agent-skills

A collection of Agent Skills; structured markdown that extends LLM agent capabilities through a standardised format.

Available Skills

Refer to § Token stats for a detailed breakdown of each skill's token count, plus overall metadata usage. I've used and tested them most with Crush, Amp, and Shelley, and to a lesser degree, Octofriend and Pi.

addressing-agent-comments: Finds AGENT: comments in code, gathers feedback, and carries out requested changes.
ast-grep: Write ast-grep rules for structural code search and analysis. Find patterns the way the compiler sees them, not just as text.
authoring-skills: Create and review Agent Skills following best practices. Covers skill structure, frontmatter, and progressive disclosure patterns.
creating-tasks-through-lunatask: Create tasks and handoffs in Lunatask via lune. Tasks are just tasks. Handoffs capture work to resume later across sessions without filling context windows.
formatting-commits: Create commits following Conventional Commits format via git-format.
frontend-accessibility: Generate accessible HTML, React, and frontend code following WCAG 2.2 AA. Prioritizes semantic HTML over ARIA, keyboard navigation, and screen reader compatibility.
handling-customer-data: Query customer data responsibly—the agent answers questions about the data without ever seeing it directly.
humanizer: Remove AI-generated patterns from text like promotional fluff, weasel words, and mechanical sentence structures. Based on Wikipedia's AI Cleanup research. Originally from blader/humanizer.
invoking-subagents: Spawn subagents with restricted tool access for parallel tasks across repositories. Requires synu and the claude CLI. Useful for summarizing git history or processing large diffs without filling the main context window.
managing-and-navigating-worktrees: Git worktree management using wt with a bare repository structure. Each branch lives in its own sibling directory. Requires wt, git, and gum.
querying-documentation: Query Go documentation with focused subagents for complex questions spanning multiple packages or external libraries. Requires Fish shell and synu. Currently Go-only.
rebasing-with-git: Git rebase workflows from simple rebases to the drop-and-repick integration branch pattern. Assumes upstream remote for forks and rerere enabled. Supports worktrees including wt-managed repos.
researching-with-rumilo: Dispatch AI research subagents via rumilo for web search and repository exploration. Research topics, look up library usage, or explore external codebases without filling the main context window.
resuming-work-through-lunatask: Resume deferred work from Lunatask handoff notes via lune.
reviewing-code: Review code for correctness, security, reliability, performance, and quality. Covers PRs, diffs, and general change review.
scripting-with-go: Create executable Go scripts using a shell trick (not a true shebang). For automation and tooling outside of Go projects.
working-with-tmux: Spawn and manage background processes via tmux. Create windows, send commands, and capture output without blocking the main session. Useful for servers and long tasks.
writing-git-tags: Generate git tag annotations from commit history following Semantic Versioning and Conventional Commits via git-format.
writing-roc-lang: Write Roc code with strong static types, helpful compiler errors, and functional programming. Covers both full applications and one-off scripts.
writing-rust: Write idiomatic Rust with strict quality gates: cargo fmt, clippy -- -D warnings, and required tests. Enforces thiserror for libs, anyhow for apps, warns against common anti-patterns.

Installation

If you're using Amp, run amp skill add https://git.secluded.site/agent-skills. Otherwise, clone the repository, then install via make or manual symlinks.

git clone https://git.secluded.site/agent-skills
cd agent-skills

With Make

Skills install to ~/.config/agents/skills/ by default. Override by setting SKILL_DIR. Tab completion works so you don't have to remember skill names or copy/paste.

