agent-skills

Note

My personal and public agent skills. Olshansky.info

What is this?

• Olshansky's day-to-day agent skills
• Follows the Agent Skills pattern for cross-tool skill distribution
• Inspired by vercel-labs/agent-skills

Table of Contents

• Quickstart
• Available Skills
  • Polished Skills
  • Personal Skills (cmd-*)
• Star History
• Demo of skills-dashboard
• How It Works

Quickstart

npx skills add olshansk/agent-skills

Then ask your agent to run any installed skill:

• "resolve merge conflicts"
• "close the loop on this session"
• "idiot proof the documentation"
• "generate skills dashboard"

Tip

Start with session-commit — it turns every coding session into durable knowledge by extracting patterns, decisions, and gotchas into your AGENTS.md. Future sessions (and future agents) pick up right where you left off.

Available Skills

Polished Skills

Skills I've polished for public use.

Skill	What it does	Trigger examples
session-commit	Captures session learnings and updates AGENTS.md safely	"run session commit", "close the loop", "update AGENTS.md"
skills-dashboard	Scrapes skills.sh and generates an interactive HTML dashboard	"generate skills dashboard", "show skills ecosystem"

Personal Skills (cmd-*)

Skills I use daily but might rename or delete in the future. I prefix them with cmd- so I can easily also leverage them as custom slash commands in claude code.

Skill	Description
cmd-chain-halt-code-reviewer	Review protocol code for chain halt risks, non-determinism, and onchain behavior bugs
cmd-email-md	Convert markdown to email-safe HTML with inline styles and cross-client compatibility
cmd-gh-issue	Create structured GitHub issues from conversation context using gh CLI
cmd-idiot-proof-docs	Simplify documentation for clarity and scannability with approval-gated edits
cmd-local-repo-skills	Scaffold cross-tool repo-local skills with canonical source in .agents/skills/ and symlinks
cmd-olshanskify	Apply Olshansky's personal style to docs, code, blog posts, or presentations via templates
cmd-persona	Prime the agent with a behavioral persona for the conversation
cmd-pr-build-context	Build high-signal PR context with diff analysis, risk assessment, and discussion questions
cmd-pr-conflict-resolver	Resolve merge conflicts with context-aware 3-tier classification and escalation
cmd-pr-description	Generate concise PR descriptions by analyzing the diff against a base branch
cmd-pr-gh-comments	Holistically triage PR comments with line-range context, adjacent sweeps, approval-gated resolution, and cmd-olshanskify updates for @olshansk feedback
cmd-pr-review-prepare	Prepare branch for code review by building context and identifying issues
cmd-pr-edgecase	Review branch changes for test gaps, logic edge cases, and failure modes
cmd-pr-test-plan	Generate manual test plans with verified commands and pass/fail criteria
cmd-productionize	Transform apps into production-ready deployments with framework-specific optimization
cmd-proofread	Proofread posts for spelling, grammar, repetition, logic, weak arguments, and broken links
cmd-follow-up	Post-implementation reflection — surface missed work, simplifications, and idiomatic fixes
cmd-latest-msg	Store or retrieve the latest agent message to /tmp/agents/{agent}/
cmd-sculpt-code	Reshape code for readability, naming, structure, TODOs, and reduced surface area
cmd-rfc-review	Review RFCs for problem clarity, compliance, security, and performance using SCQA
cmd-rss-feed-generator	Generate Python RSS feed scrapers with hourly GitHub Actions integration
cmd-scope-sweep	Final pass to identify missed items, edge cases, and risks before closing scope
makefile	Create or improve Makefiles with templates (python-uv, fastapi, nodejs, go, flutter)
mermaid-render	Render and display Mermaid diagrams inline in iTerm2 or Ghostty

Star History

Demo of skills-dashboard

A live dashboard of the skills.sh ecosystem is available at skills-dashboard.olshansky.info.

It shows publisher distribution, install counts, top skills, and the long-tail power law of adoption. Regenerate it yourself with the skills-dashboard skill.

How It Works

Symlink Architecture

Two sources of truth feed into every tool's skills directory:

graph TB
    classDef workspace fill:#d4e6f1,stroke:#2980b9,stroke-width:2px,color:#000
    classDef dotdir fill:#fadbd8,stroke:#e74c3c,stroke-width:2px,color:#000

    subgraph sources["Source of Truth"]
        thirdparty["~/.agents/skills/*<br/>(third-party skills)"]
        repo["~/workspace/agent-skills/<br/>·  skills/*  ·  agents/*  ·<br/>(your skills)"]
    end

    subgraph claude["~/.claude/"]
        direction LR
        claude_skills["skills/*"]
        claude_md["CLAUDE.md"]
    end

    subgraph gemini["~/.gemini/"]
        direction LR
        gemini_skills["antigravity/skills/*"]
        gemini_md["GEMINI.md"]
    end

    subgraph codex["~/.codex/"]
        direction LR
        codex_skills["skills/*"]
        codex_md["AGENTS.md"]
    end

    repo -->|"symlink (wins on conflict)"| claude_skills
    repo -->|"symlink (wins on conflict)"| gemini_skills
    repo -->|"symlink (wins on conflict)"| codex_skills
    repo -->|"AGENTS.md"| claude_md
    repo -->|"AGENTS.md"| codex_md
    repo -->|"MEMORIES.md"| gemini_md

    thirdparty -->|"symlink"| claude_skills
    thirdparty -->|"symlink"| gemini_skills
    thirdparty -->|"symlink"| codex_skills

    class repo workspace
    class thirdparty,claude_skills,claude_md,gemini_skills,gemini_md,codex_skills,codex_md dotdir

Loading

Skill type	Source of truth	Installed via	Symlink target
Your skills	skills/ (this repo)	make link-skills	~/workspace/agent-skills/skills/*
Third-party	~/.agents/skills/	npx skills add	~/.agents/skills/*

Makefile Workflows

graph LR
    classDef default color:#000

    skills["agent-skills/skills/*"]
    agents["agent-skills/agents/"]

    subgraph outbound["Symlink OUT"]
        link["make link-skills"]
    end

    subgraph inbound["Copy IN"]
        sync["make sync"]
    end

    claude_dir["~/.claude/"]
    gemini_dir["~/.gemini/"]
    codex_dir["~/.codex/"]
    snapshots["~/workspace/configs/"]

    skills -->|"skills symlink"| link
    agents -->|"instructions symlink"| link
    link --> claude_dir
    link --> gemini_dir
    link --> codex_dir

    claude_dir -->|"file copy"| sync --> snapshots
    gemini_dir -->|"file copy"| sync
    codex_dir -->|"file copy"| sync

Loading

Target	Description
make link-skills	Symlink repo + third-party skills into Claude, Gemini, and Codex
make list-skills	List all skills with descriptions
make sync	Backup tool configs into ~/workspace/configs/
make test	Validate skill frontmatter and repo consistency

After npx skills add

npx skills add installs third-party skills into ~/.agents/skills/ and creates symlinks in ~/.claude/skills/. Running make link-skills afterward restores your repo skills (which take precedence on name conflicts) and extends third-party skills to Codex and Gemini.