8000 GitHub - wxxb789/hoa: Installable agent skills for Claude Code, Codex, OpenCode, Pi, and other AI coding agents via npx skills. Browse reusable skills for evidence, writing, and engineering, or use gdd for goal-driven delivery. · GitHub
[go: up one dir, main page]

Skip to content
wxxb789Public

About

Installable agent skills for Claude Code, Codex, OpenCode, Pi, and other AI coding agents via npx skills. Browse reusable skills for evidence, writing, and engineering, or use gdd for goal-driven delivery.

Topics

Resources

Contributing

Stars

3 stars

Watchers

0 watching

Forks

Repository files navigation

hoa: Agent skills and goal-driven delivery

English · 简体中文

hoa CI status License: MIT

hoa is an open collection of skills for AI coding agents: get ready-to-use Claude Code skills or install them for Codex, OpenCode, Pi, and other runtimes with npx skills, or use gdd for goal-driven delivery.

Install agent skills

Pick a skill from the catalog, then use its generated command in the install guide. For a standalone skill, for example:

npx skills add wxxb789/hoa -g -s official-document-verification

-g installs globally; omit it for the current project. Add -a <agent> to choose a runtime. For skills with siblings, use the generated command so they land in the same skills root: npx skills does not install dependencies automatically. To install the collection, use the generated Without Hermes command unless you also want the Hermes-only skills.

In the source tree each skill lives at skills/<category>/<name>/; the categories are a checkout convenience only — installs keep the flat <skills-root>/<name>/ layout and every command stays name-based.

What's inside

This is a monorepo of separate parts, not a harness you have to install as a unit. Use the parts you need; individual skills can have documented dependencies.

Agent skill catalog

The installable skills/ catalog is for people looking for reusable agent capabilities across software development, work management, and self-management. Start with the generated skill catalog.

  • Evidence and verification: audit claims, check official documents, and verify delegated work or measurement campaigns.
  • Skill lifecycle: hoa-skill-* skills create, review, port, curate, and audit skills. There are also explicit-only practice and repository retros.
  • Writing: draft and revise publishable prose, including analytical arguments backed by evidence.
  • Delivery: define deliverables, check package releases, and close out project phases.
  • Other capabilities: converge on a spec, consolidate research, and work with individual tools such as Discord.
  • Hermes operations (opt-in): hermes-* skills are only for people operating Hermes, including model routing, Kanban, updates, and runtime repairs.

gdd: Goal-driven delivery

For repository owners handing off a written, machine-checkable increment, gdd defines the delivery protocol: goal, staged agent runs, operator-run gates, and a scripted merge check. For a single external CLI handoff, lightweight delegation stops at independent acceptance without a stage ledger or merge. Install the agent route separately. OMP (gdd-route-omp) is the only model-visible route; DeepSeek Harness, Codex, Claude Code, Pi, OpenCode, Omo, and GitHub Copilot routes are hidden. The optional Paseo GDD Observer is a read-only plugin, not part of skill installation.

Start with the gdd delivery and transport guide.

Routing and skill necessity evals

For maintainers and evaluators: routing cases test which skill a request should select, while the necessity runner compares task outcomes with no skill, terse instructions, and the full skill. CI validates routing cases and the recorded baseline offline; live comparisons are separate.

Start with the evaluation runner guide.

Repository tooling and CI gates

For contributors maintaining hoa: scripts/ generates the indexes, checks vendored files, and validates metadata, links, composition, and revisions. rules/ defines the repository's standing constraints. These are repo-only tools and rules, not an installed runtime.

Start with the CI gate map.

Small library of roles, patterns, prompts, and recipes

For people assembling their own agent workflows: agents/ holds a role contract, orchestration/ a verification-gate pattern, prompts/ task templates, and workflows/ a finite recipe. This is a small reference library for manual reuse, not a package.

Start with the library guide.

External reference submodules

For people studying upstream agent, skill, and harness projects: ref/ maps external repos to patterns worth studying. The shallow submodules track upstream default branches via daily Dependabot updates. They are never deployed and are not needed to install skills or run repo tests.

Reflections, goals, and plans

For readers tracing how hoa evolved: reflections/ holds comparisons, decisions, and retrospectives; docs/goals/ and docs/plans/ record work contracts and plans. These are working and historical records, not runtime inputs.

Start with the notes and project history.

Supported runtimes

Use the catalog with Claude Code, Codex, OpenCode, Pi, Kimi Code, or another npx skills-supported runtime. Check individual skills for prerequisites and runtime-specific behavior; Hermes operations require Hermes, and gdd routes have their own CLI requirements. Skills that run Python scripts require uv and Python 3.13 or newer.

License and project links

MIT license · Changelog · Architecture · Contributing and skill requests

About

Installable agent skills for Claude Code, Codex, OpenCode, Pi, and other AI coding agents via npx skills. Browse reusable skills for evidence, writing, and engineering, or use gdd for goal-driven delivery.

Topics

Resources

Contributing

Stars

3 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages

0