Mercury Agent Skills: A Curated SKILL.md Playbook Library for AI Agents

What Mercury Agent Skills Is and the Problem It Addresses

AI coding agents accept context through skills: structured Markdown files that give the agent standing instructions, frameworks, or playbooks for a particular domain. Writing these files from scratch for every workflow is tedious, and prompts written quickly tend to be generic. Mercury Agent Skills provides a starting library of 132 such files, hand-curated for use in real developer workflows.

The repository is maintained by cosmicstack-labs and targets developers who already use AI coding agents (Mercury, Claude Code, Codex CLI, Cursor, OpenClaw, Hermes, or Gemini CLI) and want to improve the quality of instructions they give those agents without writing every playbook themselves. The README frames the value proposition as curated rather than crowded: hand-written against real workflows, no duplicates.

The skills are available through a web browsing interface at skills.mercuryagent.sh, through the Mercury CLI for one-command install, or by cloning the repository and copying individual files. The format is intentionally minimal so any agent runtime that can read Markdown can use the files without custom tooling.

The SKILL.md Format: Structure and Portability

Every skill is a single file named SKILL.md with YAML frontmatter and a Markdown body. The frontmatter structure from the README:

---
name: skill-name
description: 'What this skill does and when to use it'
metadata:
  author: cosmicstack-labs
  version: 1.0.0
  category: development
  tags: [clean-code, refactoring, best-practices]
---

# Skill Name

Full instructions, frameworks, scoring rubrics, and actionable guidance.

The frontmatter provides a machine-readable identifier, version, category, and tags. The body is plain Markdown prose: instructions, decision frameworks, and specific guidance for the agent. The README notes the format is intentionally minimal so any agent runtime can parse it without custom tooling.

The directory structure mirrors the category taxonomy. Installing all skills into an agent means dropping the relevant SKILL.md files into the agent's designated skills directory. The README documents the install path for each supported agent: ~/.mercury/skills/ for Mercury, .claude/skills/ for Claude Code, .codex/skills/ for Codex CLI, .openclaw/skills/ for OpenClaw, .hermes/skills/ for Hermes, .cursor/skills/ for Cursor, and .gemini/skills/ for Gemini CLI.

Installing Skills via Mercury CLI or Direct File Copy

The Mercury CLI is the primary install tool. After installing Mercury Agent, search and install skills with:

mercury skills search "react patterns"

mercury skills install frontend/react-patterns

Multiple skills can be installed in one command:

mercury skills install ai-ml/prompt-engineering devops/docker security/threat-modeling

Installed skills land in ~/.mercury/skills/<category>/<slug>/SKILL.md and are picked up automatically by the Mercury agent. For other agents, clone the repository and copy files to the appropriate location:

git clone https://github.com/cosmicstack-labs/mercury-agent-skills.git

Then copy individual SKILL.md files from the categories/ directory into the target agent's skills folder. The README also points to skills.mercuryagent.sh, which provides per-skill install instructions for each supported agent, a full-text search index, and a trending leaderboard. A public JSON feed at /api/feed.json and per-skill JSON at /api/skills/<category>/<slug> are available for tooling builders.

23 Categories: Coverage and What Is Missing

The 132 skills divide across 23 categories. Development has 16 skills covering clean code practices, code review, debugging, testing, ADRs, documentation, refactoring, and dependency hygiene. AI and ML has 11 skills for prompt engineering, agent health checks, memory, delegation, handoffs, token budgets, error recovery, and audit logging. Automation has 11 skills including screenshot workflows, shell scripting, web scraping, data sync, and deployment automation.

Three categories stand out as domain-specific rather than engineering-general: Shop and Restaurant (10 skills covering inventory, menu engineering, table management, Zomato ordering, and Amazon assistant), Health and Wellness (5 skills on fitness, nutrition, mental health, sleep, and habits), and Career (5 skills on resume writing, interview prep, and LinkedIn). These are not useless, but they are unlikely to apply to most engineering teams.

Security has only 2 skills (audit and secure coding), and Data has 1. These are shallow given how significant those domains are in real engineering work. The library is strongest in the areas that overlap most with AI-assisted software development: Development, AI and ML, Backend, Frontend, DevOps, and Testing.

The README links to CATALOG.md for the full skill-by-skill index, and the web interface at skills.mercuryagent.sh provides the most complete browsable view of what is available.

Limitations: No Runtime and Discovery Without Mercury CLI

Mercury Agent Skills is a file library, not a runtime. The skills do nothing on their own; they are instructions for agents that can read them. If an agent's runtime does not support loading SKILL.md files from a specified directory, the files are inert. The supported agents are listed explicitly in the README; custom agent frameworks or proprietary systems are not supported unless they implement compatible skill loading.

Discovery outside of the Mercury CLI requires either browsing skills.mercuryagent.sh or navigating the CATALOG.md file. Without the Mercury CLI, installing skills means manually identifying the right category path, cloning the repository or fetching individual raw files, and placing them in the correct directory. This is workable but less convenient than a package manager.

The repository has no GitHub releases and lists no versioning strategy beyond the version field in each SKILL.md frontmatter. There is no lock file or dependency pinning for skills. A skill's content can change on any push, so teams that want stable skill versions need to copy files from a specific commit rather than always pulling from main.

The last push was on 2026-08-25, and the library is MIT licensed. CONTRIBUTING.md documents how to submit new skills via pull request.

Mercury Agent Skills vs. LangChain Hub

LangChain Hub is a registry of reusable prompts and chains for the LangChain framework, hosted at smith.langchain.com/hub. Both Mercury Agent Skills and LangChain Hub provide a shared library of reusable agent instructions, but they are built for different deployment models.

LangChain Hub skills are deeply integrated with LangChain's Python SDK: they are pulled at runtime using langchain-hub client code, use LangChain's prompt template syntax, and require LangChain to run. Mercury Agent Skills are plain Markdown files with no runtime dependency. Copying a SKILL.md file to .claude/skills/ makes it available to Claude Code without any SDK or additional tooling.

For teams already using LangChain, Hub skills can be pulled programmatically into agent chains with type-safe templates. For teams using standalone agents like Claude Code or Codex CLI, LangChain Hub is not useful because those agents do not load LangChain prompt templates. Mercury Agent Skills fills the gap for teams whose agent runtime is not LangChain. The two registries serve incompatible runtimes and are not direct substitutes.

Editorial conclusion

Mercury Agent Skills is useful for developers who use AI coding agents regularly and want a starting point for common workflows rather than writing skill files from scratch. It is not useful for teams building fully custom agent systems with proprietary context: the skills are generic enough to be reusable across projects, which means they will often need adjustment for a specific codebase. Before adopting skills from this library, check the categories against your actual workflow needs. The Development and AI-ML categories are the strongest; categories like Shop and Restaurant or Health and Wellness are less likely to match engineering workflows. The MIT licence means you can adapt them freely.

Frequently asked questions

Official sources

1. cosmicstack-labs/mercury-agent-skills on GitHub
2. Issues
3. License: MIT
4. Project website
5. README