Shape agent capabilities into your workflow.

Frame one concrete problem, list the objects already in its world, and apply Action, Variation, and Removal to every object. The skill clusters duplicates, ranks promising ideas, flags open-world trade-offs, and delivers a self-contained HTML report plus a Markdown transcript.

Plan a physical B2B SaaS outreach campaign from targeting through packaging. The skill researches 20–50 prospects, creates a branded PDF for each, groups stops into driving clusters, and exports Google Maps routes plus a My Maps KML; its template writes are limited to a development or internal tenant.

Run a findings-first audit loop: explore high-risk code, verify each candidate with a concrete failure scenario, and report ranked issues before editing. After approval, make minimal fixes, compare with a test baseline, verify using the project's real build or test command, then commit and push without force-pushing.

BetterThink gives Claude Code, Codex, and Gemini CLI six guardrails for coding work: clarify scope, finish and verify one slice before the next, delegate mechanical tasks, keep instruction files lean, load supplementary docs only when needed, and check UI at three viewport widths. It is a portable skill prompt, not a separate test runner.

springy-motion uses a shared duration-and-bounce model for Motion, CSS, and SwiftUI animations. It covers transitions, gestures, drag behavior, and micro-interactions with reduced-motion considerations, and includes a review mode for performance, continuity, interruptibility, frequency, and accessibility. Web and SwiftUI galleries demonstrate the presets.

venture-analyst takes a startup or SaaS idea through problem discovery, competitor intelligence, validation experiments, and a Bull/Bear/Judge verdict. It gathers public signals, extracts pricing and gap clues, and proposes three prioritized tests such as a fake door or concierge MVP. No API keys or paid accounts are required, but the result is a starting point for real validation.

venture-suite packages the venture-analyst skill for Claude Code. It takes an idea through problem discovery, competitor research, validation experiments, and a Bull/Bear/Judge verdict using public sources such as Hacker News, Reddit, GitHub, search, and Google Trends. The pack currently contains one skill and needs no API keys or paid account.

nvc-communication turns complaints, refusals, criticism, apologies, and anger into more usable language using the observation, feeling, need, and request framework. It supports both expressing yourself and empathizing with the other person, with focused paths for requests, boundaries, gratitude, and self-care. The output stays conversational rather than formulaic.

Eagerworks' skills collection shows a portable layout for Agent Skills: each skill has a frontmatter-based SKILL.md, optional references, and installable assets. It works with Claude Code, Cursor, Copilot, Codex, Amp, and other compatible agents through the skills.sh CLI. The current collection contains the Kamal deployment skill, covering deploys, rollbacks, proxy, SSL, builders, and troubleshooting.

LinkedIn Outreach Skill combines a profile screenshot with one business configuration file to create personalized outreach drafts. It scores prospects from role, activity, and buying signals, then suggests a non-pitching first message, follow-up cadence, value comments, and reply handling while filtering spammy or overly sales-focused wording. It is designed mainly for existing first-degree connections and responsible low-volume outreach, not mass automation.

Performance Report Assistant interviews the user, gathers notes and git history, inspects an Excel template, and proposes the exact cells or sections to change before filling a copy. It supports weekly updates, reviews, promotion narratives, and customer reports while preserving workbook structure. Complex mappings may still need confirmation, and the WeCom collector is Windows-specific.

TAR Engine's audit layer discovers and checks SKILL.md files, Codex skill definitions, Claude commands, OpenCode configs, and adjacent helper scripts from the CLI, CI, or an MCP-compatible agent. Static rules run by default; semantic and adversarial analysis requires an explicitly supplied model key. The hosted mode sends submitted skill text to tarai.dev, so sensitive material can use the self-hosted backend instead.

agency-audit inspects a prospect's website for SEO, stack, copy, calls to action, trust signals, and performance clues, then compares three to five competitors. It ranks problems, maps opportunities to agency services, and drafts a customizable proposal. Python plus scrapling, requests, and ddgs are required, but no API key is needed.

marketing-suite bundles five Claude Code skills for agency and sales work: prospect website reviews, retainer opportunity mapping, meeting briefs, personalized cold outreach, and competitor ad analysis. The skills cover the path from researching a prospect to proposing and retaining work. Python, scrapling, requests, and ddgs are installed automatically; individual skills can be installed separately.

Git & GitHub Workflow Skill gives Claude Code stack-agnostic rules for pre-flight checks, atomic commits, branches, pull requests, branch protection, CI, tags, and releases. Its default CI model requires lint, typecheck, test, and build on PRs and pushes to main, with copyable templates included. The rigor may be unnecessary for tiny throwaway repositories.

Use it before design or implementation to resolve conflicting inputs, missing goals, priorities, page states, and no-touch boundaries. After a short diagnostic interview, it creates a read-only design brief plus agent-specific execution constraints for Claude Code, Cursor, or Codex. Checkpoints can require human approval before work continues, helping prevent scope drift and repeated rework.

Run a seven-part pre-launch check covering row-level security, authentication, payment handling, frontend exposure, infrastructure, LGPD privacy duties, and a project-specific manual checklist. The skill labels each item and can fix some frontend or SQL issues automatically. It targets React and Supabase-style stacks and will not find every unknown vulnerability.

Configure bypass permissions for the Claude Code CLI, desktop app, and supported IDE extensions in one command. The script merges settings, creates backups, supports a dry run, and can revert its changes. Because this mode approves file edits, shell commands, and network calls without prompts, it is intended for carefully controlled environments.