AIFC
Back to version

Published version: AIFC-V002. This is the latest published version. All versions.

AIFC-044: Human Skills and AI Skills

Status: Draft 0.1
Standard: AI-First Community Standard
Short name: AIFC
Builds on:

Purpose of this document: Define the difference and relationship between human skills and AI skills. Explain why critical know-how must remain human-readable, why AI skills must not be the only carrier of community capability, and how skills should work together, be managed, versioned, reviewed, exported, and used in the Human Cockpit Layer.

1. Purpose of this document

This document defines Human Skills and AI Skills.

An AIFC community uses AI to accelerate work, but does not want to lose human capability.

It must therefore distinguish two types of skills:

Human skills
AI skills

A human skill helps a person understand, perform, review, transfer, and restore work.

An AI skill helps an AI agent perform work safely, consistently, and in alignment with community rules.

Both skill types matter.

But they are not interchangeable.

An AI skill without a human skill can create dependency. A human skill without an AI skill can fail to use AI’s acceleration potential. Together, they enable AI-first, human-managed operations.

2. Core principle

The core principle of this document is:

Critical knowledge must be understandable by humans and usable by AI.

AIFC says:

Do not let AI skills become the only place where community capability lives.

AI skills can be very powerful.

But if people do not understand what an agent does, the community has not gained capability. It has gained dependency.

3. Definition of human skill

Human skill is a human-readable artefact that describes a capability so a person can understand, perform, review, transfer, or restore it.

A human skill may include:

A human skill supports:

Minimum requirement

Critical community capabilities must have a human skill or corresponding human-readable knowledge artefact.

4. Definition of AI skill

AI skill is a structured artefact that defines how an AI agent or AI workflow should perform a certain type of work.

An AI skill may include:

An AI skill supports:

Minimum requirement

An AI agent or AI workflow with significant impact must have a defined AI skill or equivalent agent instructions.

5. Human skill vs AI skill

Human skills and AI skills have different primary readers.

Human skill:
written for humans

AI skill:
written for AI agents and AI workflows

But both must be understandable by humans.

An AI skill may have a more machine-precise structure, but a person must be able to understand its purpose, risk, and impact.

Comparison

Human skill answers:
How can a human understand and perform this capability?

AI skill answers:
How should an AI agent perform this capability safely and consistently?

Minimum requirement

A critical AI skill must be human-readable enough for its owner to review it.

6. Why both are needed

AIFC needs both types of skills.

Human skill without AI skill

When a human skill exists but no AI skill exists:

AI skill without human skill

When an AI skill exists but no human skill exists:

Human skill + AI skill

When both exist:

Minimum requirement

Critical AI-assisted capabilities must have both human-readable explanation and AI-facing instructions.

7. Skill pairing

AIFC recommends pairing human skills and AI skills.

A skill pair is:

human skill
+
AI skill

for the same or related capability.

Example:

Human skill:
How to review an AI-generated Jira ticket

AI skill:
How to draft a Jira ticket from a change proposal

Example:

Human skill:
How to design a clear dashboard for management attention

AI skill:
How to generate dashboard UX review suggestions

A skill pair helps maintain the balance between human understanding and AI acceleration.

Minimum requirement

Critical AI skills must have an explicit link to a human skill or human-readable knowledge.

8. Human skill as capability anchor

A human skill is the anchor of community capability.

It ensures that capability is not locked into:

A human skill does not mean a person will always do the work manually.

It means a person or community understands the capability enough to:

Minimum requirement

A critical capability must not be anchored only in an AI skill.

9. AI skill as acceleration layer

An AI skill is the acceleration layer over community capability.

It helps AI work:

An AI skill should build on human-readable know-how.

It should not replace it.

Minimum requirement

An AI skill must reference relevant Source of Truth artefacts when it performs critical work.

10. Skill hierarchy

AIFC recommends distinguishing skill layers.

Community principles
-> Human skills
-> AI skills
-> Workflow-specific instructions
-> Runtime prompts / agent execution

Community principles say why and according to which values the work is done.

Human skills say how people understand and perform the capability.

AI skills say how AI performs it.

Workflow-specific instructions say how the skill is used in a specific workflow.

Runtime prompts are the concrete execution.

Minimum requirement

Runtime prompts must not be the only layer of a critical skill.

11. Skill granularity

Skills should be specific enough, but not too fragmented.

Too general:

Write good documentation.

This is not very usable.

Too small:

Rewrite the third sentence in section 2 of document X.

This is not reusable.

Better:

Write scannable AI-first Markdown documentation with clear purpose, status, owner, rules, examples and next actions.

Minimum requirement

Skills must be designed for repeated use in a reasonably bounded context.

12. Skill ownership

Every critical skill must have an owner.

The owner is responsible for:

The owner is not necessarily the author.

The owner is responsible for the skill being usable and safe.

Minimum requirement

Critical human skills and AI skills must have an owner.

