Skills

Skills are reusable capabilities delivered as markdown files that teach Claude Code how to approach specific tasks. Every piece of Catalyst functionality — from committing code to researching a codebase — is a skill.

How Skills Work

There are two types of skills, distinguished by who activates them:

User-invocable skills are structured workflows you trigger with a slash command. They orchestrate multi-step processes — spawning agents, reading context, interacting with you, and saving artifacts.

Model-invocable skills are reference knowledge that Claude activates automatically when it detects relevant context. For example, when Claude sees a ticket ID like ACME-123, the linearis skill activates and teaches Claude how to use the Linearis CLI — without you having to explain it. These skills shape Claude’s behavior the way a README or style guide would, but they load on demand instead of consuming context all the time.

Some skills are both — they can be triggered by you or activated by Claude when relevant.

A third category, CI skills, are non-interactive variants designed for automation pipelines. They follow the same conventions but skip all user prompts.

For more on how Claude Code skills work under the hood, see Anthropic’s skills documentation.

Invoking Skills

Type / followed by the skill name:

/research-codebase
/commit
/create-plan

Claude Code has excellent auto-complete — start typing /res and it will suggest /research-codebase. You don’t need to include the plugin name; Claude Code resolves skills across all installed plugins automatically.

How Skills Are Indexed

Claude Code discovers skills through a two-layer system:

1. Metadata layer — Every skill’s name and description from its YAML frontmatter are always loaded into Claude’s context (~100 words each). This is how Claude decides whether to consult a skill.
2. Body layer — The full SKILL.md instructions are loaded only when the skill is triggered. This keeps context costs low while making the full capability available on demand.

For model-invocable skills, the description is the primary triggering mechanism. Claude reads all skill descriptions and activates the skill whose description best matches the current task. This is why descriptions include specific trigger phrases — they tell Claude “activate me when you see these patterns.”

For user-invocable skills, the / command is the trigger. The description helps Claude understand what the skill does when presenting auto-complete suggestions.

Trigger Contexts — catalyst-dev

Each skill’s description includes specific phrases and contexts that tell Claude when to activate it. Here’s what triggers each dev skill:

Skill	Trigger Phrases / Contexts
research-codebase	”research”, “investigate”, “explore the codebase”, “how does X work”, “find out about”, deep analysis of existing code
create-plan	”plan this”, “create a plan”, “design the approach”, structured TDD implementation planning
iterate-plan	”update the plan”, “change the plan”, “requirements changed”, “revise the approach”
implement-plan	”implement the plan”, “start implementing”, “build from the plan”, executing a TDD plan
validate-plan	”validate the plan”, “check if the plan was implemented correctly”, “verify the implementation”
oneshot	”oneshot”, “do everything end to end”, “full workflow”, ticket-to-merged-PR autonomously
orchestrate	”orchestrate”, “run multiple tickets in parallel”, “coordinate workers”, multi-ticket autonomous execution
code-first-draft	”build this feature”, “implement this PRD”, “code this up”, “create the initial implementation”
commit	”commit this”, “save my changes”, “let’s commit”, saving session work
create-pr	”create a PR”, “open a pull request”, “ship this”, “ready for review”
describe-pr	”describe the PR”, “update PR description”, after pushing new commits
merge-pr	”merge the PR”, “merge this”, “ship it”, merging an approved PR
create-handoff	”create a handoff”, “hand this off”, “save progress for later”, context usage >60%
resume-handoff	”resume handoff”, “pick up where we left off”, “continue from handoff”
linear	”create a ticket”, “update the ticket”, “move ticket to”, “search Linear”
create-worktree	”create a worktree”, “work in parallel”, parallel feature development
setup-orchestrate	”start orchestration”, “set up orchestrator”, “bootstrap orchestrate”, create orchestrator worktree and launch command
fix-typescript	”fix type errors”, “fix typescript”, “type-check is failing”, TypeScript compilation errors
scan-reward-hacking	After /fix-typescript, “scan for hacks”, “check for type cheats”, verifying TS fixes
validate-type-safety	”validate types”, “check type safety”, “run type validation”, before PRs with TS changes
review-comments	”address comments”, “fix review feedback”, “handle PR comments”, “respond to reviewers”
agent-browser	”open in browser”, “check the site”, “take a screenshot”, “fill the form”, visual browser interaction
linearis	Activates when ticket IDs like ACME-123 appear, or when working with Linear CLI
ci-commit	Non-interactive — used by CI pipelines and automation only
ci-describe-pr	Non-interactive — used by CI pipelines and automation only

Legend: User column: checkmark = invoke with /skill-name | Model column: checkmark = Claude activates automatically | CI = non-interactive, for automation pipelines

catalyst-dev

The core development plugin. Skills covering research, planning, implementation, and shipping.

