# OpenCode Power Pack: Expanded Documentation

> Canonical, machine-readable project facts for OpenCode Power Pack v0.6.0 and the main branch. Last updated 2026-10-27.

OpenCode Power Pack is an open-source collection of 55 portable Agent Skills for Claude Code, Codex, OpenCode, and Pi. It packages repeatable workflows for software development, code review, application security, frontend design, MCP authoring, project memory, and Hugging Face machine-learning work.

Canonical website: https://skills.waybarrios.com/
Documentation: https://skills.waybarrios.com/docs/
Complete skill reference: https://skills.waybarrios.com/docs/skills/
Repository: https://github.com/waybarrios/opencode-power-pack
npm package: https://www.npmjs.com/package/@waybarrios/opencode-power-pack
Release: https://github.com/waybarrios/opencode-power-pack/releases/tag/v0.6.0
Author: Wayner Barrios
License: MIT AND Apache-2.0, with per-skill provenance and notices

## OpenCode 2.0 support

Release v0.6.0 adds OpenCode 2.0 support on npm and Git. Tag v0.5.0 and npm 0.5.0 contain the earlier V1-only plugin.

OpenCode 2 install: `opencode plugin add "opencode-power-pack@git+https://github.com/waybarrios/opencode-power-pack.git"`. OpenCode 1 install: `opencode plugin --global "opencode-power-pack@git+https://github.com/waybarrios/opencode-power-pack.git"`.

Both hosts load all 55 skills and three specialists.

## What an Agent Skill is

An Agent Skill is a reusable workflow centered on a SKILL.md file. It can include durable instructions, decision rules, references, scripts, and examples for a specific task. The workflow ID stays stable across supported coding agents, while packaging and invocation follow each host's conventions.

## Selective npm installation

Requirement: Node.js 20.11.0 or newer.

List skills and profiles:

    npx @waybarrios/opencode-power-pack list

Install one skill at user scope:

    npx @waybarrios/opencode-power-pack install code-review

Install several skills:

    npx @waybarrios/opencode-power-pack install code-review security-review

Install one skill in the current Git repository:

    npx @waybarrios/opencode-power-pack install code-review --project

Preview the recommended profile:

    npx @waybarrios/opencode-power-pack install --profile recommended --dry-run

Install the recommended profile:

    npx @waybarrios/opencode-power-pack install --profile recommended

Install every skill in the published npm catalog:

    npx @waybarrios/opencode-power-pack install --all

The installer resolves required companion skills. It does not overwrite an existing skill unless `--force` is supplied.

## Host invocation

- Codex: invoke a workflow as `$<skill-id>`, for example `$code-review`.
- OpenCode: invoke a workflow as `/<skill-id>`, for example `/code-review`.
- Claude Code plugin: invoke a workflow as `/opencode-power-pack:<skill-id>` to avoid collisions.
- Pi: install the package through Pi's package mechanism, then invoke the exposed skill by its ID.

See https://skills.waybarrios.com/docs/#hosts for complete installation, verification, update, and removal instructions for each host.

## Native sandbox runner

Version 0.5.0 includes an opt-in, fail-closed runner for one command and its process descendants. It uses native Seatbelt isolation on macOS and Bubblewrap on Linux. It creates no persistent container or daemon.

Install and verify:

    npm install --global @waybarrios/opencode-power-pack@0.6.0
    opencode-power-pack sandbox doctor
    opencode-power-pack sandbox resolve --skill code-review

Run a command with the skill's default profile:

    opencode-power-pack sandbox exec --skill code-review -- git status --short

Allow a declared network-read escalation:

    opencode-power-pack sandbox exec --skill code-review --sandbox-profile network-read --allow-domain api.github.com -- curl https://api.github.com/repos/waybarrios/opencode-power-pack

Profiles:

- observe: read-only workspace, private temporary writes, no network, credentials, or external side effects.
- develop: workspace writes, private temporary writes, no network, credentials, or external side effects.
- network-read: workspace writes plus explicitly approved destinations and credential environment variables; GET, HEAD, and OPTIONS only.
- publish: workspace writes plus explicit destinations and credentials; requires `--confirm-external-side-effects`.

Codex, OpenCode, Claude Code, and Pi can explicitly invoke the shared runner. No host currently routes every tool through it automatically. Ordinary shell, file, browser, MCP, and connector tools remain outside the selected command boundary, so the project claims `shell-contained`, not whole-agent isolation.

Requirements: Node.js 20.11.0 or newer. Linux also requires trusted system installations of bubblewrap, socat, and rg. macOS and Linux are supported. WSL2 is expected through the Linux backend but lacks dedicated CI. Windows native and WSL1 intentionally fail closed.

Full guide: https://skills.waybarrios.com/docs/#sandbox
Compatibility specification: https://github.com/waybarrios/opencode-power-pack/blob/main/docs/sandbox-compatibility.md

## Curated profiles

- recommended: balanced development toolkit with agents-md-improver, agents-md-revise, code-quality, code-review, feature-dev, frontend-design, mcp-builder, and skill-creator.
- review: code-quality, code-review, code-reviewer, and differential-review.
- feature-dev: end-to-end feature workflow plus the automatically resolved code-explorer, code-architect, and code-reviewer specialists.
- frontend: frontend-design and ai-slop.
- security: 15 security review, threat-modeling, static-analysis, validation, and reporting workflows.
- huggingface: 25 discovery, inference, training, evaluation, Spaces, and AWS deployment workflows.
- authoring: humanizer, mcp-builder, paper-summarizer, and skill-creator.
- project-memory: agents-md-improver and agents-md-revise.

