Shape agent capabilities into your workflow.

skills-best-practices is a guidance-oriented repository for creating and maintaining agent skills. Its documented workflow covers skill descriptions, example phrases, model-assisted validation, and organizing a reusable skill library while keeping context lean. The verifiable repository content centers on a SKILL.md and supporting guidance rather than a standalone desktop application.

This Claude skill is designed for Customer Success Managers preparing Executive Business Reviews and Quarterly Business Reviews. It focuses on organizing review material into executive-facing insights and a coherent slide structure, giving users a clearer narrative and presentation sequence for customer business-review meetings.

Upload or point to a DDEX ERN XML file to detect its version, validate it against official schemas, and check fields such as ISRCs, UPC or EAN codes, territories, and required metadata. The result includes a release summary, error locations, and concrete fix suggestions. It is designed as a proof of concept for pre-delivery checks by labels and distribution teams.

Ionic Capacitor Skill teaches Claude Code the expected page structure, lifecycle hooks, forms, theming, navigation, and native plugin patterns for React, Angular, and Vue projects. Its references cover Ionic components, official Capacitor plugins, and production concerns such as debugging, accessibility, security, performance, and deployment. It is intended to prevent generic web patterns from replacing framework-appropriate mobile implementations.

m_AI_l coordinates ongoing projects with a roster, per-thread mailboxes, and archives stored as ordinary Markdown files. Each Claude conversation can leave messages, act on its inbox, and hand work to a successor when its context fills, without a database or custom server. It is aimed mainly at non-developers seeking minimal setup and requires Claude Desktop with filesystem MCP access; the web version is not supported.

DeepThink starts by clarifying why a topic matters to you, researches it through Perplexity, optional X/Twitter search, and curated RSS feeds, then shifts into a Socratic session with follow-up questions and counterarguments. It stores topics, research, sessions, and changes in your beliefs across conversations. A Perplexity API key is required for the core research stage, while the X source is optional.

After installing the skill, its specialist agents, and the `bzrk` CLI, you can ask for error searches, trace analysis, metric exploration, service health checks, or production incident triage in natural language. The main skill routes each request to a focused log, trace, metric, cluster, or incident agent so detailed query output does not overwhelm the primary conversation.

In Create mode, DesignLint makes the agent understand the user, choose a creative tension, and commit to nine decisions across aesthetics, typography, color, space, motion, interaction, voice, and departure before implementation. In Audit mode, it documents the current UI, scans for repetitive AI patterns, scores the design, and proposes the three highest-impact improvements. It supports Claude Code, Cursor, Codex, and other agents.

The skill examines relevant work across static code, dependencies, secrets, infrastructure, authentication and authorization, business logic, and runtime observability. It can generate hardened code, threat models, bug-bounty reports, and compliance gap assessments mapped to OWASP, CWE, NIST CSF, ISO 27001, and CIS Controls. Each result includes a scorecard and residual risks rather than claiming the system is secure.

Classify Claude Code actions such as reads, file writes, network requests, credential handling, and destructive commands into proportional risk tiers. Routine work can continue, while sensitive actions trigger notices, approval pauses, or stops and are logged for review. This repository is the Markdown-only skill, so enforcement depends on model compliance; the companion plugin adds deterministic hooks.

This skill produces top-down development summaries for busy, non-technical readers. It leads with the bottom line, groups changes by outcomes and implications, surfaces material open issues, and ends with one concrete next step instead of replaying the work chronologically. Ready-to-install wrappers are provided for both Claude Code and Codex.

ELI5 detects who the explanation is for and adjusts vocabulary, analogies, tone, depth, and framing accordingly. It can explain code to a manager, simplify an error for a parent, or describe a technical concept to a child without using the same generic style for every audience. The skill installs in Claude Code and triggers from ordinary requests.

This skill reviews an X post against the engagement signals described by the open-source Phoenix recommendation system, then suggests a format, rewrites, and a posting plan. Its weighted score also incorporates observed 2026 engagement data. Because the released source does not include the platform's exact private weights, the result is an optimization framework rather than a precise reach forecast.

overnight-skill audits a codebase, asks the user to confirm protected files, and then launches fresh Claude Code sessions from a tmux runner to fix queued issues. Each iteration runs verification, commits and pushes to a dedicated branch, and reverts on failure, while a draft PR, progress log, and morning review capture the outcome. It runs headlessly with bypassPermissions, so its shell-enforced safeguards and scope require careful review before use.

emailwriter helps Claude create cold emails, welcome series, abandoned-cart flows, win-back campaigns, and other lifecycle messages. Outputs can include subject-line alternatives, preview text, follow-up timing, and notes on the frameworks used, along with deliverability and compliance considerations for different industries. The material supports structured writing and testing, but it does not guarantee replies, meetings, or revenue.

Triad reviews a Notion partner database alongside recent Gmail and Calendar activity to identify two people who serve the same buyers with complementary, non-competing offers. It drafts a concise introduction and records the connection back in Notion, supporting a repeatable relationship-building routine. A compatible Notion setup plus Gmail and Google Calendar access are required.

AWRSHIFT gives coding agents a single adaptive flow for identifying a task, researching it, defining success metrics, comparing hypotheses, fact-checking a plan, and testing inside an experiments folder. The user chooses what happens at every phase transition. Main project files remain read-only until a GO decision and an explicit file-by-file implementation preview are approved.

openskill reads signals such as CLAUDE.md, package.json, and framework configuration to understand a project before looking for relevant Claude skills. It searches installed sources, community repositories, and an official registry when available, filters duplicate suggestions, and lets the user choose global or project-scoped installation. Searches can also be narrowed to areas such as frontend, backend, testing, or API design.