Physical Skill Cards: The Structured Format for How-To Knowledge
Every tradesperson carries knowledge that took years to earn. The plumber who knows exactly how much thread tape to use on a 15mm fitting. The electrician who can identify a bad neutral by the hum of a light fitting before they've opened a single panel. The carpenter who angles a nail at 30 degrees through a stud not because someone told them to, but because they've driven 10,000 nails and they know what holds.
That knowledge — hard-won, tacit, procedural — lives mostly in heads and in informal conversation. It gets passed down through apprenticeships, YouTube videos, and dog-eared instruction manuals. But it's not structured. It's not machine-readable. And increasingly, as AI moves from language tasks into physical action, that's becoming a problem.
Physical Skill Cards are the solution.
What is a Physical Skill Card?
A Physical Skill Card is a structured, standardised format for capturing procedural knowledge about real-world tasks. Not concepts. Not theory. The actual steps required to do something physical — with tolerances, tools, materials, failure modes, and safety warnings baked in.
Think of it as the intersection of three things:
- The precision of an engineering spec sheet
- The accessibility of a how-to article
- The machine-readability of structured data
The result is a document that a human can follow step by step, and that a machine agent — whether that's a home robot, an AI assistant guiding a repair, or an automated quality control system — can parse, reason about, and execute.
Physical Skill Cards are not recipes (though recipe cards follow a similar logic). They're not manuals (which are typically exhaustive reference documents). They're not YouTube videos (which require you to watch, guess timestamps, and translate visual demonstrations into physical action). A Physical Skill Card is a single, bounded, executable procedure.
The Format Specification
A complete Physical Skill Card contains the following fields:
Task Name — A clear, action-oriented title. "Hang a picture frame" not "Picture hanging." Verb-noun format.
Difficulty — A 1–5 scale:
- 1: No experience needed
- 2: Basic household competence
- 3: Some DIY or trade experience helpful
- 4: Trade-level competence required
- 5: Professional qualification recommended
Time Estimate — Realistic time including setup and cleanup. Format: "15–20 minutes" (always a range, never a point estimate — tasks vary).
Tools Required — An explicit list. No "and any other tools you might need." If you need a stud finder, list it. If you need a specific drill bit size, specify it.
Materials — Consumables, fixings, and materials. Include quantities and specifications (e.g., "2× M6 × 40mm wall anchors" not just "wall anchors").
Prerequisites — What the operator needs to know or have done before starting this skill. Can reference other skill cards. Example: "Wall surface must be dry. If patching was done, patching must be fully cured (minimum 24h)."
Step-by-Step Procedure — Numbered steps. Each step is a single action. Include:
- The action ("Drive the screw until resistance increases")
- The tolerance ("…until resistance increases but the anchor has not yet compressed — stop before the wall surface deforms")
- Any "check" steps ("Verify the anchor is flush with the wall surface")
Safety Warnings — Flagged separately. Not buried in steps. Explicit about the failure mode: "If drilling into a plaster-over-brick wall without first checking for cables, you risk hitting electrical wiring. Use a cable detector before drilling."
Failure Modes — What does it look like when this goes wrong? How do you recognise a failure in progress? What do you do? This section is often missing from how-to content and is what separates a Skill Card from a tutorial.
Tips — Practitioner notes that don't fit cleanly into the procedure. "Professional painters use a chalk line, not a spirit level, for picture arrangements over 3 pieces because chalk lines are faster and erasable." This is the tacit knowledge layer.
Why Machine-Readable Matters
Here's the uncomfortable truth about most how-to content: it was written for a human sitting at a computer, not for an agent trying to guide a physical action.
YouTube videos assume you can pause and replay. WikiHow assumes you can zoom in on a photo. Instruction manuals assume you have 45 minutes to read before you start. None of that works when:
- Your hands are inside a wall cavity and you need guidance via an earpiece
- A home assistant robot is attempting a simple repair without human intervention
- An AI agent is checking a technician's work against a quality standard in real time
- A field service worker in a remote location needs step-by-step guidance with no connectivity to video
Machine-readable procedural knowledge changes all of these scenarios. If the steps are structured — typed actions with explicit tolerances, discrete conditions, clear success criteria — then a language model can serve as a real-time guide. An embodied AI can cross-reference current sensor data against expected states. A quality system can verify that each step was completed before allowing the next.