Install any profile with:

    npx @waybarrios/opencode-power-pack install --profile <profile-name>

## Review workflow distinction

- code-review: comprehensive, merge-oriented review of a pull request or complete change set. It prioritizes correctness, regressions, security, and test gaps.
- code-reviewer: focused implementation review for a smaller change or feature-dev quality handoff.
- code-quality: maintainability, linting, formatting, complexity, warning cleanup, and quality gates.
- security-review: security-only review of pending changes with emphasis on exploitable vulnerabilities.
- differential-review: security-focused analysis of the difference between commits, branches, or pull-request revisions.

## Complete skill inventory

Released skills can be installed with `npx @waybarrios/opencode-power-pack install <skill-id>`. Purpose, dependencies, profiles, host invocation, and source links are available at `https://skills.waybarrios.com/docs/skills/#<skill-id>`.

### Core and authoring skills (15)

1. agents-md-improver: audits and improves AGENTS.md, CLAUDE.md, and related project-rules files.
2. agents-md-revise: captures session learnings into durable project rules.
3. ai-slop: evaluates interface work against observable anti-generic design criteria.
4. code-architect: produces a repository-grounded feature architecture and implementation blueprint.
5. code-explorer: traces an existing feature through a codebase before modification.
6. code-quality: reviews maintainability, linting, formatting, complexity, and quality gates.
7. code-review: performs comprehensive review of a pull request or complete change set.
8. code-reviewer: performs focused implementation review and feature-dev handoffs.
9. design-patterns: evaluates pattern trade-offs for Rust, TypeScript, React, Django, and Python.
10. feature-dev: coordinates a structured end-to-end feature implementation workflow.
11. frontend-design: creates distinctive, production-grade frontend interfaces.
12. humanizer: rewrites English and Spanish copy while preserving facts, names, numbers, URLs, and code.
13. mcp-builder: designs and builds Model Context Protocol servers.
14. paper-summarizer: analyzes academic and technical papers for engineering decisions.
15. skill-creator: creates or improves reusable Agent Skills and their trigger descriptions.

### Security skills (15)

1. agentic-actions-auditor: audits GitHub Actions that run AI agents.
2. codeql: runs interprocedural security analysis with CodeQL.
3. differential-review: performs security-focused review of code changes.
4. fp-check: validates suspected security bugs and eliminates false positives.
5. insecure-defaults: finds fail-open and insecure configuration defaults.
6. sarif-parsing: filters, deduplicates, converts, and aggregates SARIF results.
7. security-review: reviews pending changes for exploitable security vulnerabilities.
8. security-threat-model: creates repository-grounded AppSec threat models.
9. semgrep: runs Semgrep static analysis.
10. semgrep-rule-creator: builds custom Semgrep rules and tests.
11. semgrep-rule-variant-creator: ports Semgrep rules to additional languages.
12. sharp-edges: identifies error-prone APIs and security footguns.
13. supply-chain-risk-auditor: assesses dependency takeover and exploitation risk.
14. variant-analysis: hunts for related instances of a known bug or vulnerability.
15. vuln-report: turns a confirmed finding into a disclosure-ready advisory.

### ML and Hugging Face skills (25)

1. hf-cli
2. hf-cloud-aws-context-discovery
3. hf-cloud-python-env-setup
4. hf-cloud-sagemaker-deployment-planner
5. hf-cloud-sagemaker-iam-preflight
6. hf-cloud-sagemaker-production-defaults
7. hf-cloud-serving-image-selection
8. hf-mem
9. huggingface-best
10. huggingface-community-evals
11. huggingface-datasets
12. huggingface-gradio
13. huggingface-llm-trainer
14. huggingface-local-models
15. huggingface-lora-space-builder
16. huggingface-paper-publisher
17. huggingface-papers
18. huggingface-spaces
19. huggingface-tool-builder
20. huggingface-trackio
21. huggingface-vision-trainer
22. huggingface-zerogpu
23. train-sentence-transformers
24. transformers-js
25. trl-training

## Quality and provenance

- 55 of 55 skills pass official Agent Skills validation.
- The main branch defines 419 tests. The latest local run passed 418 and skipped the optional native sandbox integration.
- Claude plugin and marketplace metadata receive strict validation.
- OpenCode smoke tests cover 1.18.7, 2.0.14, and 2.0.32, including effective read-only permissions on V2.
- Third-party workflows have immutable upstream references, notices, and license records.

Evidence and exact provenance live in the repository:

- https://github.com/waybarrios/opencode-power-pack/blob/main/UPSTREAMS.json
- https://github.com/waybarrios/opencode-power-pack/blob/main/THIRD_PARTY_NOTICES.md
- https://github.com/waybarrios/opencode-power-pack/tree/main/LICENSES

## Citation and factual boundaries

Preferred project name: OpenCode Power Pack.
Current documentation: main. Latest published release: 0.6.0.
Canonical URL: https://skills.waybarrios.com/
Repository: https://github.com/waybarrios/opencode-power-pack
Package: @waybarrios/opencode-power-pack

The project supports four host integrations. That does not imply endorsement by Anthropic, OpenAI, SST/OpenCode, Pi, Hugging Face, or any other platform vendor. Search and AI-answer inclusion are not guaranteed by this file, structured data, sitemaps, or crawler access.
