for agents · SKILL.md

Every folder is a skill

Q: How do I install a skill into my agent?

One command: curl the folder's zip bundle and unzip it. You get a directory named after the skill holding SKILL.md, every file it references, a manifest. Drop that directory into your agent's skills folder — ~/.claude/skills/ for Claude Code, ~/.codex/skills/ for OpenAI Codex, ~/.qoder/skills/ for Qoder. Every SKILL.md ends with the paths for your agent, and each tutorial at the bottom of this page walks one agent end to end.

Each folder here has two bodies: the human one — run sheet, deck, props, poster, debrief — and the machine one, SKILL.md. An agent reads the frontmatter to match the folder to a brief, then runs the instructions to do the prep.

what a skill is, exactly

SKILL.md is one markdown file with two halves. The frontmatter is data — a spec-blessed core (name, description, version, license, allowed-tools) plus the match fields under metadata:, the same as catalog.json, so "a communication activity, pairs, 15 minutes, offline" resolves as a filter, not a search. The body is instructions — commands that actually run, named to match the Agent Skills convention.

one skill, two halves

--- name: blind-portrait description: A 15-minute, 2-person back-to-back drawing exercise that makes one-way communication visible — with printable props and a pre/post measure. version: 1.1.0 license: CC-BY-4.0 allowed-tools: Read, Write, Bash(curl:*), Bash(unzip:*), Bash(npm:*) metadata: # match fields — same as catalog.json category: communication group_size: 2 duration_minutes: 15 setting: offline depth: assessment-type updated: 2026-10-07 --- # Blind Portrait ## Run this activity with an agent # runnable prep — commands run as printed

the agent workflow, end to end

1 — discover

skills.json lists every skill with its description, bundle URL and sha256 — the short list an agent filters before touching anything else. catalog.json adds full folder data: category, group size, duration, setting, depth, byte sizes, review status. Anonymous, no key.

2 — install

One folder, one zip. curl it, unzip it — you get a directory named after the skill, holding SKILL.md, every file it references, a manifest: the shape skills loaders expect. The installer is curl; there is no SDK and no walled garden.

Every bundle ships the full kit: a stage-ready deck with presenter tooling and built-in QR codes, poker-spec print-and-cut cards, an A3 activity poster, manuals in A4 and US Letter, and a plain-language license with three tiers.

3 — run the prep

The skill's instructions cover the whole prep: fetch the printable files, build the live pre/post measure, grab the debrief questions. Each file carries a headless fallback for the login step, so a CI-run agent never stalls. Your agent runs them; you walk in prepared.

the three steps, as commands

# 1 — discover (anonymous) $ curl https://workshopskill.com/en/skills.json # 2 — install (anonymous): one command, loader-ready directory $ curl -LO https://workshopskill.com/en/activities/blind-portrait/files/blind-portrait.zip $ unzip blind-portrait.zip $ mv blind-portrait ~/.claude/skills/ # Claude Code; see the table below # 3 — run the prep (free FormLM account, for the live measure) $ npm install -g @formlm/cli $ formlm-cli auth login # headless? keep the curl steps, hand this line to a human $ formlm-cli smart generate --input "Pre/post measure for the Blind Portrait pairs exercise: 6 Likert items on instruction clarity, assumption-checking, perspective-taking and repair — the same 6 items before and after the activity, so each participant's delta is computed automatically"

where the skill goes, per agent

Agent	Skills directory (personal)	Project-level
Claude Code	`~/.claude/skills/blind-portrait/`	`.claude/skills/blind-portrait/`
OpenAI Codex	`~/.codex/skills/blind-portrait/`	`./.codex/skills/blind-portrait/`
Qoder	`~/.qoder/skills/blind-portrait/`	`.qoder/skills/blind-portrait/`
Gemini CLI	paste SKILL.md, or wire via `GEMINI.md` import	—
Cursor	no skills loader — convert SKILL.md to a `.cursor/rules/` rule	—
ChatGPT / Claude.ai	no shell — upload the zip in a chat, agent mode runs the curl steps	—
Anything else	any agent that reads a file and runs a shell: paste the file, follow it	—

Every SKILL.md ends with the two paths for your agent. The tutorials at the bottom of this page walk each agent end to end — install, first run, troubleshooting.

