skill tutorial · agent
Workshop skills in opencode
opencode is the open-source terminal agent that reads the same conventions its cousins do: AGENTS.md for standing instructions, rules files for context, a real shell for the printed commands. A WorkshopSkill bundle installs into any of those shapes.
Install
personal drawer
mkdir -p ~/.config/opencode/skills && cd "$(mktemp -d)" curl -LO https://workshopskill.com/en/activities/two-truths/files/two-truths.zip unzip two-truths.zip mv two-truths ~/.config/opencode/skills/
Standing instructions
~/.config/opencode/AGENTS.md
# Workshop skills ~/.config/opencode/skills/<name>/SKILL.md — Agent Skills format, index: https://workshopskill.com/en/skills.json On prep briefs: match metadata, read the one match, run its commands as printed. Ask before account-login steps.
Per-project variant
Committing ./skills/ plus a project AGENTS.md to the training repo makes the drawer portable: anyone who clones gets the folders, whoever opens opencode there gets the same behavior — no global config required.
First run
opener for 12 strangers, 15 minutes, guardrails needed
resolves to two-truths (its description carries guardrails); opencode fetches the card deck, prints the eyes-closed voting rule from the run sheet, and offers the 4-item pre/post measure with the same consent gate.
keep going
questions
Does opencode have skills yet?
It gains features quickly; either way the AGENTS.md pattern is behaviorally identical — and survives version changes because it is just files.
Can it approve the curls once?
Permission modes per tool pattern: allow curl to workshopskill.com and the prep runs clean.
Does it work with local models?
Anything opencode runs — matching is model-side, knowledge is file-side. Smaller models: name the skill explicitly.