13. Skill status

Skills must have lifecycle status.

Recommended statuses:

draft
proposed
under_review
active
deprecated
retired
archived
rejected

AI agents must not use a deprecated or rejected skill as a current rule.

People must be able to see what is valid.

Minimum requirement

Critical skills must have status.

14. Skill versioning

Skills evolve.

They must therefore have a version or change history.

Versioning helps:

Minimum requirement

Critical AI skills must be versioned or auditable for change.

15. Skill review

Skills may degrade.

Review should verify:

Minimum requirement

Critical skills must have a review cycle.

16. Skill examples

Examples help people and AI.

A good example shows:

Examples are especially important for AI skills because they reduce ambiguity.

Minimum requirement

Critical or ambiguous skills must contain examples.

17. Skill anti-patterns

Anti-patterns show what should not be done.

They are important for:

Example:

Anti-pattern:
AI rewrites an approved decision record as if it were a draft proposal.

Example:

Anti-pattern:
Human reviewer approves an AI output without understanding the assumptions.

Minimum requirement

Skills with significant misuse risk must contain anti-patterns.

18. Human skill content structure

AIFC recommends this structure for a human skill:

Title
Purpose
When to use
When not to use
Inputs
Steps
Quality criteria
Examples
Anti-patterns
AI assistance
Human review checklist
Fallback
Related workflows
Related AI skills
Owner
Status
Review cycle

Not every skill must have every section.

Critical skills should have at least purpose, steps, quality criteria, examples, anti-patterns, owner, and review.

Minimum requirement

A critical human skill must enable a person to understand and review the work.

19. AI skill content structure

AIFC recommends this structure for an AI skill:

Title
Agent role
Purpose
Scope
Allowed inputs
Forbidden inputs
Allowed actions
Forbidden actions
Output format
Quality criteria
Examples
Anti-patterns
Uncertainty handling
Approval rules
Write-back rules
AI-NDA Boundary reference
Escalation rules
Related human skill
Owner
Status
Review cycle

An AI skill must be more explicit than a human skill in boundaries, actions, and outputs.

Minimum requirement

A critical AI skill must define allowed actions, forbidden actions, output format, and approval rules.

20. Hybrid skills

Some skills are hybrid.

A hybrid skill contains human and AI parts in one artefact.

This can be suitable for smaller communities or tightly connected workflows.

Example:

Skill:
Create and review a change proposal with AI assistance

Human part:
How to judge whether proposal makes sense

AI part:
How AI drafts proposal from observed signal

A hybrid skill can be useful if it remains clear.

Minimum requirement

A hybrid skill must clearly separate human responsibilities from AI responsibilities.

21. Skill and decision boundaries

Skills must not blur the boundary between proposal and decision.

An AI skill may say:

AI may propose.
AI must not approve.

A human skill may say:

Human owner reviews, approves or rejects the proposal.

This must be explicit especially for:

Minimum requirement

Skills that influence decision-making must clearly distinguish proposal, recommendation, decision, and approved change.

22. Skill and AI-NDA Boundary

An AI skill must respect the AI-NDA Boundary.

It must state:

Minimum requirement

An AI skill working with non-public data must reference the AI-NDA Boundary or data classification.

23. Skill and Operational DNA

Skills may be part of Operational DNA.

Critical skills often describe how the community actually functions.

This means they must be protected.

AI skills over Operational DNA are especially sensitive because they may allow agents to reproduce or change the community’s operating capability.

Minimum requirement

Skills containing Operational DNA must have appropriate sensitivity classification and access control.

24. Skill and Source of Truth

Skills must be stored in the Source of Truth or managed skill repository.

A skill stored only in:

is not reliable enough for critical work.

Minimum requirement

Critical skills must be discoverable outside runtime AI interaction.

25. Skill and Human Cockpit Layer

The Human Cockpit Layer should make skills available in the context of work.

For example:

Skills should not be only documentation deep in a repository.

They should be working interfaces.

Minimum requirement

Critical skills must be available to people at the moment they need them.

26. Skill and onboarding

Human skills are the foundation of onboarding.

They help new members understand:

AI can support onboarding.

But onboarding must not exist only as a chat with AI.

Minimum requirement

Critical onboarding must not depend only on an AI agent without a human-readable skill path.

27. Skill and fallback

Human skills support fallback.

A fallback skill says:

An AI skill may include when to switch to fallback.

Minimum requirement

Critical AI-assisted workflows must have a human fallback skill or documented fallback procedure.

28. Skill and review

Human review is itself a skill.

A critical AI output requires a reviewer skill.

The reviewer must know:

Minimum requirement

Critical AI workflows must have a human review skill or review checklist.

29. Skill and portability

Skills support exit strategy.

If skills are human-readable and exportable, the community can:

If skills are locked in a proprietary tool, skill lock-in emerges.

Minimum requirement

Critical skills must be portable or exportable.

