Reusable skills and agent roles for coding, documentation, code review, and education-content production. Canonical repository data produces first-class Claude and Codex installations, with maintained Cursor, OpenCode, Grok, and Hermes compatibility outputs.
Clone the repository, then run the guided installer from the repository root. It installs beneath the current operating-system home and defaults to platforms with existing skill directories. Claude and Codex are the default when no skill directories are found.
./install_skills.pyThe interview marks the skill directories it finds and shows the destinations before asking for confirmation. You can enter any supported combination of platforms. See docs/INSTALL.md for destinations and update behavior.
Skills are linked directly from each platform destination back to this clone, so the installer does
not duplicate skill trees and source edits are immediately visible. Codex preserves the repository
categories at ~/.codex/skills/<category> (Hermes does the same); Claude remains flat at
~/.claude/skills/<skill> (as do Cursor, OpenCode, and Grok). Authored Claude agents use the same
link model for Claude and Grok. Native Codex, Cursor, and OpenCode agent projections are the only
generated files. Re-running the installer unlinks stale links into this clone left by renamed skills.
skills/<category>/<name>/SKILL.mdcontains each skill's authored instructions.CATEGORY.mdprovides the category title, description, order, visibility, and required paths. Every skill also carriesagents/openai.yamlwith display metadata and a default prompt.agents/<name>.mdcontains the authored Claude-compatible agent instruction body. agents/CATALOG.yaml supplies each agent's identity, responsibility, Gas Town role, access level, authority, and escalation metadata.install_targets/<platform>/TARGET.mddeclares an adapter, support tier, skill layout, and destinations. Claude and Codex areprimary; Cursor, OpenCode, Grok, and Hermes arecompatibility.- install_skills.py and install_lib/ create reproducible local installations without installer state or hidden configuration.
Generate or check published artifacts from their canonical sources:
source source_me.sh && python3 index_lib/build_all.py --checkThe index and manifests come from skills and category metadata. The searchable agent index comes from the agent catalog plus authored Markdown, while installation renders agent adapters directly from those canonical sources. See docs/USAGE.md for maintenance commands and docs/CODE_ARCHITECTURE.md for extension points.
- docs/INSTALL.md: Platform tiers, destinations, and installer lifecycle.
- docs/USAGE.md: Canonical sources, generated outputs, and verification.
- docs/SKILLS_INDEX.md: Generated skill catalog.
- docs/AGENTS_INDEX.md: Generated agent roles and authority index.
- Game design expert: Controls, surprise, visual identity, progression, and independent assessment for new and existing games.
- docs/CODE_ARCHITECTURE.md: Components, data flow, and extension points.
- docs/FILE_STRUCTURE.md: Directory map and generated-output owners.
- docs/TROUBLESHOOTING.md: Common checks and recovery guidance.
Representative workflow skills include audit-code-reviewer, blueprint-plan-drafter,
delegate-manager-to-subagents, docset-updater, readme-docs, and repo-rules-reader.
Domain-expert skills cover computational geometry, Apple Liquid Glass, computer vision, CSS,
Podman, PySide6, SolidJS, TypeScript, PDF work, and education-content generators. Browse the
complete generated catalog in docs/SKILLS_INDEX.md.
- Agent Skills standard: Overview of the open
SKILL.mdformat. - Anthropic Skills: Claude skill examples and resources.