Public PAI skill tree.
A readable map of the public agent skill collection: router skills, leaf workflows, operational tooling, and visualization systems exported without private workspace state.
Agents route by metadata, then load only what the task needs.
This atlas embeds all public skill metadata for human browsing. Agent runtimes should not eagerly load every skill body. They use the standard skill pattern: metadata for routing, activated instructions for execution, and resource files only when needed.
Small frontmatter metadata is available for routing. The description carries the trigger contract.
The full body loads only after the user request matches a relevant skill description.
Scripts, references, workflows, and assets are read or executed on demand, not preloaded.
Every folder with its own `SKILL.md` is represented below, including nested skills such as Beads audits, Frontend/Impeccable commands, n8n specialists, document tools, and browser attach workflows.
Required frontmatter, public links, local-path hygiene, and secret-pattern gates pass. Large-skill warnings remain visible as candidates for future progressive splitting.
Broad routers first, concrete workflows below.
The tree is intentionally uneven. Root skills act as routers. Subskills are concrete operating guides with their own descriptions, paths, and activation triggers.
Filter by category, name, description, or path.
Use this as the practical navigation layer before installing or linking a skill into Codex, Claude, or another agent runtime. Click any card for details; searches also surface nested workflows, references, tools, prompts, configs, and data files.