agent-skills
59 agent skills and 9 autonomous agents. Polyglot dev, web3, ZK, genomics, UI/UX, systems programming. Default host is the Raxol agent; Claude Code also works.
How skill loading works
Skills are portable SKILL.md files -- the agentskills.io format: YAML frontmatter (a name, a short trigger description) plus a markdown body. Any compatible host loads the same files. Two are documented here.
Raxol agent (default). Raxol.Agent.Skills.Store scans ~/.agents/skills/**/SKILL.md on boot and holds them as read-only procedural memory. Skills are pull-based: the agent sees only each skill's name and description (via the skills_list tool) and reads a full body on demand (skill_view) when it judges one relevant -- the trigger clause in the description guides that choice. Nothing but the list metadata sits in context until the agent pulls a skill, so all 59 cost almost nothing until used.
Claude Code (secondary). Lazy in a different way: Claude Code reads the trigger clause from each frontmatter and auto-injects the full skill content when the trigger matches your conversation. Sub-files (examples, checklists, reference tables) stay out of context until they're needed.
Either way, 59 skills don't bloat your sessions. The noir ZK skill isn't eating tokens while you're reviewing a PR.
Skills
Each skill lives in skills/<name>/ with a SKILL.md entry point.
Domain
| Skill | What it does |
|---|---|
claude-api |
Anthropic SDK reference (Python, TS, Go, Elixir, Rust, Lua, cURL) |
droo-stack |
Polyglot patterns (Elixir, TS, Go, Rust, C, Zig, Python, Lua, Shell, Noir, Chezmoi) |
raxol |
Elixir TUI/agent framework (TEA agents, MCP, headless sessions) |
noir |
ZK circuit design, Aztec contracts, constraint optimization |
solidity-auditor |
Solidity dev standards, vulnerability taxonomy, Foundry-first audit |
zk-x-ray |
Pre-audit briefing for ZK + EVM hybrids (Noir + Solidity), public-input parity check |
ethskills |
Ethereum tooling, framework selection, EIP/ERC standards |
nix |
Nix language, flakes, NixOS, Home Manager, packaging |
native-code |
NIF development (C/Rust/Rustler), SIMD (Zig), BEAM native boundary |
ffmpeg-asm |
ffmpeg upstream + integration: x86inc.asm, NEON + Apple M4 SME, HBD templating, dispatch, checkasm, upstream patch workflow |
coingecko |
CoinGecko/GeckoTerminal API: prices, markets, DEX pools, trending tokens |
blockscout |
Blockscout MCP: 16 tools for on-chain data across 8+ chains |
signoz |
SigNoz MCP: ~41 tools for OTel metrics/traces/logs, alerts, and regen correlation |
web-asset-generator |
Favicon, app icon, OG image, devicon generation and optimization |
cancer-predisposition-variant-analyst |
Ultra-rare variant interpretation, mechanistic paradox resolution, ACMG/ClinGen evidence |
Workflow
| Skill | What it does |
|---|---|
tdd |
Test-driven development: vertical slices, mutation testing, polyglot |
code-review |
PR review: blast radius, security scan, SOLID checks, 40-item checklist |
prd-to-plan |
PRD -> phased tracer-bullet vertical slices (and optionally GitHub issues) |
focused-fix |
5-phase bug fix: SCOPE -> TRACE -> DIAGNOSE -> FIX -> VERIFY |
release |
Conventional commits, semver bumping, changelog, readiness checks |
qa |
Bug triage and issue creation; interactive QA with background explorer |
interface-designer |
"Design It Twice" -- parallel sub-agents with divergent constraints |
ubiquitous-language |
DDD glossary extraction, canonical terms |
design-ux |
UI/UX design patterns, design tokens, accessibility, TUI aesthetics |
property-testing |
Generative/property-based testing (Hypothesis, proptest, StreamData, fast-check) |
refactoring-strategy |
Strangler fig, large renames, safe restructuring for polyglot codebases |
Infrastructure
| Skill | What it does |
|---|---|
mcp-server-builder |
OpenAPI -> MCP server scaffolding (Python FastMCP + TypeScript) |
ci-cd-pipeline-builder |
Stack detection -> GitHub Actions/GitLab CI generation |
dependency-auditor |
Multi-language vuln scanning + license compliance |
observability-designer |
SLO/SLI design, burn rate alerting, dashboard generation |
database-designer |
Schema analysis, ERD generation, index optimization |
performance-profiler |
Polyglot profiling (Node/Python/Go/Elixir/Rust) |
git-guardrails |
PreToolUse hooks to block dangerous git operations |
git-worktree-manager |
Parallel dev with deterministic port allocation |
env-secrets-manager |
Leak detection, rotation, pre-commit setup |
tech-debt-tracker |
Debt scanning, cost-of-delay prioritization |
security-auditor |
Security vulnerability scanning and compliance assessment |
playwright |
Browser automation and e2e testing tooling (Python + TS) |
prepper |
Pre-session project briefings (git, GitHub, CI, deps, recall) |
Meta
| Skill | What it does |
|---|---|
polymath |
Split-brain research: three-tier roster, polymath persona composition |
architect |
ADR workflows, dependency classification, pattern detection |
agent-designer |
Multi-agent architecture patterns, tool schemas, guardrails |
adversarial-reviewer |
Three-persona devil's advocate review |
self-improving-agent |
Auto-memory curation, pattern promotion lifecycle |
codebase-onboarding |
Auto-generate onboarding docs, audience-aware |
rag-architect |
RAG pipeline design: chunking, embedding, retrieval, evaluation |
llm-cost-optimizer |
7 optimization techniques in priority order |
digest |
Multi-platform activity digest (18 sources + differential mode) |
recall |
Knowledge base: query past decisions, patterns, gotchas |
autoresearch |
Check experiment status, run iterations, view dashboards |
watchdog |
Scan repos for stale PRs, failing CI, security advisories |
sentinel |
Monitor on-chain contracts for anomalous transactions |
regen |
Read Fluidify Regen incidents, extract SigNoz correlation keys |
patchbot |
Scan and update outdated dependencies across ecosystems |
writing-voice |
Writing voice calibration from studied authors, combinatorial blending |
humanizer |
Strip AI-isms and add real voice; de-slop AI-generated prose |
skill-creator |
Scaffold new skills with frontmatter, triggers, and sub-files |
Agents
Nine standalone tools, each with a Typer CLI, pydantic models, and a FastMCP server. They run independently of the skill system -- install them separately, talk to them over MCP.
