conservative posture — for example --allow-rebase-that-resets-approvals false to preserve approvals.

    Optional extensions that are off by default: --require-successful-pipeline-history true (restrict management to approved MRs whose history already shows a successful pipeline; by default approval alone is enough) and --advisory-reviewers <bot,names> (comma-separated usernames, e.g. AI review bots, whose unresolved discussions are advisory and never block automation). Explain the behavior change and confirm before enabling either unless the user already explicitly requested it.

    Polling visibility: quiet cycles report a MONITOR_HEARTBEAT by default, so a silent Monitor is distinguishable from a dead one. Every heartbeat wakes the session, so when the user sets a short poll_interval_seconds, point out --heartbeat-min-interval-seconds <seconds> to throttle idle reports (real events always report immediately regardless), or --heartbeat-when-idle false to silence them entirely and rely on monitor.last_cycle in status.

  6. Run a read-only status check and summarize which merge requests are monitored, paused for review, waiting for CI, or blocked:

    "${CLAUDE_PLUGIN_ROOT}/bin/gitlab-mr-guardian" \
      --plugin-data-dir "${CLAUDE_PLUGIN_DATA}" \
      status
    
  7. Report the real monitoring state from the monitor_enabled field of the CONFIG_SAVED output — do not assume polling is off. If it is true, background polling is already active and will pick up the new configuration on its next cycle. If it is false, tell the user that /gitlab-mr-guardian:start enables background polling and /gitlab-mr-guardian:stop disables it; polling runs only while an interactive Claude Code session is open.

Do not weaken the approval or unresolved-discussion checks — those still gate every mutation. allow_rebase_that_resets_approvals is on by default, so tell the user a rebase can invalidate approvals and set it to false when they want them preserved.

Keep looking

Skills are one crate of 608,666. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.