The format isn't an academic exercise. It's infrastructure for a world where AI is increasingly involved in physical work.
A Worked Example: Hanging a Picture Frame
Here's a complete Physical Skill Card in the standard JSON schema format used by Skills Warehouse:
{
"skillCard": {
"version": "1.0",
"type": "physical",
"id": "hang-picture-frame-plasterboard",
"title": "Hang a Picture Frame on a Plasterboard Wall",
"difficulty": 1,
"timeEstimate": "10–15 minutes",
"tools": [
"Pencil",
"Tape measure",
"Spirit level (mini, 300mm)",
"Hammer",
"Electric drill with 6mm masonry bit (for wall anchor variant)",
"Screwdriver (matching screw head type)"
],
"materials": [
"Picture hook rated for frame weight (check packaging — standard hook: up to 10kg)",
"OR: 1× 6mm plasterboard anchor + 1× M4 × 30mm screw (for frames over 5kg)",
"1× pencil mark",
"Optional: 1× length masking tape (prevents plasterboard crumbling when drilling)"
],
"prerequisites": [
"Know the approximate weight of the frame (check back of frame or estimate)",
"Know if wall is plasterboard (hollow sound when tapped) or brick/concrete (solid)",
"This card covers plasterboard walls only — see 'Hang picture frame on masonry' for brick/concrete"
],
"procedure": [
{
"step": 1,
"action": "Mark the desired centre-top position of the frame on the wall with a pencil.",
"tolerance": "Mark should be light — erasable later.",
"check": null
},
{
"step": 2,
"action": "Measure down from the mark by the distance between the hanging wire/hook on the back of the frame and the top edge of the frame.",
"tolerance": "Typical range: 30–80mm depending on frame design.",
"check": "Confirm the new mark is where the hook/anchor will sit."
},
{
"step": 3,
"action": "For frames under 5kg: hammer the picture hook nail at 45 degrees upward through the hook's nail guide at the second mark.",
"tolerance": "Nail angle should be 40–50 degrees from horizontal — steeper angle = stronger hold.",
"check": "Hook should feel firm. No wobble or movement in the plasterboard."
},
{
"step": 4,
"action": "For frames over 5kg: apply a 30mm square of masking tape over the mark, then drill a 6mm hole 30mm deep at the mark.",
"tolerance": "Drill at 90 degrees to the wall surface. Masking tape prevents plasterboard paper from tearing.",
"check": "Hole should be clean-edged. No large chips or cracks in surrounding plaster."
},
{
"step": 5,
"action": "For wall anchor variant: insert the 6mm plasterboard anchor into the hole, flat side flush against the wall.",
"tolerance": "Anchor face should be flush — not proud (protruding) and not recessed.",
"check": "Anchor should not spin freely in hole. If it does, hole is too large — see failure modes."
},
{
"step": 6,
"action": "Drive the M4 × 30mm screw into the anchor, leaving 5mm of screw protruding to hang the frame wire on.",
"tolerance": "Do not overtighten — plasterboard anchors grip by expanding behind the board. Stop when resistance noticeably increases.",
"check": "Screw should feel solid. No continued movement after initial resistance increase."
},
{
"step": 7,
"action": "Hang the frame. Use the spirit level to verify it is horizontal.",
"tolerance": "Bubble centred in the spirit level vial. Less than 1 degree of tilt is acceptable for most frames.",
"check": "Step back 2m and visually confirm level."
}
],
"safetyWarnings": [
{
"hazard": "Hidden electrical cables",
"detail": "Plasterboard walls may contain electrical wiring running vertically from power points and horizontally between studs. Use a cable/stud detector before drilling anywhere except directly over a confirmed safe zone.",
"consequence": "Risk of electrocution and/or tripped circuit breaker."
