A collection of skills for the TritonAI harness (a Codex-based agent). Each top-level directory is a self-contained skill, source-controlled here and synced to GitHub for sharing with the team.
Each skill is a standard Codex skill: a directory containing a SKILL.md with name and
description frontmatter, plus optional references/, scripts/, assets/, and agents/
resources. The description frontmatter controls when the skill is triggered. Skills may
carry their own README.md with a scoped overview and any provenance or creation records —
see each directory.
- From the TritonAI Harness, click
Settings(gear icon) thenSkills - Scroll to the "Add From Link" section
- Add the URL for the skill directory in this repository, e.g.,
https://github.com/ucsdlib/dx-tools/tree/main/uc-library-search
- Follow the standard Codex skill structure; keep
SKILL.mdlean and move reference material intoreferences/. - Keep developer-only artifacts (e.g., testing logs) as hidden dotfiles (
.name) so agent file discovery ignores them, and never reference them from the skill itself. - Give each skill its own
README.mdfor a scoped overview; keep it current with the skill. - Use descriptive commit messages; commit history is the changelog.
- Authored by Doug Worsham, Digital Experience Manager, UC San Diego Library, with AI assistance (developed iteratively with Claude/Anthropic and Codex/OpenAI).
- 2026-09-15 — the collection was added to this repo (commit
128176c). Skills previously distributed as Claude-style.skillZIP files; the plain directories here are now the source of truth. - 2026-09-15 — adapted for the TritonAI/Codex harness: search/read guidance maps to the
collaborative browser (
preview_*) andexec_command, and model identification is provider-agnostic.
Claude Code consumes skills as single-file .skill ZIP packages. Rebuild one from its
directory if needed:
zip -r <skill-name>.skill <skill-name>/Unless noted in a skill's own files, skills do not yet carry a license — decide and add one before public sharing if needed.