Explains fully — still no preamble, still no closer, but the body runs as long as the topic needs, with headers to skim back.

rm -rf, force push, schema migration, dropping a table: it confirms before acting. Safety wins over brevity.

Stops iterating on code, names the assumption that might be wrong, and asks one diagnostic question.

One short clarifying question beats guessing and rewriting.

/ecc:plan “describe the feature”, then the tdd-workflow skill writes the failing test before the code.

Reproduce it with a failing test first, then use tdd-workflow to implement the fix and prove it passes.

/code-review runs a fresh-context reviewer that never saw the code being written.

/save-session or /learn-eval distils the session into summaries and reusable patterns; /resume-session picks it back up.

/security-scan with AgentShield checks your own hooks, MCP servers, permissions and secrets.

Set the intensity or turn it off; no argument reports the current level. “ultra exists for when the codebase has wronged you personally.”

Reviews the current diff for over-engineering and hands back a delete-list.

Audits the whole repo for over-engineering, not just the diff.

Harvests the shortcuts you deferred into a ledger, so “later” doesn’t become “never.”

Shows the measured impact scoreboard: less code, less cost, more speed.

Rewrites your CLAUDE.md using <important if> blocks to improve instruction adherence.

Narrows React component prop types to match live code paths instead of Storybook, test or mock-only states.

Builds a repo-local skill plus an iterated coding-agent GitHub Actions workflow, prompt, memory file and reference templates.

Interviews you to design a sensor-controller-actuator loop for your codebase, then builds it plus a scheduled coding-agent workflow.

Explains the current topic with concise diagrams, code-shape sketches and focused HTML artifacts.

Inspect a Figma design and implement it in code, create Code Connect templates for your components, or build a screen in Figma — the plugin’s own suggested prompts.

Turn specs into implementation plans, synthesize research into structured docs, prep meeting materials, and capture decisions into Notion pages.

Create products, prices and payment links; retrieve and manage payments, subscriptions, invoices, refunds, disputes and customers.

Jesse Vincent’s skills framework, bundled in OpenAI’s own marketplace: brainstorming, planning, test-driven development, debugging and code review.

Components, services, storage and boundaries. Give it the scope, core components and primary path.

CI/CD pipelines, approvals, tool calls and runbooks. Give it participants, order, branches and exceptions.

API calls, cache fallback, auth and async traces over time. Give it callers, callees, returns and timing.

Pipelines, lineage, PII boundaries and consumers. Give it sources, transforms, stores and boundaries.

States, retries, waits and terminal outcomes. Give it states, events, retry and cancellation paths.

Research a GitHub repo’s architecture, stack, contributors, issues and activity: 5 calls, 62 KB of context instead of 986 KB (94% saved).

Clone a repo and analyze the last 500 commits — top contributors, frequency by month, most-changed files — in one call, 5.6 KB of context.

Fetch the Hacker News front page, extract titles, scores and domains, group by domain: one call, 3.2 KB of context.

Fetch the React useEffect docs, index them, and find the cleanup pattern with code examples: 2 calls, 1.8 KB of context.

Start a multi-step task, let the chat compact after 20+ tool calls, and the model continues from your last prompt with tasks, files and decisions intact.

Run /grill-me or /grill-with-docs first. The agent asks you detailed questions about what you’re building until you and it are aligned.

/grill-with-docs builds a CONTEXT.md glossary of your project’s terms so the agent stops using 20 words where 1 will do.

/tdd makes the agent write a failing test first, then fix it. /diagnosing-bugs wraps debugging into a disciplined, phase-gated loop.

/improve-codebase-architecture surveys the codebase for simplification candidates and hands you the list. Matt suggests running it every few days.

“It’s not just X, it’s Y,” a dramatic one-line closer after every section, “Let’s dive in.” State the point directly.

Forced triads, the same sentence opening three times, dashes as the universal connector. Use the number of items the meaning needs.

“Delve,” “testament,” “marking a pivotal moment,” “experts believe.” Keep the fact, drop the significance.

Bold as decoration, title-case headings with emojis, curly quotes. Sentence case; turn a labeled list into prose.

“Great question! I hope this helps!”, knowledge-limit disclaimers, a heading repeated in the first sentence. Remove the wrapper.

vercel-deploy, netlify-deploy, render-deploy and cloudflare-deploy push a project live; yeet stages, commits, pushes and opens a pull request in one flow.

gh-fix-ci reads failing GitHub Actions logs, summarizes the failure and drafts a fix plan; gh-address-comments works through PR review comments.

figma-implement-design turns a Figma file into application code with 1:1 visual fidelity; a family of figma-* skills covers design systems and libraries.

notion-*, linear and sentry skills plus pdf, playwright (drive a real browser), screenshot, transcribe and speech.

migrate-to-codex carries instruction files, skills, agents and MCP config from .claude/ into Codex’s own files.