m4ttstack
OrganizationThe mattstack skill collection for Claude Code: browser, orchestration, and infra skills.
Categories
Indexed Skills (40)
checkout
Use when someone else's branch needs a local worktree for review or testing -- given a branch name, an MR/PR link or number, or a ticket id -- without starting your own work on it. Does not open an editor (checkout-and-open does).
rebase-worktree
Use when one worktree's feature branch has fallen behind the default branch -- 'rebase this worktree', a stale MR branch -- or as the per-branch step of sync-open-mrs. Not for resolving conflicts on its own or moving uncommitted work.
sync-open-mrs
Use when every open MR or PR should be brought current with the default branch in one sweep -- 'rebase all my open MRs', 'my branches are stale, sync them', batch maintenance after the default branch moved.
shepherdr
Shepherd a herd of Claude Code agents via herdr panes. Use when the user wants to fan out work across multiple agents, run parallel brainstorms, delegate parallel tasks, or says 'shepherdr', 'shepherd', 'fan out', 'spawn agents', 'delegate this', 'split this across agents', 'herd this', 'run these in parallel with herdr', 'spread this across my accounts', or asks for multi-account or cswap-aware fan-out.
ship
Use when work on the current branch should leave the machine as an MR or PR -- 'ship this', 'ship it', 'push and open an MR', 'create the PR' -- outside a pipeline run.
stage-evidence
Pipeline stage: capture the before-state evidence the plan committed to, while the before still exists. Reached only through a resolved pipeline; not for direct invocation.
stage-plan
Pipeline stage: triage the approach and commit to it visibly before any implementation. Reached only through a resolved pipeline; not for direct invocation.
stage-provision
Pipeline stage: establish where the unit of work happens -- workspace, branch, ticket. Reached only through a resolved pipeline; not for direct invocation.
stage-ship
Pipeline stage: publish the unit of work for review -- push, open the MR/PR, attach evidence. Reached only through a resolved pipeline; not for direct invocation.
stage-watch-ci
Pipeline stage: watch the pipeline the ship stage triggered and triage a red result. Reached only through a resolved pipeline; not for direct invocation.
watch-ci
Use when the user wants CI watched or triaged outside a pipeline run -- 'watch CI', 'is the pipeline green', 'babysit this MR', or after a push when they want the red/green verdict and failures classified.
work
Use when running a unit of work through a configured pipeline -- 'run the feature pipeline', 'do this ticket end to end', 'start a unit of work', or when a repo's .mattstack/skills.jsonc defines pipelines and a ticket or task should flow through its stages.
receive-review
Use when processing review feedback received on your OWN MR/PR -- "address the review comments", "go through the reviewer's comments and reply", "respond to the review", "handle the feedback on my change". For someone else's change use the domain's review skill; for your own branch before it has feedback use the self-review flow.
review
Use when reviewing someone else's MR or PR before it merges -- a pasted MR/PR link or !iid, 'review this MR', 'check my co-worker's change', 'is this MR solid'. For your own uncommitted work use self-review; for feedback on your own MR use receive-review.
self-review
Use when reviewing, sanity-checking, or gut-checking work this session produced on the current branch, before shipping or between tasks -- "review my work", "is this solid?", "gut-check my changes", "self-review this branch". For a teammate's MR/PR use the domain's review skill instead.
checkout-and-open
Use when reviewing someone else's branch in an editor in one step -- 'check out that branch and open it', 'open this MR locally' -- given a branch, MR/PR link, or ticket id. For a checkout without an editor, use checkout.
map-open-mrs
Use when acting on all of the user's open MRs or PRs at once -- a batch sweep, rebase, or audit that needs each open item paired with the local worktree holding its branch. The discovery step of sync-open-mrs.
stage-gates
Pipeline stage: apply the domain's mandatory gates for the paths this unit of work touches, before implementation. Reached only through a resolved pipeline; not for direct invocation.
stage-implement
Pipeline stage: build the change under the approach the plan stage committed to, test-first. Reached only through a resolved pipeline; not for direct invocation.
stage-self-review
Pipeline stage: fresh-eyes review of the unit of work before it ships. Reached only through a resolved pipeline; not for direct invocation.
cswap-accounts
cswap provider for the account-pool@1 contract: given a herd's model mix and the accounts already assigned this run, names where to launch the next worker. Reached through a wrapper's accounts slot; not for direct invocation.
