Team Skills
The features you use. Every entry-point skill is a slash command you can run (
/team,/team-fix, …). There are no methodology skills: what agents used to preload now lives in ordinary playbooks and references they read by path. A principle is a guarded command — explicitly invoked, never applied by the model on its own — that a consuming procedure reads by installed path.Source of truth: the skill bodies themselves,
skills/*/SKILL.md. This page is a hand-maintained reference. When it disagrees with aSKILL.md, theSKILL.mdwins.
Each entry starts with one sentence copied from that skill’s frontmatter
description. **Uses:** lists the skills this skill consumes: the ones it
loads through the Skill tool and the ones whose SKILL.md its files read by
path. **Used by:** lists the skills that consume it. Both use
comma-separated lists, or None.
The load form is
Call the Skill tool with `<name>`. The path form is a reference to
<name>/SKILL.md, relative or root-relative. A reference to any other file in
a skill’s directory is neither a load nor a use, and neither is a reference
that only locates a skill’s install directory — “the directory containing
skills/<name>/SKILL.md” names a location, not the skill.
The edges are therefore directed, and reading them transitively gives the
graph. team-implement loads team-pr, which loads pr-screenshots. None of
the three loads back. **Uses:** None marks a
leaf with no further consumption.
References are collected across every .md file in the skill’s directory, so a
reference written in a references/ file counts the same as one in SKILL.md.
This page carries both directions of each skill-to-skill edge.
For what separates a load from a citation, and for how a skill is loaded, see
architecture.md §6.
The catalog has 27 registered skills, all commands. The six shared principle documents, playbooks, templates, and operational rules are ordinary installed resources. They are read at the consuming operation and add no picker entries.
Entry-point skills
Each carries argument-hint, so it is a slash command, and each either kicks off a
full run or drives one phase of the QRSPI pipeline.
team
Runs the 8-phase QRSPI feature pipeline, or a leading-argument route.
Used by: None
Uses: how, team-implement, team-pr, team-worktree, why
team-question
Decomposes a feature into task and question artifacts.
Used by: None
Uses: None
team-research
Researches a codebase area before changes.
Used by: None
Uses: None
team-design
Drafts and adversarially reviews a design.
Used by: None
Uses: None
team-structure
Breaks a reviewed design into verified slices.
Used by: None
Uses: None
team-plan
Produces the tactical implementation plan.
Used by: None
Uses: None
team-worktree
Prepares isolated git worktrees.
Used by: team, team-fix
Uses: pr-cleanup
team-implement
Executes and verifies implementation slices.
Used by: team
Uses: team-pr
team-pr
Opens PRs with project terms, evidence, and risk.
Used by: team, team-implement
Uses: pr-screenshots
team-fix
Runs the compressed bug-fix pipeline.
Used by: None
Uses: pr-screenshots, principle-fix-root-causes, team-worktree, why
eng-design-doc-review
Reviews a technical design document with fresh context.
Used by: None
Uses: None
Standalone utilities
Each carries argument-hint (so it is a slash command) but is not a
QRSPI phase: a self-contained action a user runs on demand.
shipit
Lands a reviewed pull request.
Used by: None
Uses: None
pr-open-comments
Triages unresolved PR review comments.
Used by: pr-watch-as-author, pr-watch-as-reviewer
Uses: pr-screenshots
pr-watch-as-author
Watches an authored PR for feedback.
Used by: None
Uses: pr-open-comments
pr-watch-as-reviewer
Watches a reviewed PR and approves settled feedback.
Used by: None
Uses: pr-open-comments
groom-backlog
Grooms a project backlog and proposes tracker changes.
Used by: None
Uses: None
pr-cleanup
Cleans PR state.
Used by: pr-rebase, team-worktree
Uses: None
pr-verify
Verifies a PR test plan with evidence-rated verdicts.
Used by: None
Uses: None
pr-screenshots
Attaches local images to a PR body.
Used by: pr-open-comments, team-fix, team-pr
Uses: None
pr-rebase
Rebases a branch onto its base.
Used by: None
Uses: pr-cleanup
retro
Mines a session for durable learnings.
Used by: None
Uses: None
why
Investigates design rationale behind code.
Used by: code-review, how, team, team-fix
Uses: how
how
Explains subsystem architecture and runtime flow.
Used by: team, why
Uses: why
code-review
Reviews a diff with fresh context.
Used by: None
Uses: why
no-comments
Removes low-value source comments and encodes valid constraints.
Used by: None
Uses: None
agent-prompt
Composes an agent-optimized prompt for a task.
Used by: None
Uses: None
Principles
Guarded commands: an explicit invocation starts them; the model never applies them on its own. Consuming procedures read them by installed path.
principle-fix-root-causes
Requires diagnosis and repair of root causes.
Used by: team-fix
Uses: None
Prose composition and evaluation
The writing standards reference protects exact text and semantic force, scans the untouched draft, and records the applicable checklist before style edits. It then resolves every recorded match, rescans, and self-audits meaning and protected text.
The live-model suite covers zero, one, and many matches; exact code, user, and vendor text; pipeline authors; the technical-writer semantic veto; the fresh DESIGN reviewer; isolated Research producers and assembly; nested helpers; and the named-parent fallback. These evals are stochastic, paid, periodic, and non-gating. Their assertions use required contracts and score floors rather than expecting identical prose across runs.
Name-collision pairs
Several skills and agents share a stem, which is an easy trap. The pattern is consistent: the skill is the orchestrator or methodology, while the agent is the specialist that does the work.
| Skill | Agent | How they differ |
|---|---|---|
team-research |
researcher |
Skill dispatches the Research phase. The agent is the doer that runs the research. |
team-question |
questioner |
Skill drives the Question phase. The agent decomposes the intent. |
team-design |
design-author |
Skill drives the Design phase. The agent drafts the alignment doc. |
eng-design-doc-review |
design-author |
The review skill dispatches a read-only Explore subagent, not the design-author agent, which keeps the audit independent of the author. |
See also
- Architecture: the design rationale behind skills (two flavors, three-tier discovery, load limits) in §6.
- Vision: the loop-driven end state Team builds toward.
- Ethos: the principles behind the pipeline.
- Overview: the landing page and pipeline overview.
skills/team/registry.json: the phase-tagged inventory of the 13 specialist agents, in the source tree.
Shared resources
Read these ordinary documents at their consuming step. They add no registrations or picker entries.
- bug fix
- bug diagnosis
- boil the ocean
- durable state
- focused work
- human control
- independent review
- verified results
- feature playbook
- question playbook
- research playbook
- design playbook
- structure playbook
- plan playbook
- implement playbook
- verify playbook
- decisions
- dependencies
- code standards
- writing standards
- design template
- structure template
- testing
- PRD template
- execution
- external data
- code reviewer brief
- security reviewer brief
- documentation reviewer brief
- ux reviewer brief
- finding format
- design reviewer brief
- comment reviewer brief
- cross-model review
- agent dispatch
- commit discipline
- changelog rules
- tracking rules
- watch loop
- worktree playbook
- cleanup playbook