},
{
"hazard": "Load rating",
"detail": "A standard picture hook nail in plasterboard holds up to approximately 10kg under static load. Do not use a single hook for frames heavier than rated — use two anchor points.",
"consequence": "Frame falls, damages wall, may injure people below."
}
],
"failureModes": [
{
"failure": "Anchor spins in hole",
"cause": "Hole drilled too large, or wall is too thin for standard anchor.",
"fix": "Use a larger anchor (10mm) or toggle bolt for very thin plasterboard. Do not try to hold the anchor still by hand while driving the screw — this rarely works."
},
{
"failure": "Hook nail bends on entry",
"cause": "Hit a stud or dense patch. Nail angle too shallow.",
"fix": "Remove nail. Reposition 20mm left or right to avoid stud, or steepen angle. Do not try to hammer a bent nail — it will not hold."
},
{
"failure": "Frame hangs at an angle despite level mounting",
"cause": "Hanging wire on back of frame is off-centre, or wire is attached at unequal heights.",
"fix": "Adjust wire position on frame hooks before re-hanging. Measure from both ends of wire to top of frame to verify symmetry."
}
],
"tips": [
"For gallery walls with multiple frames, use paper templates (trace each frame on paper, cut out, tape to wall) before driving a single nail. Rearrange freely until composition is right.",
"Chalk lines are faster than spirit levels for horizontal arrangements of 3+ frames.",
"If the plasterboard is textured (sand texture, orange peel), the hook nail will feel slightly more resistance — this is normal."
]
}
}
This is a complete, executable Skill Card. Every field has a purpose. Nothing is missing. An AI agent can parse this JSON and:
- Ask clarifying questions based on the prerequisites
- Guide a user step by step, confirming each check before proceeding
- Flag safety warnings at the appropriate moment in the procedure
- Diagnose failures by matching observed symptoms to the failure modes table
How Physical Skill Cards Differ from YouTube Tutorials
YouTube is extraordinary for learning. But it is a terrible format for execution.
| | YouTube Tutorial | Physical Skill Card | |---|---|---| | Format | Video, linear | Structured data, any interface | | Navigation | Scrub to timestamp | Jump to any step | | Machine-readable | No | Yes | | Success criteria | Implied | Explicit | | Failure modes | Rarely covered | Required field | | Tolerances | "Roughly this much" | Specific measurements | | Safety | Often omitted | Mandatory field | | Ads | Yes | No | | Searchable within step | No | Yes | | Usable by AI agent | No | Yes |
The YouTube creator's incentive is watch time. The Physical Skill Card's purpose is successful execution. These are fundamentally different goals, and they produce fundamentally different formats.
The Embodied AI Angle
We are at the beginning of a shift in AI capability. Language models are moving from text tasks to physical tasks — through robotics, computer use, and AI-assisted repair and manufacturing. The next generation of tools won't just write code; they'll help you fix your sink.
But embodied AI needs a knowledge layer. A robot doesn't benefit from watching a YouTube video. It needs a structured description of the physical state it should find at each step, the action to take, and the tolerance within which that action is acceptable.
Physical Skill Cards are that knowledge layer. They are to embodied AI what training data is to language models — the raw structured knowledge that allows a system to act correctly in the physical world.
This is why Skills Warehouse exists for physical skills, not just agentic software procedures. The future of AI isn't only software automation. It's a world where AI assists — and eventually performs — the physical work that currently requires trained humans to be physically present.
When a home AI assistant knows how to hang a picture frame, install a tap washer, or reset a circuit breaker, it will be because someone wrote a Physical Skill Card for it. That card will be here.
Start Contributing
Skills Warehouse is accepting Physical Skill Card submissions during the founding creator period. If you have trade knowledge, domain expertise, or procedural know-how that deserves to be preserved in structured form — we want it.
Founding creators lock in a 75/25 revenue split and get a permanent "Founding Creator" badge on their profile.
The knowledge in your head has value. Time to put it somewhere machines can find it.
Be first right now
The marketplace is open. Publish free, founding-creator terms, and a permanent Warehouse Verified badge.
Publish your skill →