# List available skills
make list

# Install all skills to default dir
make all

# Install specific skills to Claude Code
SKILL_DIR=$HOME/.claude/skills make authoring-skills formatting-commits

# Remove a specific skill from Amp
make uninstall-authoring-skills SKILL_DIR=$HOME/.config/agents/skills

# Remove all skills from Codex
make uninstall SKILL_DIR=$HOME/.codex/skills

With symlinks

# Install all skills
for s in skills/*/; do ln -s "$PWD/$s" ~/.config/agents/skills/; done

# Install specific skills
ln -s "$PWD/skills/authoring-skills" ~/.config/agents/skills/

# Remove a skill
rm ~/.config/agents/skills/formatting-commits

Agent skill directories

Agent	Dir
Crush, Octofriend, Shelley, Amp	`~/.config/agents/skills/`
Pi	`~/.pi/agent/skills/`
Anthropic's Claude Code	`~/.claude/skills/`
OpenAI's Codex	`~/.codex/skills/`

Updating

cd ~/path/to/agent-skills
git pull

Symlinked skills update automatically.

Contributions

Patch requests are in amolith/llm-projects on pr.pico.sh. You don't need a new account to contribute, you don't need to fork this repo, you don't need to fiddle with git send-email, you don't need to faff with your email client to get git request-pull working...

You just need:

Git
SSH
An SSH key

# Clone this repo, make your changes, and commit them
# Create a new patch request with
git format-patch origin/main --stdout | ssh pr.pico.sh pr create amolith/llm-projects
# After potential feedback, submit a revision to an existing patch request with
git format-patch origin/main --stdout | ssh pr.pico.sh pr add {prID}
# List patch requests
ssh pr.pico.sh pr ls amolith/llm-projects

See "How do Patch Requests work?" on pr.pico.sh's home page for a more complete example workflow.

Token stats

Real token stats as reported by Synthetic.new's /messages/count_tokens endpoint with ./skill-stats.go.

=== addressing-agent-comments ===

Token breakdown:
  Name:           22 tokens
  Description:    96 tokens
  Body:          706 tokens (40 lines)
  ───────────────────────────────────────────────
  Total:         824 tokens

=== ast-grep ===

Token breakdown:
  Name:           18 tokens
  Description:   198 tokens
  Body:         7503 tokens (469 lines)
  References:
    rule_reference.md                         6814 tokens
  ───────────────────────────────────────────────
  Total:       14533 tokens
  ⚠️  Body exceeds recommended 5000 token budget!

=== authoring-skills ===

Token breakdown:
  Name:           20 tokens
  Description:    86 tokens
  Body:         2000 tokens (102 lines)
  References:
    checklist.md                               806 tokens
    patterns.md                               2415 tokens
    process.md                                 833 tokens
    specification.md                          3890 tokens
  ───────────────────────────────────────────────
  Total:       10050 tokens

=== creating-tasks-through-lunatask ===

Token breakdown:
  Name:           30 tokens
  Description:   108 tokens
  Body:         1927 tokens (88 lines)
  References:
    cli.md                                     589 tokens
    handoff.md                                2769 tokens
    installing-lune.md                         176 tokens
  ───────────────────────────────────────────────
  Total:        5599 tokens

=== formatting-commits ===

Token breakdown:
  Name:           22 tokens
  Description:    98 tokens
  Body:          992 tokens (43 lines)
  References:
    installing-git-format.md                    64 tokens
  ───────────────────────────────────────────────
  Total:        1176 tokens

=== frontend-accessibility ===

Token breakdown:
  Name:           20 tokens
  Description:   140 tokens
  Body:         2891 tokens (148 lines)
  References:
    antipatterns.md                           3638 tokens
    patterns.md                               6233 tokens
  ───────────────────────────────────────────────
  Total:       12922 tokens

=== handling-customer-data ===

Token breakdown:
  Name:           22 tokens
  Description:   123 tokens
  Body:         1983 tokens (107 lines)
  ───────────────────────────────────────────────
  Total:        2128 tokens

=== humanizer ===

Token breakdown:
  Name:           15 tokens
  Description:    84 tokens
  Body:         8536 tokens (462 lines)
  ───────────────────────────────────────────────
  Total:        8635 tokens
  ⚠️  Body exceeds recommended 5000 token budget!