Research & Planning

Skill	User	Model	Description	Source
research-codebase	✓	—	Parallel codebase research with specialized agents. Produces a research document with file:line references.	Source
create-plan	✓	—	Interactive TDD implementation planning. Works best after /research-codebase.	Source
iterate-plan	✓	—	Revise existing plans with research-backed modifications after feedback or changed requirements.	Source

Implementation

Skill	User	Model	Description	Source
implement-plan	✓	—	Execute plans phase by phase using TDD (Red-Green-Refactor). Supports team mode for parallel implementation.	Source
validate-plan	✓	—	Verify implementation against plan success criteria and TDD adherence.	Source
oneshot	✓	—	End-to-end autonomous workflow — research, plan, implement, validate, ship, and merge with context isolation between phases.	Source
orchestrate	✓	—	Multi-ticket parallel coordinator — dispatches Level 2 workers across worktrees with wave-based execution and adversarial verification.	Source
code-first-draft	✓	—	Initial feature implementation from a PRD or feature description. Also generates standalone prototypes when no codebase exists.	Source
fix-typescript	✓	—	Fix TypeScript errors with strict anti-reward-hacking rules. Ensures runtime type safety over silencing errors.	Source
scan-reward-hacking	✓	—	Scan for forbidden patterns (as any, @ts-ignore, non-null assertions, async issues) in recent changes. Companion to /fix-typescript.	Source
validate-type-safety	✓	—	5-step type safety gate: type check, reward hacking scan, test inclusion, tests, lint. Detects project tooling automatically.	Source

Shipping

Skill	User	Model	Description	Source
commit	✓	—	Auto-detect commit type, scope, and ticket reference. Conventional commit format for changelog generation.	Source
create-pr	✓	—	Full PR creation: commit, rebase, push, create PR, generate description, and update Linear ticket.	Source
describe-pr	✓	—	Generate or incrementally update PR descriptions. Preserves manual edits across updates.	Source
merge-pr	✓	—	Safe squash merge with test execution, CI verification, approval checks, branch cleanup, and Linear update.	Source
review-comments	✓	—	Pull PR review comments, analyze context, implement fixes, and push updates. Used by /oneshot Phase 5.	Source

Session Management

Skill	User	Model	Description	Source
create-handoff	✓	—	Save session context, learnings, and next steps for continuation in a fresh session.	Source
resume-handoff	✓	—	Resume work from a handoff document. Verifies codebase state and creates an action plan.	Source
create-worktree	✓	—	Create git worktree for parallel development without switching branches.	Source
setup-orchestrate	✓	—	Bootstrap an orchestrator worktree and print a ready-to-run launch command for /orchestrate.	Source

Integrations & References

Skill	User	Model	Description	Source
linear	✓	—	Linear ticket operations: create from thoughts documents, update status, manage workflow.	Source
agent-browser	—	✓	Browser automation CLI reference — activates when visual browser interaction is needed (OAuth, dashboards, screenshots).	Source
linearis	—	✓	Linearis CLI reference — activates when ticket IDs like ACME-123 appear or when Linear CLI syntax is needed.	Source

CI / Automation

Skill	User	Model	Description	Source
ci-commit	—	CI	Non-interactive variant of /commit for CI pipelines. Never prompts the user.	Source
ci-describe-pr	—	CI	Non-interactive variant of /describe-pr for CI pipelines. Auto-detects current PR.	Source

catalyst-pm

Project management workflows. 46 skills covering strategy, research, planning, and reporting.

