AI Skill Encoding Stack
Turn recurring work into an executable process with examples and checks
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 6
- Confidence
- 99%
An AI skill packages a recurring unit of work into three layers. First, encode the process: the ordered actions the agent should perform whenever the skill runs. Second, supply examples that demonstrate what good output looks like, giving the agent a concrete quality target. Third, add scripts or deterministic checks when the task is complex enough to require consistent validation. A practical way to capture tacit work is to ask the AI to interview the operator one question at a time about how the process is performed, then have it formalize and execute that process. Once tested, the skill can run on demand or through automation. The result is not another specialized agent but a reusable capability that one general agent can select when appropriate.
Origin
Joshua Wöhle described Rebel skills as the process, examples, capabilities, and optional scripts that let one agent execute specialized work reliably.
Core principles
- 01Reliable AI work needs an explicit process
- 02Examples define what good output looks like
- 03Deterministic checks stabilize complex execution
- 04One capable agent can select among thoroughly developed skills
How to run it
- 1
Select the unit of work
Choose a recurring task with a clear trigger, output, and quality standard.
Pro tip Start with work that is frequent, laborious, neglected, or routinely done below standard.
Watch out Do not encode an unstable process before understanding it.
- 2
Extract the process
Ask the AI to interview the operator one question at a time and document how the task is actually completed.
Watch out Capture exceptions and decision points, not only the happy path.
- 3
Define what good looks like
Attach representative examples of successful outputs and explain the important qualities they share.
Pro tip Use real examples whenever available.
- 4
Add deterministic support
For complex tasks, add scripts or fixed checks that verify required properties and help execution remain consistent.
Pro tip Reserve scripts for steps that must behave predictably every time.
- 5
Test and refine
Run the skill on real work, compare the result with the examples, and correct process gaps.
Watch out Do not automate an unreliable skill.
- 6
Deploy on demand or schedule
Let the agent invoke the proven skill when requested or trigger it automatically in the appropriate workflow.
Watch out Match autonomy to the risk of the action.
In the wild
Wöhle built a skill that checks Slack, email, WhatsApp, and HubSpot for deals without recent activity, then drafts an email to revive each forgotten opportunity. The workflow combines a repeatable search process with contextual communication.
→ Dormant deals are surfaced and given a tailored follow-up without a manual cross-system review.
Hala Taha encoded podcast qualifications such as downloads, reviews, social links, and contact details. The skill researches prospects, crafts a custom email, finds addresses, and saves drafts in her inbox.
→ A multi-step prospecting process becomes a reusable AI capability.
Common mistakes
Writing instructions without examples
A process says what to do, but examples show the agent the quality and form expected from the result.
Automating before proving reliability
Scheduling an untested skill scales errors and can turn a weak workflow into an operational risk.
Is it for you?
Best for
It is best for frequent, multi-step knowledge work with recognizable standards and examples.
Not ideal for
It is not ideal for rare, ambiguous work whose process or definition of quality changes every time.
From the transcript
“So a skill is a combination of a process. Basically a process that the AI can execute”
“It is a set of examples so that the AI understands what does good look like”
“sometimes a skill can include some scripts, which is basically some deterministic steps”
From the episode
Joshua Wöhle: Turn AI Into Your Competitive Advantage and 50X Your Productivity
Joshua Wöhle