=== invoking-subagents ===

Token breakdown:
  Name:           20 tokens
  Description:   113 tokens
  Body:          796 tokens (13 lines)
  References:
    installing-synu.md                         140 tokens
  ───────────────────────────────────────────────
  Total:        1069 tokens

=== managing-and-navigating-worktrees ===

Token breakdown:
  Name:           32 tokens
  Description:   164 tokens
  Body:         1912 tokens (96 lines)
  ───────────────────────────────────────────────
  Total:        2108 tokens

=== querying-documentation ===

Token breakdown:
  Name:           22 tokens
  Description:   132 tokens
  Body:          796 tokens (22 lines)
  References:
    dependencies.md                            108 tokens
  ───────────────────────────────────────────────
  Total:        1058 tokens

=== rebasing-with-git ===

Token breakdown:
  Name:           22 tokens
  Description:   188 tokens
  Body:         5696 tokens (202 lines)
  ───────────────────────────────────────────────
  Total:        5906 tokens
  ⚠️  Body exceeds recommended 5000 token budget!

=== researching-with-rumilo ===

Token breakdown:
  Name:           25 tokens
  Description:   159 tokens
  Body:         1319 tokens (49 lines)
  ───────────────────────────────────────────────
  Total:        1503 tokens

=== resuming-work-through-lunatask ===

Token breakdown:
  Name:           30 tokens
  Description:   113 tokens
  Body:         1241 tokens (42 lines)
  References:
    cli.md                                     528 tokens
    installing-lune.md                         176 tokens
  ───────────────────────────────────────────────
  Total:        2088 tokens

=== reviewing-code ===

Token breakdown:
  Name:           18 tokens
  Description:   125 tokens
  Body:         1880 tokens (121 lines)
  ───────────────────────────────────────────────
  Total:        2023 tokens

=== scripting-with-go ===

Token breakdown:
  Name:           20 tokens
  Description:   120 tokens
  Body:         1968 tokens (138 lines)
  ───────────────────────────────────────────────
  Total:        2108 tokens

=== working-with-tmux ===

Token breakdown:
  Name:           22 tokens
  Description:    84 tokens
  Body:         1465 tokens (87 lines)
  ───────────────────────────────────────────────
  Total:        1571 tokens

=== writing-git-tags ===

Token breakdown:
  Name:           22 tokens
  Description:   101 tokens
  Body:         2232 tokens (107 lines)
  References:
    installing-git-format.md                    64 tokens
  ───────────────────────────────────────────────
  Total:        2419 tokens

=== writing-roc-lang ===

Token breakdown:
  Name:           22 tokens
  Description:   142 tokens
  Body:         1846 tokens (117 lines)
  References:
    builtins-llms.md                         67881 tokens
    llms.md                                  11324 tokens
    scripting.md                              1722 tokens
  ───────────────────────────────────────────────
  Total:       82937 tokens

=== writing-rust ===

Token breakdown:
  Name:           18 tokens
  Description:   120 tokens
  Body:         3628 tokens (139 lines)
  References:
    idioms.md                                 4605 tokens
  ───────────────────────────────────────────────
  Total:        8371 tokens

============================================================
SUMMARY
============================================================

Skills: 20
Metadata: 2936 tokens
Combined bodies: 51317 tokens
Overall: 169028 tokens
Validation errors: 0

Largest skills (by total tokens):
  1. writing-roc-lang                         82937 tokens
  2. ast-grep                                 14533 tokens
  3. frontend-accessibility                   12922 tokens
  4. authoring-skills                         10050 tokens
  5. humanizer                                 8635 tokens

Some other tools if these interested you

wt - CLI for managing git worktrees
git-format - CLI that turns LLM input into well-formatted conventional commits and tags
garble - transform stdin with an LLM (fix typos, translate, reformat)
lune - CLI and MCP server for Lunatask.app

README.md