catalyst-dev

Fast browser automation CLI for AI agents. **ALWAYS use instead of Playwright MCP tools** for web testing, screenshots, form filling, and UI verification. Use when user says "open in browser", "check the site", "take a screenshot", "fill the form", "test the UI", or any browser interaction. Also use as a fallback when a task requires visual browser interaction that CLIs and APIs cannot handle (e.g., OAuth flows, complex dashboards, visual verification).

Ask/decision tickets ("what needs me") — the Catalyst-wide SOP for raising a human decision as a Linear ticket, replying in-thread as the app actor, and closing it when the answer lands. Use whenever an agent needs a human to decide or click something, or for any ticket labelled `catalyst-ask` / `ask/decision`.

Interactive walk-through of today's morning briefing. Loads the briefing markdown at thoughts/briefings/YYYY-MM-DD.md (built by morning-briefing), parses its decisions block, and walks the user through each open decision — approve / reject / defer, schedule a calendar entry, file a Linear ticket, launch a relay-ticket session, draft an email, resolve ADR drift, or apply/edit/defer/reject a compound-engineering ADR proposal — then writes resolutions back to the briefing markdown. Use after /catalyst-dev:morning-briefing has produced today's briefing, or whenever the user says "walk the briefing" / "follow up on the briefing".

Protocol reference for the Catalyst event broker daemon. Covers agent identity (agent.checkin/ checkout), auto-correlation of ticket↔PR interests, ticket_lifecycle deterministic routing for Linear events, and the existing pr_lifecycle + Groq prose routing paths. Use when an agent needs to wait for events related to its own ticket/PR, watch another ticket's lifecycle, or register any semantic interest in the event stream.

Protocol guide for the `catalyst-comms` file-based agent communication CLI. Use when agents need to coordinate across worktrees, sub-agents, teams, or orchestrators — e.g. orchestrator passes `CATALYST_COMMS_CHANNEL`, user asks to "coordinate with", "ask the other agent", or workers need to share state without HTTP.

Protocol reference for the catalyst-filter semantic event routing daemon. Use when an orchestrator needs to wait for relevant events using a natural-language intent description instead of a precise jq predicate. Covers registration, waiting, deregistration, the wake event structure, prompt writing, context fields, and the fallback path when the daemon is not running.

Automatically analyzes code changes and creates git commits with conventional commit messages when the user indicates they want to save their work, commit changes, or says 'commit this', 'save my changes', 'let's commit'. Auto-detects commit type, scope, and ticket reference from branch name.

Closing ritual for the AI-native estimation feedback loop. **ALWAYS use when** a ticket's PR has merged and merge-pr's post-merge deploy-verification (CTL-2232) has resolved a terminal sentinel for it — the relay-native trigger this skill now shares with `ticket-retro` and `ticket-compound`, see references/trigger.md (CTL-2244) — or when the user says 'compound-estimate', 'close the estimation loop', 'record actuals', 'compound-log', or wants to log post-merge actuals for a shipped Linear ticket. Writes a structured entry (linear.key, pr_number, merged_at, estimate_at_start, estimate_actual, cost_usd, wall_time_hours, what_worked, what_surprised_me) to thoughts/shared/retros/estimate/YYYY-WW-compound-log.md — one file per ISO week, appends.

The ONE agent a human talks to. Use when a human comments on their `Concierge — <human>` ticket, on the status board, or on any scope no steward owns; when a new project must be scaffolded from a request; or when an ask has gone unanswered. Owns the board, the ask inbox, and routing. Never commands a steward.

Create handoff document for passing work to another session. **ALWAYS use when** the user says 'create a handoff', 'hand this off', 'save progress for later', 'I need to stop here', or when context usage is high (>60%) during implementation and work needs to continue in a fresh session.

Create detailed implementation plans through an interactive process. **ALWAYS use when** the user says 'plan this', 'create a plan', 'let's plan the implementation', 'design the approach', or wants a structured TDD implementation plan before writing code. Works best after /research-codebase.

Create pull request with automatic Linear integration. **ALWAYS use when** the user says 'create a PR', 'open a pull request', 'ship this', 'ready for review', or wants to push changes and create a GitHub PR. Handles commit, rebase, push, PR creation, description generation, and Linear ticket update.

Create a git worktree for parallel work and optionally launch implementation session. **ALWAYS use when** the user says 'create a worktree', 'work in parallel', 'start a worktree for', or needs to work on multiple features simultaneously without switching branches.

Generate or update PR description with incremental changes. **ALWAYS use when** the user says 'describe the PR', 'update PR description', 'generate PR description', or after pushing new commits to an existing PR. Supports incremental updates that preserve manual edits.

