Agent Skills

A personal collection of reusable Agent Skills for coding, research, analysis, decision support, and tool-assisted workflows. Each skill packages focused instructions and optional scripts, references, assets, tests, or client metadata behind a discoverable SKILL.md file.

The repository favors portable skills that can be used by Codex, Claude Code, OpenCode, and other clients that support the Agent Skills format. Client-specific integration is kept optional where practical.

Skills

Skill	Purpose
agent-steering	Authors precise behavioral constraints that prevent recurring agent failure modes.
aws-expert	Produces sourced AWS technical answers and Slack-ready output.
critical-thinking-discussion	Guides issue-led Socratic discussion while maintaining a live reasoning pyramid.
ddg-search	Performs current web research through DuckDuckGo and synthesized search services.
deep-resolve	Applies exhaustive root-cause analysis to difficult, high-cost decisions.
fin-analyzer	Analyzes financial transaction CSVs and generates interactive reports.
liquid-glass	Builds Apple-style refractive Liquid Glass interfaces in HTML, CSS, and SVG.
macos-calendar-plus	Creates and manages macOS Calendar events through AppleScript.
mcp-skill-authoring	Creates portable, token-efficient MCP-backed skills using the official MCP Inspector CLI.
opencode-agents	Delegates work to specialized oh-my-openagent agents through OpenCode.
smart-shopper	Researches and compares products while caching results for fast re-filtering.

Read each skill's frontmatter before use. Its description defines when the skill should trigger, important exclusions, and any required dependencies.

Installation

Install individual skills with the Skills CLI:

npx skills install axot/agent-skills --skill <skill-name>

For example:

npx skills install axot/agent-skills --skill mcp-skill-authoring

Usage

Most compatible clients can select a skill automatically from its description. Clients may also support explicit invocation using a skill name, such as $mcp-skill-authoring or /critical-thinking-discussion; exact syntax varies by client.

Supporting files are relative to the skill directory. When a SKILL.md directs the agent to read a reference or run a script, preserve that directory structure when installing the skill.

Authoring Guidelines

Every skill must have a directory whose name matches the name field in its SKILL.md frontmatter. The frontmatter must also include a specific description that states positive triggers and material exclusions.

Keep workflows scoped and executable. Prefer deterministic scripts for parsing, calculation, or repetitive protocol work. Keep credentials out of instructions, examples, logs, and committed configuration. Verify mutable facts against current primary documentation, and validate observable behavior before claiming completion.

Use mcp-skill-authoring when creating a portable skill that invokes an MCP server without permanently loading its complete tool catalog into model context.

Validation

Before committing a skill:

1. Parse the YAML frontmatter and confirm the directory name matches name.
2. Validate referenced files, scripts, and client metadata.
3. Run syntax checks and skill-specific tests.
4. Scan for credentials and generated output that should not be committed.
5. Run git diff --check and inspect the complete staged diff.

The MCP authoring skill includes an offline smoke test:

bash mcp-skill-authoring/tests/smoke.sh

Project-wide operating rules are defined in AGENTS.md.