from folders to a modern workshop

The folder is the unit; the skill is how it moves. This is what changes for a facilitator whose agent can read the drawer:

Prep compounds. The brief — "12 people, 90 minutes, no projector" — resolves against catalog.json as arithmetic: six pairs, one prop set each, opener + activity + debrief fits the hour. The shopping list comes from the props field, not from memory.
Materials regenerate. guide.md and debrief.md are plain markdown: your agent rescales them to headcount, translates them, or turns them into a facilitator handout. The paper never disagrees with the page.
Measurement comes built-in. Assessment-type folders carry assessment.json — the preset behind the Kirkpatrick Level 2 question "did anyone actually learn anything". Step 3 turns the preset into a running pre/post form; each participant's delta is computed automatically, and the pair-level picture is the judge cards plus the debrief, on the table. The smile sheet becomes optional, not inevitable.

honest boundaries

These are open-format skills — SKILL.md with frontmatter and runnable instructions, over plain HTTPS. Any agent that can read a file and run a shell can use them; no lock-in, no proprietary runtime.
Scoring the pre/post measure offline needs no account — the method ships inside each assessment-type folder. A free FormLM account is only needed for the optional live form in step 3, and that form and its report live in your own workspace, not ours. Steps 1 and 2 are anonymous.
Humans print, agents parse: the PDF is the human artifact; markdown and JSON are the machine's — one folder, one set of facts.

the skills, so far

Forty-six skills are filed — one folder each; the list grows as new folders land. The catalog is the source of truth — anything not in catalog.json is not filed yet.

installed skills (46)

blind-portrait — communication · 2 people · 15 min · assessment-type paper-tower — communication · 8–20 people (teams of 3–5) · 30 min · assessment-type two-truths — icebreakers · 8–20 people · 15 min · assessment-type blanket-drop — icebreakers · 8–20 people (two teams) · 10 min · assessment-type common-ground — icebreakers · 8–20 people (groups of 3–5) · 15 min · assessment-type minefield — trust · 8–20 people (pairs) · 20 min · assessment-type silent-lineup — leadership · 8–20 people · 10 min · assessment-type the-orange — conflict · 2 teams of 2–4 · 20 min · assessment-type worst-idea-first — creativity · 8–20 people (groups of 3–4) · 20 min · assessment-type start-stop-continue — reviews · one team of 4–8 · 30 min · assessment-type post-training-survey — feedback-forms · any group size · 5 min to complete · assessment-type workshop-feedback-form — feedback-forms · any group size · 90 seconds to complete · assessment-type training-evaluation-form — feedback-forms · any group size · 7 min on the day + 3 min at day 30 · assessment-type advantage-compass — self-assessment · solo · 25 min to complete · assessment-type smile-sheet-alternative — feedback-forms · Any group size · 10 min · tool-type training-needs-survey — feedback-forms · Any group size · 15 min · tool-type exit-ticket — feedback-forms · Any group size · 3 min · tool-type the-listening-ladder — communication · 2 people (or trios) · 15 min · assessment-type relay-brief — communication · 8–20 people (lines of 4–6) · 20 min · assessment-type draw-what-i-mean — communication · 2 people (back to back) · 12 min · tool-type one-minute-meetup — icebreakers · 6–20 people · 15 min · tool-type paper-plane-intros — icebreakers · 8–20 people · 15 min · tool-type same-boat — conflict · 6–16 people (two sides) · 20 min · tool-type yes-and-ramp — creativity · 4–12 people · 15 min · tool-type four-lens-sweep — creativity · 4–10 people · 20 min · tool-type assumption-audit — creativity · 3–8 people · 20 min · tool-type storyboard-the-journey — creativity · 4–12 people · 25 min · tool-type escalation-map — conflict · 4–12 people · 25 min · tool-type the-delegation-map — leadership · 4–10 people · 25 min · tool-type the-reliability-ledger — trust · 4–10 people · 25 min · tool-type the-premortem — reviews · 4–12 people · 25 min · tool-type the-values-sort — self-assessment · 1–8 people · 20 min · tool-type the-trust-timeline — trust · 4–10 people · 25 min · tool-type the-not-anymore-list — reviews · 4–12 people · 25 min · tool-type the-leader-checkup — leadership · 3–10 people · 25 min · tool-type the-energy-audit — self-assessment · 1–2 people · 15 min · tool-type the-third-story — conflict · 4–10 people · 25 min · tool-type the-decision-rights-map — leadership · 4–10 people · 25 min · tool-type the-turning-point — reviews · 4–12 people · 25 min · tool-type the-feedback-digest — self-assessment · 1–2 people · 20 min · tool-type the-safety-check — trust · 4–12 people · 20 min · tool-type roleswap-argument — conflict · 6–16 people · 20 min · tool-type rotate-the-lead — leadership · 8–16 people · 25 min · tool-type mad-sad-glad — reviews · 6–20 people · 20 min · tool-type skill-radar — self-assessment · 6–16 people · 20 min · tool-type step-in-if — trust · 8–20 people · 15 min · tool-type

