skill
skill-tap
Portable, reusable skills for agents and everyday workflows. Clank responsibly.
Set up
Skills are installed by placing their folder where your agent loads skills.
The download comes from github.com/batteryshark/skill-tap itself, not from our server, so what you get is exactly what its author published.
Install instructions come from the README
README
synced Oct 28, 2026Skill Tap
Skill Tap is a public collection of portable, reusable skills for agents. It is a general-purpose catalog of repeatable workflows, organized by category as it grows. Each skill is self-contained, readable without a specific agent platform, and backed by small command-line tools where deterministic checks help.
Skills
Development
| Skill | Use it to |
|---|---|
architecture-for-comprehension | Review whether a system is coherent, navigable, and explainable as a whole. |
code-quality-review | Audit or improve local code quality with evidence and behavior-preserving refactoring. |
codebase-archeology | Map inherited or archived projects and decide what to preserve, merge, or retire. |
codebase-writeup | Produce an evidence-backed technical explanation of how a codebase works. |
deliverable-hygiene | Keep a repository shaped like finished work and recover one that has accumulated process residue. |
agent-create | Scaffold a portable Markdown role prompt with explicit inputs, constraints, and output contract. |
excalidraw | Create and validate editable Excalidraw diagram JSON before rendering and delivery. |
engineering-diagrams | Create evidence-backed system, component, data-flow, sequence, state, and trust-boundary diagrams. |
github-publish | Preflight and publish a local project to GitHub without leaking or staging unrelated state. |
publish-ready | Run a final, evidence-backed cleanup before sharing or releasing a repository. |
mcp-elicitations | Build and test protocol elicitations and FastMCP Apps interaction boundaries. |
mcp-server-create | Scaffold a library-first FastMCP 3 server with portable configuration and tests. |
mcp-server-design | Design providers, transforms, auth, tasks, outputs, discovery, and safety boundaries. |
mcp-server-upgrade | Upgrade an MCP server while preserving and verifying observable compatibility. |
mcp-testing | Test domain logic, MCP components, interactions, security, lifecycle, and transports. |
norman | Apply interface-design principles while building: signifiers, interaction cost, and controls that earn their place. |
norman-audit | Audit an interface from source for eight failure classes, ranked worst first with fixes. |
skill-create | Scaffold a complete portable skill package that satisfies this tap's contract. |
skill-node-portability | Audit Node.js skill dependencies, lockfiles, launchers, and compiled artifacts. |
skill-python-portability | Audit Python skill imports, PEP 723 metadata, dependencies, and launchers. |
skill-refine | Audit and refine a skill package against the public portability contract. |
windows-mingw-build | Build and troubleshoot portable 32-bit or 64-bit Windows C/C++ artifacts with MinGW-w64. |
work-product-audit | Find public prose that leaks internal reasoning, defensive framing, or generated-sounding filler. |
orchestrate | Route multi-part work across model tiers and reasoning budgets, then verify and repair what comes back. |
tier-plan | Turn one request into a routed work graph with a tier, a budget, and a check per unit. |
model-fitness | Probe a model for the properties delegation depends on and give it a tier ceiling. |
Productivity
| Skill | Use it to |
|---|---|
decision-records | Preserve consequential choices, constraints, alternatives, consequences, and reconsideration criteria. |
project-retrospective | Extract concrete lessons and improvements from completed or paused work. |
project-tracker | Maintain compact durable project memory across sessions and handoffs. |
claude-session-handoff | Build a redacted, evidence-backed continuation brief from a Claude Code JSONL transcript. |
apple-reminders | Manage Apple Reminders through a native EventKit command with mutation safeguards. |
reverse-brief | Reconstruct an evidence-labeled implementation brief from existing artifacts. |
goal-epic-craft | Turn rough intent into a bounded goal or epic, grade it against a standard, or split it into vertical slices. |
Utilities
| Skill | Use it to |
|---|---|
intellidiff | Compare files or folders, calculate SHA-256 hashes, and find exact duplicates. |
docker-cleanup | Survey and selectively reclaim Docker disk usage without broad implicit deletion. |
external-ip | Cross-check public IPv4 or IPv6 observations through multiple HTTPS services. |
nano-banana-image | Generate or edit images with Gemini using portable environment credentials. |
system-recon | Collect a bounded read-only system inventory without harvesting secrets. |
windows-ai-controls | Audit and plan reversible Windows AI policy changes for a specific build. |
windows-elevation | Plan the narrowest Windows elevation path with a guarded SYSTEM fallback. |
Use a skill
Copy a complete skill directory into your agent's skill path, or point the agent directly at its SKILL.md. Keep the directory intact: scripts, references, and role prompts are part of the skill.
git clone https://github.com/batteryshark/skill-tap.git
cp -R skill-tap/skills/development/publish-ready /path/to/your/agent/skills/
Every Python tool can run with uv or the system Python. The executable in each skill's bin/ directory prefers uv and falls back to python3 when the tool has no third-party dependencies.
skills/development/publish-ready/bin/publish-ready .
Add a skill
Put new skills under a broad category in skills/, optionally nest related skills in organizational groups, follow the skill contract, and run:
python3 scripts/validate_skills.py
python3 -m unittest discover -s tests
License
MIT. See LICENSE.