skill tutorial · agent

Workshop skills in Claude Code

Claude Code loads every folder in its skills directory as an Agent Skill and matches by description. WorkshopSkill bundles ship in exactly that shape — one curl, one unzip, one move — so the whole discover → install → run loop closes in under two minutes.

What Claude Code can do with a workshop skill

Once blind-portrait is installed, Claude Code sees its name and description and pulls the full SKILL.md into context whenever your ask matches: prep a 15-minute pairs exercise for a communication session resolves to the skill, and the skill's run-as-printed commands fetch the prop cards, the debrief and the run sheet — then build the live pre/post measure through formlm-cli if you have a FormLM account.

Skills are context + instructions, not code: nothing runs until the agent executes a command you asked for, and each bundled command is an anonymous curl or an explicit CLI install.

Install — personal or project level

Personal skills (~/.claude/skills/) are available in every project; project skills (.claude/skills/) travel with the repo, so a training team can commit the drawer once and every member's Claude Code picks it up.

one skill into Claude Code

mkdir -p ~/.claude/skills cd "$(mktemp -d)" curl -LO https://workshopskill.com/en/activities/blind-portrait/files/blind-portrait.zip unzip blind-portrait.zip # -> blind-portrait/ (SKILL.md + every file + manifest) mv blind-portrait ~/.claude/skills/ # project-level instead: same last step, different target mv blind-portrait ./.claude/skills/

First run

Start a new Claude Code session in your workspace folder and ask in plain terms:

the ask, and what comes back

> prep blind portrait for 12 people, offline, 90-minute session Claude loads the skill, then: fetches the poker-spec prop-card deck, paper-form, debrief, run sheet (curl steps) computes: 12 people = 6 pairs, one deck per pair, 2+15+2 = 19 min asks whether to build the live pre/post measure (formlm-cli) hands you the shopping list and the print order

Troubleshooting

Skill not matching? Descriptions carry the trigger vocabulary (group size, duration, category). Say the constraint out loud, or name the skill: use the blind-portrait skill.
Stale copy? Each skill has metadata.updated and a manifest.json with per-file hashes; re-curl the bundle when the catalog's updated date moves.
formlm-cli stalls at login? The skill's headless fallback covers it: run the curl steps, keep the assessment.json, hand the login line to a human.

keep going

The skills, end to endDiscover, install, run — the three-step agent workflow on one page. Install into OpenAI CodexSame bundle, different loader: the ~/.codex/skills route. No skills loader? Paste-and-runFor any chat-only agent — the universal fallback.

questions

Does Claude Code need a restart to see a new skill?

No. Skills are re-scanned per session; start a new conversation (or clear context) after moving the directory in.

Can the agent run the whole prep without me?

The curl steps run unattended — they are anonymous downloads. Only the FormLM live-measure step needs a human login, and the skill tells the agent to stop there and ask.

Are these skills free for commercial workshops?

Yes — CC BY 4.0: run it in paid workshops, rebrand it, keep the attribution line.