Reporting & Analysis

Skill	User	Model	Description	Source
analyze-cycle	✓	—	Cycle health report with risk analysis and recommendations	Source
analyze-milestone	✓	—	Milestone progress toward target dates	Source
report-daily	✓	—	Quick daily standup summary	Source
groom-backlog	✓	—	Backlog health analysis and cleanup	Source
sync-prs	✓	—	GitHub-Linear PR correlation and gap identification	Source
context-daily	✓	—	Context engineering adoption dashboard	Source

Product Strategy

Skill	User	Model	Description	Source
define-north-star	✓	—	Define north star metrics and strategic goals	Source
write-prod-strategy	✓	—	Write product strategy documents	Source
strategy-sprint	✓	—	Run a strategy sprint session	Source
expansion-strategy	✓	—	Plan expansion and growth strategies	Source

User Research

Skill	User	Model	Description	Source
interview-guide	✓	—	Create structured interview guides	Source
interview-prep	✓	—	Prepare for user interviews	Source
user-interview	✓	—	Conduct and document user interviews	Source
interview-feedback	✓	—	Process and organize interview feedback	Source
user-research-synthesis	✓	—	Synthesize findings from multiple research sessions	Source
journey-map	✓	—	Create user journey maps	Source

Feature Development

Skill	User	Model	Description	Source
prd-draft	✓	—	Create a modern PRD with guided questions and multi-agent review	Source
prd-review-panel	✓	—	Multi-agent PRD review panel	Source
feature-metrics	✓	—	Define and track feature-level metrics	Source
feature-results	✓	—	Analyze feature launch results	Source
launch-checklist	✓	—	Create comprehensive launch checklists	Source

Experimentation

Skill	User	Model	Description	Source
experiment-decision	✓	—	Make data-driven experiment decisions	Source
experiment-metrics	✓	—	Design experiment metrics and success criteria	Source
impact-sizing	✓	—	Size the impact of proposed changes	Source

Meetings & Communication

Skill	User	Model	Description	Source
meeting-agenda	✓	—	Create structured meeting agendas	Source
meeting-notes	✓	—	Transform meeting transcripts into structured action items	Source
meeting-cleanup	✓	—	Clean up and organize meeting artifacts	Source
meeting-feedback	✓	—	Process meeting feedback	Source
slack-message	✓	—	Draft Slack messages for various contexts	Source
status-update	✓	—	Generate status updates for stakeholders	Source

Planning & Prioritization

Skill	User	Model	Description	Source
daily-plan	✓	—	Create daily work plans	Source
weekly-plan	✓	—	Create weekly work plans	Source
weekly-review	✓	—	Conduct weekly review sessions	Source
prioritize	✓	—	Prioritize features and work items	Source

Prototyping

Skill	User	Model	Description	Source
prototype	✓	—	Build quick prototypes	Source
generate-ai-prototype	✓	—	Generate AI-powered prototypes	Source
prototype-feedback	✓	—	Collect and organize prototype feedback	Source
napkin-sketch	✓	—	Quick napkin-sketch ideation	Source

Analysis

Skill	User	Model	Description	Source
competitor-analysis	✓	—	Conduct competitor analysis	Source
retention-analysis	✓	—	Analyze user retention patterns	Source
activation-analysis	✓	—	Analyze user activation funnels	Source
metrics-framework	✓	—	Set up leading vs lagging indicators	Source

Other

Skill	User	Model	Description	Source
decision-doc	✓	—	Create structured decision documents	Source
create-tickets	✓	—	Create Linear tickets from requirements	Source
connect-mcps	✓	—	Connect and configure MCP servers	Source
ralph-wiggum	✓	—	Devil’s advocate PRD/document review — surfaces risks and assumptions with adversarial questioning	Source

catalyst-analytics

PostHog integration for product analytics. ~40K token context cost — enable only when needed.

Skill	User	Model	Description	Source
analyze-user-behavior	✓	—	User behavior patterns and cohorts	Source
segment-analysis	✓	—	User segment analysis for targeted insights	Source
product-metrics	✓	—	Key product metrics and conversion rates	Source

catalyst-debugging

Sentry integration for production error monitoring. ~20K token context cost — enable only when needed.

Skill	User	Model	Description	Source
debug-production-error	✓	—	Investigate production errors with Sentry data	Source
error-impact-analysis	✓	—	Analyze error impact across users and releases	Source
trace-analysis	✓	—	Trace error paths through the stack	Source

catalyst-meta

Workflow discovery and management for advanced users and plugin developers.

Skill	User	Model	Description	Source
discover-workflows	✓	—	Research external Claude Code repositories for workflow patterns	Source
import-workflow	✓	—	Import and adapt workflows from other repositories	Source
create-workflow	✓	—	Create new agents or skills from templates	Source
validate-frontmatter	✓	—	Check frontmatter consistency across all workflows	Source
audit-references	✓	—	Audit plugin health and find broken references	Source
reorganize	✓	—	Analyze and reorganize directory structures	Source