Fix TypeScript errors with strict anti-reward-hacking rules. **ALWAYS use when** the user says 'fix type errors', 'fix typescript', 'type-check is failing', or when TypeScript compilation errors need to be resolved. Ensures runtime type safety — fixes root causes instead of silencing errors with casts.

Shape every ticket around a scannable use-case before it's filed. **ALWAYS use when** the user says 'file a ticket', 'create a ticket', 'file tickets for', 'open an issue', 'add a ticket', 'log a bug', or whenever drafting, titling, or rewriting a ticket. Turns vague, implementation-first tickets into an outcome title (`<actor> should <outcome> so that <benefit>`) plus tiered Gherkin (Given/When/Then) acceptance criteria — even for backend bugs and chores.

Implement approved technical plans from thoughts/shared/plans/. **ALWAYS use when** the user says 'implement the plan', 'start implementing', 'build from the plan', or wants to execute a previously created implementation plan using TDD (Red-Green-Refactor). Supports team mode for parallel implementation.

Update existing implementation plans based on feedback or changed requirements. **ALWAYS use when** the user says 'update the plan', 'change the plan', 'the requirements changed', 'revise the approach', or wants to modify an existing plan in thoughts/shared/plans/ after review feedback or discovered issues.

Manage Linear tickets with workflow automation. **ALWAYS use when** the user says 'create a ticket', 'update the ticket', 'move ticket to', 'search Linear', or wants to create tickets from thoughts documents, update ticket status, or manage the Linear workflow. Uses Linearis CLI.

Safely merge PR with verification and Linear integration. **ALWAYS use when** the user says 'merge the PR', 'merge this', 'ship it', or wants to merge an approved pull request. Runs tests, checks CI, verifies approvals, squash merges, cleans up branches, and moves Linear ticket to Done.

Reference for the canonical event-driven wait pattern in Catalyst skills. Use when a skill needs to block on a state change (PR merged, CI completed, push to branch, ticket transitioned) WITHOUT polling. Pairs the `catalyst-events` CLI with the Claude Code `Monitor` tool and `wait-for` for short-lived workers.

Generate a daily briefing markdown at thoughts/briefings/YYYY-MM-DD.md with six sections — Review yesterday, Surface decisions, Plan today, relay-dispatch candidates, Friction since last briefing, and Learnings since last briefing — synthesized from Linear, GitHub, Granola, Google Drive, Google Calendar, and the compound-engineering stores in parallel. Then fans the briefing out to four destinations (Slack DM, Slack channel, Notion page, Loom script file). Use when the user says "morning briefing" / "run my briefing", or on a weekday-morning schedule via the CMA Routine wrapping this same skill.

Run the project orchestrator — the long-lived, single-threaded owner of ONE project that moves ready backlog tickets to Todo, lets the fleet scheduler dispatch them, watches the work, and communicates in threaded Linear comments. Use when asked to run/own/coordinate a project or its backlog. It never dispatches a worker; Todo is its only dispatch verb.

Fixes what /catalyst-dev:validate-plan just found, in the same session. Consumes validate-plan's actual output — the Validation Report it renders into the conversation, since validate-plan has no Write tool and produces no verify.json — and applies the fixes directly via Edit/Write, re-runs a scoped gate, and commits. Use right after validate-plan reports FAIL or PARTIAL, especially inside a /relay-ticket session: relay-ticket's phase list names '(→ remediate)' but the only other remediation-shaped skill, phase-remediate, is bound to the daemon-era verify.json contract relay-ticket does not produce — this is the relay-native replacement (CTL-2243). Not for a fresh implementation pass; scope stays to what the Validation Report named.

Conduct comprehensive codebase research using parallel sub-agents. **ALWAYS use when** the user asks to 'research', 'investigate', 'explore the codebase', 'how does X work', 'find out about', or needs deep analysis of how existing code is structured. Produces a research document in thoughts/shared/research/ with file:line references.

Resume work from a handoff document. **ALWAYS use when** the user says 'resume handoff', 'pick up where we left off', 'continue from handoff', or provides a handoff document path. Verifies current codebase state against handoff, validates changes, and creates an action plan.

Systematically pull, categorize, and address all PR review comments — code change requests, questions, and suggestions. This skill fetches comments via gh api, groups them by file, implements fixes, handles disagreements diplomatically, and pushes a single commit. You should not try to handle PR review feedback manually — this skill ensures nothing gets missed. **ALWAYS consult this skill when** the user says 'address comments', 'fix review feedback', 'handle PR comments', 'respond to reviewers', 'address review', 'review feedback', or mentions that a PR has unresolved comments or review threads. Also used by /oneshot Phase 5 to process reviewer feedback before merging.