execution-strategy
Use when deciding how a unit of work should be carried out before it is handed to an executor -- assigning a method to a dispatched worker or subagent, choosing between test-first, plan-driven, and full-design approaches, or writing a brief that must tell its executor which method to run.
gate-protocol
Use when a gated pane or wrapper needs to publish a human decision point and carry it to an answer -- opening the daemon's gate registry, presenting the in-pane form, blocking in gate wait, handling a CAS rejection, or reconciling a doorbell push. Not for direct invocation; a gated verb includes this part.
model-tiering
Use when choosing which model to spawn a sub-agent, worker, or sub-claude on -- any decision point where a less capable model could handle the work. Covers spawn-time selection (shepherd picking worker models) and delegation-time selection (a worker dispatching sub-agents for subtasks).
review-dispatch-body-after
Fresh-context dispatch shapes after the reviewer slot, inlined by the review verbs. Not for direct invocation.
ci-forge-gitlab
Forge adapter: GitLab implementation of ci-forge@1 for the watch-ci stage's forge slot. Reached only through a slot binding; not for direct invocation.
review-core-body-after
Shared review flow body between the criteria slot and the dispatch, inlined by the review verbs. Not for direct invocation.
review-core-body-tail
Review flow tail after the reviewer dispatch, inlined by the review verbs. Not for direct invocation.
review-core-body
Shared review flow body, inlined by the review verbs. Not for direct invocation.
review-dispatch-body
Fresh-context dispatch rule and standard blocks, inlined by the review verbs. Not for direct invocation.
wrap-up-form
Use when wrapping up a session, checking in before continuing, ending a turn with open decisions, when the user asks what you need from them, what decisions are open, or what the next steps are, when they invoke wrap-up or check-in, or when a pipeline gate needs its decision presented as a form.
review-posting
Use when a decided review selection ({levels, disposition}) plus a review draft are ready to post to an MR/PR -- posting inline threads for the selected levels, composing the summary, executing the chosen disposition, and closing out. Not for producing the review, and not for deciding what posts -- the caller decides; this part only executes.
editing-skills
Use when adding, editing, publishing, or debugging why a change isn't live in any mattstack-connected skill surface -- the mattstack plugin, a team pack (acme), or a compiled/vendored pipeline verb built with `rt skills compile` -- e.g. "add a mattstack skill", "why isn't my skill or pipeline change showing up", "rt skills compile / check", "update the work orchestrator", or any change under mattstack-skills, a teams/<team> pack, or a shared work/review engine.
mattstackmodel-tiering
Use when choosing which model to spawn a sub-agent, worker, or sub-claude on -- any decision point where a less capable model could handle the work. Covers spawn-time selection (shepherd picking worker models) and delegation-time selection (a worker dispatching sub-agents for subtasks).
mattstackshepherdr
Shepherd a herd of Claude Code agents via herdr panes. Use when the user wants to fan out work across multiple agents, run parallel brainstorms, delegate parallel tasks, or says 'shepherdr', 'shepherd', 'fan out', 'spawn agents', 'delegate this', 'split this across agents', 'herd this', or 'run these in parallel with herdr'.
gitlab-mr-threads
Use when leaving comments on a GitLab MR via glab -- a positioned inline comment on a diff line, a reply to an existing discussion thread, or when a positioned comment silently landed as a general note instead of a DiffNote.
parameterized-skills
Use when authoring or reviewing a parameterized skill (a wrapper skill that takes other skills as named arguments), declaring metadata slots or provides on a SKILL.md, writing or debugging a .mattstack/skills.jsonc bindings manifest, or wiring resolve-args.sh into a wrapper skill.
run-identity
Use when a standalone verb has just resolved the target its run is about -- recording the run's ticket, branch, and mr fields so the console board and run detail can show them. Not for direct invocation.
getting-current-time
Use only when no hook-injected "Current time:" stamp is in context, or when the work is elapsed-time-sensitive enough that a stamp up to 5 minutes old is too stale. Normally the current time is injected automatically ... a UserPromptSubmit hook stamps every turn and a throttled PostToolUse hook re-stamps during long turns ... so read the most recent stamp from context instead of running anything.
mattstackgetting-current-time
Use whenever the current time matters ... answering "what time is it", timestamping output, scheduling, computing relative dates ("yesterday", "in 2 hours"), or reasoning about timezones. The context window only carries the date, never the time of day, so the clock must be read from the machine.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.