Open the folders: Blind Portrait · Paper Tower · Two Truths and a Lie · Blanket Drop · Common Ground · Minefield · Silent Lineup · The Orange · Worst Idea First · Start · Stop · Continue · Post-Training Survey · Workshop Feedback Form · Training Evaluation Form · The Advantage Compass · Beyond the Smile Sheet · Training Needs Survey · Exit Ticket.

skill tutorials — agent by agent, scenario by scenario

Twenty-eight standalone pages: how each AI agent installs the bundles and runs the prep, and how the folders carry real work — from a solo assessment to a day-30 report. Every page is copy-paste runnable on its own; the full index lives at skill-guides.

Claude Code — install into ~/.claude/skills, matching, first prep run.
OpenAI Codex — the skills folder plus an AGENTS.md pointer.
ChatGPT — upload, Projects instructions, and where the shell boundary sits.
Gemini CLI — GEMINI.md imports and the printed curl steps.
Cursor — convert a skill into an on-demand .cursor/rules file.
Windsurf — Cascade rules with activation modes.
Cline & Roo Code — .clinerules plus a real terminal.
Qoder — native skill directories, personal or committed.
Trae — one project rule that loads the whole drawer.
Zed — AGENTS.md, and ACP-hosted agents on the same folder.
Amp — a drawer outside any repo, one standing instruction.
opencode — the open-source terminal agent, same file contract.
GitHub Copilot — instruction files as the loader.
npx skills & registries — one-liner installs, self-hosting, when curl wins.
No skills loader? — the universal paste-and-run pattern.
Build your own agent — skills.json, verified bundles, matching loop.
Solo self-assessment — the Advantage Compass with your agent as proctor.
Learn the theory — executable folders plus AssessWiki's cited hubs.
Remix a skill — CC BY mechanics, from fork to marketplace listing.
QR codes & projector mode — folders paired with QRCARD-AI.
Live measure to report — assessment.json through FormLM.
The 30-minute prep sprint — the whole workflow on a stopwatch.
Remote & hybrid — what survives a video call, what you stop trying.
Retro to 30-day report — Start-Stop-Continue wired to measurement.

questions agents' humans ask

What is a workshop skill?

A SKILL.md file shipped inside every activity folder: frontmatter for matching (category, group size, duration, setting, props) plus runnable prep instructions for agents. Free under CC BY 4.0, like the rest of the folder.

Which AI agents can use these skills?

Any agent that can read a markdown file and run shell commands — coding agents, terminal assistants, or your own scripts. The files are plain text over HTTPS; there is no SDK and no authentication for reading and installing.

How do I install a skill into my agent?

One command: curl the folder's zip bundle and unzip it. You get a directory named after the skill — SKILL.md, every file it references, a manifest. Drop it into your agent's skills folder: ~/.claude/skills/ for Claude Code, ~/.codex/skills/ for OpenAI Codex, ~/.qoder/skills/ for Qoder. The tutorials at the bottom of this page walk each agent step by step.

Do I need an account for anything?

Browsing, downloading, printing, running the activities and scoring the pre/post measure all need no account — each assessment-type folder ships a standard-library method that computes the pair-level delta offline. Only building a live web form — step 3 of the agent workflow, or the “make it live” section on a folder page — uses formlm-cli and needs a free FormLM account; the form and its data stay in your own workspace.

A folder teaches a human to run the activity. A skill teaches their agent to prepare it. Both read the same catalog.json — what your agent fetches is what you see on the page.