Shape agent capabilities into your workflow.

Trigger Sessionclips during a Claude conversation and it selects useful lessons, shipped work, opinions, receipts, process notes, or funny moments, then drafts three to five short social posts with character counts. Each draft follows the rules in an editable voice.md file, while sessions containing credentials, unpatched vulnerabilities, private client work, or other flagged material are refused.

These installable skills teach an AI agent to use the Chia Health MCP for browsing GLP-1, peptide, and longevity treatments, checking eligibility, verifying the patient, completing medical intake and consent, ordering, paying, and managing follow-up care. The workflow serves US telehealth operations, while licensed healthcare providers—not the skills or MCP—evaluate and make prescription decisions.

Give Claude a claim, contract, or argument and it responds as a rigorous opponent, identifying the most dangerous weaknesses, unanswered questions, and plausible worst-case outcome. The skill is written primarily for Russian-language legal work but can also challenge business ideas and general reasoning. Its output is preparation material, not a substitute for qualified legal advice.

Talk Slow replaces long, one-shot explanations with progressive disclosure: a brief headline first, then one concept or step at a time, with a chance to redirect before more work is produced. It is designed for plans, debugging walkthroughs, technical explanations, and large code changes, while simple questions still receive a normal direct answer. You can disable the mode or request the full response at any time.

This Claude Code skill analyzes a repository, proposes modules and effort estimates, then creates an Epic → Module → Subtask structure in Jira and advances issues only after testing. It resolves project-specific issue and transition IDs dynamically, verifies that the target Epic belongs to the user before destructive operations, and defaults deletions to dry-run. It requires Jira Cloud credentials and an Epic where the user is the assignee or reporter.

Claude Prompt Booster chooses among three modes from the request: rewrite an existing prompt, design one from scratch, or audit it against a ten-criterion checklist. It applies structured prompt patterns, examples, model-specific guidance, and hallucination guardrails drawn from Anthropic materials. Boost and build modes return a copy-ready prompt with rationale, while audit mode prioritizes fixes without rewriting unless the user explicitly asks.

Beauty Diagram teaches an agent to use the bd CLI to polish existing Mermaid or PlantUML, render generated diagram source, and return share or direct-embed links for README files, Notion, and blogs. Node.js is required. Anonymous use is limited and watermarked, while unwatermarked exports, higher quotas, and prompt-to-diagram generation require an API key; AI generation is restricted to Pro or Premium plans.

Repo Docs System builds a Chinese knowledge base around a codebase, including an AGENTS.md context map, an architecture overview, design documents, product specifications, execution plans, and a documentation coverage score. It is intended for both human maintainers and AI agents that need a reliable project map. Repeated runs update and extend the existing documentation rather than deleting prior content.

Provide a finished DOCX, a reference-manager export, and one of eleven supported dental journals to receive a reformatted manuscript, a validation report, and an optional cover letter. The tool adjusts page setup, heading structure, citation rendering, and journal limits while preserving the manuscript text. It does not submit the paper, preserve tracked changes or comments, or insert the newly rendered bibliography back into the output DOCX in the current version.

Meeting Todos reads a chosen or recent meeting note, separates tasks assigned to you from other people's commitments, and surfaces unresolved decisions. It shows a preview before writing, preserves the format and order of the existing to-do file, and avoids duplicate entries. The plugin expects Claude Code, Markdown meeting notes, and a to-do file; on first load it also sends an anonymous install signal that can be disabled with an environment variable.

Read Deeper works with books stored in NotebookLM or files uploaded directly as PDF, EPUB, DOCX, or text. Its twelve modes cover core arguments, chapter distillation, 30-day actions, permanent notes, blind spots, debate, cross-book links, recall testing, and other focused outcomes instead of a generic summary. All output is in Thai with source markers. NotebookLM use requires a configured MCP server, while uploaded files need no additional integration.

Agent OS Patterns documents practical shapes for products that need persistent state, memory controls, subagents, scheduled or proactive behavior, persona, and process isolation. The seven patterns cover filesystem-backed state, behavioral flags in memory, sibling sessions, linked work items and run directories, three-gate proactive checks, named failures with partial output, example-driven personas, and a localhost sidecar between UI and runtime. It is an architecture reference, not a prompt-only or library-debugging tool.

Image Generation is a model-agnostic prompting guide for agents working with Imagen, Gemini Image, GPT Image, DALL·E, Stable Diffusion, Flux, Ideogram, Midjourney, and similar tools. It organizes prompts around subject, action, setting, composition, lighting, and style, with patterns for photography, illustration, logos, product shots, infographics, UI mockups, and image text. It also covers edit-preservation lists, failure diagnosis, and practical parameters such as aspect ratio, resolution, seed, and reference fidelity; it does not generate images by itself.

Prompt Injection Auditor reviews a system prompt for direct and indirect injection, role hijacking, instruction leakage, goal manipulation, memory poisoning, delimiter escapes, and multi-agent or tool-call attacks. It returns severity-ranked findings, concrete exploit examples, and a hardened replacement prompt without requiring external tools. Its scope is prompt-level security, so it does not detect application bugs, model-specific bypasses, steganographic attacks, probabilistic failures, or future zero-day jailbreaks.

Guide Claude Code through a short setup interview, then tailor a small set of skills, project rules, hooks, settings, and optional MCP integrations to the way you work. Its status command reports the installed setup and context budget, while the wizard explains what it skips. Recommendations are intentionally limited.

Make Claude Code prose less formulaic with six layers: persistent rules, a slash command, a Write/Edit hook, a reviewer subagent, and a scoring CLI. The framework covers 38 patterns and four domain profiles, so academic, documentation, blog, and commit writing can use different exceptions instead of one blunt filter.

Bridge findings from Cowork, Claude.ai, Claude Desktop, or Claude for Work into Claude Code through a shared markdown inbox. The /cowork skill classifies scope, asks for one confirmation, runs in-scope items sequentially, archives completed findings, and produces an aggregate report. Data-destructive work always pauses for explicit confirmation, and upstream inbox population remains your choice.

Use ultra-plan when a feature spans several modules or requires multiple AI sessions. It probes project rules, dispatches parallel read-only research, builds a module dependency graph, writes a global plan plus one document per module, and then asks how to execute. Separate Claude Code and Codex versions adapt the prompts and dispatch tools; routine single-file work is intentionally out of scope.