| Agent | What it does | MCP tools |
|---|---|---|
digest |
Multi-platform activity digest (18 sources, differential, structured views) | 7 |
recall |
Knowledge capture and retrieval (SQLite + FTS5) | 8 |
scribe |
Session insight extractor (writes to recall) | 3 |
autoresearch |
Autonomous experiment runner (ML, Noir, Solidity) | 3 |
watchdog |
Repo health monitor (PRs, CI, deps, advisories) | 2 |
prepper |
Pre-session context builder (git, GitHub, deps, recall, sentinel) | 3 |
sentinel |
On-chain contract monitor via Blockscout (11 chains) | 2 |
patchbot |
Polyglot dependency updater (Elixir, Rust, Node, Go, Python) | 3 |
regen |
Incident reader + SigNoz OTel correlation (Fluidify Regen) | 7 |
MCP integration
Each agent doubles as an MCP server. Add to ~/.mcp.json:
{
"mcpServers": {
"digest": { "command": "digest", "args": ["serve"] },
"recall": { "command": "recall", "args": ["serve"] },
"scribe": { "command": "scribe", "args": ["serve"] },
"autoresearch": { "command": "autoresearch", "args": ["serve"] },
"watchdog": { "command": "watchdog", "args": ["serve"] },
"prepper": { "command": "prepper", "args": ["serve"] },
"sentinel": { "command": "sentinel", "args": ["serve"] },
"patchbot": { "command": "patchbot", "args": ["serve"] },
"regen": { "command": "regen", "args": ["serve"] },
"coingecko": { "url": "https://mcp.api.coingecko.com/mcp" }
}
}
See TODO.md for the roadmap.
Install
Skills live at ~/.agents/skills/<name>/SKILL.md. Get them on disk once (below), then point your host at that directory -- the Raxol agent reads it natively.
Raxol agent (default)
~/.agents/skills is already the Raxol agent's default external skills dir. Enable the provider:
# config/config.exs (or runtime.exs)
config :raxol_agent, skills_provider: Raxol.Agent.Skills.Store
That supervises Raxol.Agent.Skills.Store (which scans ~/.agents/skills/**/SKILL.md on boot) and exposes the skills_list, skill_view, and skill_manage tools to the agent. Verify:
length(Raxol.Agent.Skills.Store.list()) # => 59
To load skills straight from a repo checkout instead, add its path:
config :raxol_agent, skills_external_dirs: ["~/.agents/skills", "~/CODE/agent-skills/skills"]
Get the skills on disk
npx -- pick individual skills or grab everything:
npx skills@latest add DROOdotFOO/agent-skills/tdd
npx skills@latest add DROOdotFOO/agent-skills
chezmoi (recommended -- this is how the skills are kept fresh) -- add the external, then a single chezmoi apply pulls the whole collection to ~/.agents/skills:
# ~/.chezmoiexternal.toml
[".agents/skills"]
type = "archive"
url = "https://github.com/DROOdotFOO/agent-skills/archive/main.tar.gz"
stripComponents = 2
include = ["*/skills/**"]
refreshPeriod = "168h"
Apply to install, and force a refresh whenever you want the latest before the 168h window elapses:
chezmoi apply # install / update (auto-refreshes every 168h)
chezmoi apply --refresh-externals # force-pull the newest skills right now
The Raxol agent reads ~/.agents/skills directly after apply -- no further step. To add a new skill: push it to agent-skills main, then chezmoi apply --refresh-externals. For the full reproducible rig (this external plus the repo clone for the MCP agent CLIs and the ~/.claude/skills symlinks), see DROOdotFOO/dotfiles -- chezmoi apply --refresh-externals is the one command that applies or refreshes everything.
Manual -- clone and symlink:
git clone https://github.com/DROOdotFOO/agent-skills.git ~/.agents/skills-repo
ln -s ~/.agents/skills-repo/skills ~/.agents/skills
Claude Code (secondary)
Self-contained plugin (no ~/.agents/skills needed):
/plugin install agent-skills@DROOdotFOO/agent-skills
# or add the marketplace first:
/plugin marketplace add DROOdotFOO/agent-skills
Already installed to ~/.agents/skills via the methods above? Symlink them into Claude Code:
mkdir -p ~/.claude/skills
for d in ~/.agents/skills/*/; do
ln -sf "../../.agents/skills/$(basename "$d")" ~/.claude/skills/
done
Agents
Each agent installs independently:
cd agents/<name> && pip install -e .
Lint
./scripts/skills-lint.sh
Checks frontmatter, trigger clauses, file references, and cross-skill links.
License
MIT
No comments yet
Be the first to share your take.