A collection of skills for AI agents.
This repository contains reusable skills that enhance AI agent capabilities. Skills are modular, composable units that provide specialized knowledge and workflows for agents.
The open-source skills CLI runs directly through npx, so it does not require a separate global installation.
Browse the skills in this repository:
npx skills add bastos/skills --listInstall interactively and choose a skill and detected agent:
npx skills add bastos/skillsTo install a specific skill, pass its name from the table below. For example:
npx skills add bastos/skills --skill unhobble-agent-contextInstallations are project-scoped by default. Add --global to make the skill available across projects and --agent to target a particular agent:
npx skills add bastos/skills --skill unhobble-agent-context --global --agent codexUse --yes for a non-interactive installation. Verify global skills for an agent with:
npx skills list --global --agent codexReview a skill's SKILL.md, scripts, and references before installing it because skills run with the agent's permissions. See the CLI documentation for supported sources, agents, and management commands.
| Skill | Description |
|---|---|
| apple-cktool | Automate CloudKit schema and test-data workflows with Apple's Xcode-bundled cktool CLI on macOS, including safe authentication, validation, reset, query, and deletion procedures. |
| apple-cktool-js | Build typed CloudKit automation for Node.js, browsers, and CI with Apple's CKTool JS packages, Promise APIs, secure token injection, and guarded schema and record operations. |
| audit-training-experiment | Audit ML experiments for reproducibility, evidence integrity, leakage, checkpoint identity, evaluation validity, and overclaimed conclusions. |
| awk | Expert GNU awk (gawk) one-liner generation and transformation for text field processing, column extraction, aggregation, CSV/TSV reformatting, and log analysis. |
| close-out-ml-experiment | Preserve experiment evidence, reconcile report discrepancies, record the canonical decision and limitations, and prepare a concise handoff. |
| compare-models-blindly | Create seeded blinded model-comparison packets and evaluate candidate outputs with reversed order to control identity and presentation bias. |
| conventional-commits | Format commit messages using the Conventional Commits specification for automated tooling, changelog generation, and semantic versioning. |
| design-resumable-model-evaluation | Design strict model benchmarks that persist per-case evidence, resume without repeating work, and use mathematically valid early stopping. |
| ios-design-review | Review iOS app UI/UX design quality using Apple's design principles and the Liquid Glass-era design language from WWDC25/26. |
| mikrotik-routeros | Configure and automate MikroTik RouterOS routers through the terminal, CLI, .rsc imports, scripts, Scheduler, Netwatch, firewall, NAT, DHCP, bridge, VLAN, routing, and backups. |
| obsidian-daily | Manage Obsidian Daily Notes via the official Obsidian CLI. Create, open, and append entries to daily notes, read past notes by date, and search vault content. |
| orbstack-cli | Drive OrbStack from the terminal on macOS: the orb/orbctl CLI for Linux VMs, Docker engine usage, .orb.local domains and networking, SSH, Kubernetes, config keys, and headless/CI automation. |
| package-private-ml-artifacts | Inventory, hash, scan, and verify ML artifacts before an explicitly approved private upload. |
| prepare-training-corpus | Build deterministic train, validation, and test corpora with provenance, group-safe splits, manifests, hashes, and leakage checks. |
| proxmox-admin | Administer Proxmox VE hosts, create and manage KVM virtual machines with qm, manage LXC containers with pct, and configure storage, networking, and clusters via the CLI. |
| review-ios-ui-apple-hig | Review iOS app UI against Apple Human Interface Guidelines and iOS platform conventions using page-specific Apple reference notes. |
| review-training-data-quality | Audit training corpora for distribution collapse, ambiguity, context sufficiency, hard negatives, abstention behavior, leakage, and label defensibility. |
| run-hugging-face-training-smoke | Run a bounded private Hugging Face training smoke with immutable inputs, preflight checks, stop gates, checkpoint verification, and reproducible evidence. |
| sed | Expert GNU sed one-liner generation for stream editing, in-place file editing, bulk find/replace, log filtering, and text transformation pipelines. |
| select-and-verify-best-checkpoint | Verify evaluation cadence, best-checkpoint selection, and held-out inference checkpoint identity before trusting test metrics. |
| send-to-minion | Delegate a prompt to exactly one helper agent, supervise it to completion, verify its result, and return the parent agent's final answer. |
| swift-review | Review Swift/Xcode codebases, pull requests, local changes, or individual files against Swift best practices for style, API design, memory management, performance, and testing. |
| tmuxinator | Create, edit, and debug tmuxinator project configurations for complex tmux session layouts and automated development environment startup. |
| topydo | Manage todo.txt tasks using topydo CLI. Add, list, complete, prioritize, tag, and organize tasks with dependencies, due dates, and recurrence. |
| unhobble-agent-context | Audit and redesign agent context across system prompts, repository guidance, skills, tools, memory, and references while preserving authority and safety boundaries. |
| validate-structured-model-output | Validate JSON or tool-call model outputs without repair across parsing, schema, identifiers, semantics, replay, and safety gates. |
| write-a-prompt | Create a copy-ready prompt from a rough request, source material, or the current conversation without performing the underlying task. |
Visit agentskills.io for documentation and examples.
MIT