Scan TypeScript code for reward hacking patterns — shortcuts that make linters pass without actually fixing type safety. This skill has a comprehensive checklist of 8 forbidden patterns with severity tuning (libraries vs apps) that you cannot reliably check on your own. **ALWAYS consult this skill when** the user says 'scan for hacks', 'check for type cheats', 'reward hacking', 'verify no shortcuts', wants to check for `as any`, `as unknown as`, `@ts-ignore`, non-null assertions (`value!`), `forEach(async`, or void tricks after fixing TypeScript errors. Also use after /fix-typescript completes, or when verifying TypeScript changes before marking work done. Accepts optional file/directory arguments to scope the scan.

The long-running owner of ONE initiative or project. Use when asked to run, drive, coordinate or own a project, when a human comments inside a scope you hold, or when resuming from a handoff. Dispatches by launching `/relay-ticket <TICKET>` sessions, reading their reports, answering in threads, keeping status current.

Safely delete orchestrator runtime state — worktrees and ~/catalyst/runs/{orchId} — after archiving artifacts to ~/catalyst/archives/{orchId}. **ALWAYS use when** the user says 'teardown', 'delete orchestrator', 'cleanup orchestrator', or 'remove worktree' for a finished orchestrator. Refuses to delete unless the archive sweep succeeded; use --force to bypass.

Compound-engineering capture + curation for a finished ticket — the engineering feedback loop. Harvests the Friction sections that phase agents left in their artifacts plus the git diff, writes a structured entry to the shared learnings store (thoughts/shared/learnings/), prunes/amends stale notes there autonomously, captures new domain vocabulary to thoughts/shared/CONCEPTS.md, and PROPOSES (for human approval) any ADR change. Non-blocking — runs after a ticket ships or fails, never on the critical path. **Trigger (CTL-2244):** invoke once merge-pr's post-merge deploy-verification (CTL-2232) resolves a terminal sentinel for the ticket's merge — the same relay-native signal `compound-estimate` and `ticket-retro` use, see ../compound-estimate/references/trigger.md. Also use when the user says "compound this ticket", "capture learnings", "what did we learn", or run as /catalyst-dev:ticket-compound <TICKET> [mode:headless].

Cross-ticket retrospective VIEW (CTL-789 Loop C / CTL-814). **ALWAYS use when** a ticket's PR has merged and merge-pr's post-merge deploy-verification (CTL-2232) has resolved a terminal sentinel for it (the workflow's compound closing step — relay-native trigger, CTL-2244; see `../compound-estimate/references/trigger.md`), or when the user says 'ticket retro', 'run a retro', 'retrospective', 'what did we learn lately', or 'how are the estimates calibrating'. Synthesizes everything the compound loops captured since the last retro — friction logs, learnings, compound-log calibration, catalyst.db / merged-PR actuals — into thoughts/shared/retros/ticket/<date>.md with a persisted watch-items block, and surfaces top patterns in the morning briefing's Plan today.

Drive an aging pull-request backlog to zero. Inventories every open PR, finds the STRUCTURAL blockers first (a required check that can never run, a reviewer that never fires, chronically red CI), triages every unresolved review thread in parallel and VERIFIES each finding against the code before fixing it, then merges serially. **ALWAYS use when** the user says 'burn down the PRs', 'stale PRs', 'aging PRs', 'PR backlog', 'get these PRs merged', 'clear the PR queue', or asks why PRs are not merging. Repo-agnostic — works in any repo with the gh CLI.

Validate that implementation plans were correctly executed. **ALWAYS use when** the user says 'validate the plan', 'check if the plan was implemented correctly', 'verify the implementation', or after completing /implement-plan to confirm all phases were properly executed and success criteria met.

Run the full 5-step TypeScript validation gate: type check, reward hacking scan, test inclusion, tests, and lint. This skill provides a structured multi-step pipeline that you cannot replicate on your own — it detects the project's package manager and linter automatically, checks tsconfig strictness, and invokes /scan-reward-hacking internally. **ALWAYS consult this skill when** the user mentions 'validate types', 'check type safety', 'type validation', 'type safety gate', wants to verify TypeScript changes before a PR, after completing a plan phase, or says anything about running type checks + tests + lint together. Even if you think you can run tsc yourself, use this skill — it catches issues you'd miss (tsconfig strictness, test exclusions, reward hacking patterns).

Reference for the bounded-poll pattern — a foreground, single-session, quota-conscious way for a relay worker to wait on a GitHub event (CI completion, review, merge). Replaces the retired broker/filter.wake wait, which needs a daemon, a subagent, or a background loop this model does not have. Use whenever a relay-ticket phase needs to block on GitHub state without a poll storm or a silent stall. Not a slash command — reference doc for skill authors.