30. Skill and metrics

A skill can be measured.

Metrics include:

Minimum requirement

Significant skill changes must have expected impact or a verification method.

31. Skill repository

An AIFC community may have a skill repository.

It may contain:

/skills
  /human
  /ai
  /hybrid
  /review
  /fallback
  /deprecated

A skill repository must support:

Minimum requirement

If the community uses multiple critical skills, it must have a managed place where they are recorded.

32. Suggested metadata

Example metadata for a human skill:

human_skill:
  id:
  title:
  status: draft | proposed | active | deprecated | retired | archived
  owner:
  purpose:
  related_values:
  related_workflows:
  related_ai_skills:
  criticality: low | medium | high | critical
  sensitivity: public | internal | restricted | operational_dna
  human_capability_support: true | false
  fallback_relevant: true | false
  review_cycle:
  last_reviewed:
  version:

Example metadata for an AI skill:

ai_skill:
  id:
  title:
  status: draft | proposed | active | deprecated | retired | archived
  owner:
  agent_role:
  purpose:
  scope:
  related_human_skill:
  related_workflows:
  related_ai_team_members:
  allowed_inputs:
  forbidden_inputs:
  allowed_actions:
  forbidden_actions:
  output_format:
  approval_rules:
  ai_nda_boundary:
  sensitivity: public | internal | restricted | operational_dna
  autonomy_impact: low | medium | high | critical
  human_capability_risk: low | medium | high | critical
  review_cycle:
  last_reviewed:
  version:

Example metadata for a skill pair:

skill_pair:
  id:
  title:
  human_skill:
  ai_skill:
  status: active | incomplete | under_review | deprecated
  owner:
  related_workflow:
  criticality:
  human_capability_coverage: complete | partial | missing
  ai_acceleration_coverage: complete | partial | missing
  last_reviewed:

These structures are illustrative.

The final schema should be defined in the agent-actionable layer of the standard.

33. Anti-patterns

AIFC rejects the following anti-patterns.

33.1 AI skill as only skill

An agent has instructions, but people do not have a corresponding human skill.

33.2 Human skill without AI boundary

A human skill recommends AI use, but does not state the boundaries.

33.3 AI skill without forbidden actions

An agent knows what to do, but not what it must not do.

33.4 AI skill without approval rules

An agent creates proposals or changes without clear approval.

33.5 Skill hidden in prompt

A critical skill exists only in a prompt or chat.

33.6 Skill hidden in agent memory

An agent remembers know-how that is not in the Source of Truth.

33.7 Skill without owner

The skill has no responsible role.

33.8 Skill without review

The skill becomes stale but is still used.

33.9 Skill too abstract

The skill is so general that it does not help the work.

33.10 Skill too tool-specific

The skill is so tied to one tool that it cannot be transferred.

33.11 Skill without examples

Neither people nor AI see a concrete good pattern.

33.12 Skill without fallback

An AI-assisted skill has no non-AI variant for critical work.

34. Minimal requirements

In the area of Human Skills and AI Skills, an AIFC community must at minimum:

  1. Distinguish human skills from AI skills.
  2. Ensure critical capabilities have a human skill or human-readable knowledge artefact.
  3. Ensure an AI agent or AI workflow with significant impact has an AI skill or equivalent instructions.
  4. Ensure critical AI skills are human-readable enough for review.
  5. Link critical AI skills to a human skill or human-readable knowledge.
  6. Ensure critical capability is not anchored only in an AI skill.
  7. Ensure AI skills reference relevant Source of Truth artefacts for critical work.
  8. Ensure runtime prompts are not the only layer of a critical skill.
  9. Design skills for repeated use in a reasonably bounded context.
  10. Assign owners to critical skills.
  11. Assign status to critical skills.
  12. Version or audit critical AI skills for change.
  13. Give critical skills a review cycle.
  14. Include examples in critical or ambiguous skills.
  15. Include anti-patterns in skills with significant risk.
  16. Ensure critical human skills enable people to understand and review the work.
  17. Ensure critical AI skills define allowed actions, forbidden actions, output format, and approval rules.
  18. Clearly separate human responsibilities and AI responsibilities in hybrid skills.
  19. Ensure decision-influencing skills distinguish proposal, recommendation, decision, and approved change.
  20. Ensure AI skills working with non-public data reference the AI-NDA Boundary or data classification.
  21. Apply appropriate sensitivity classification and access control to skills containing Operational DNA.
  22. Make critical skills discoverable outside runtime AI interaction.
  23. Make critical skills available to people in the context of work.
  24. Ensure critical onboarding does not depend only on an AI agent.
  25. Ensure critical AI-assisted workflows have a human fallback skill or fallback procedure.
  26. Ensure critical AI workflows have a human review skill or review checklist.
  27. Ensure critical skills are portable or exportable.
  28. Give significant skill changes expected impact or a verification method.
  29. Provide a managed place for recording skills when the community uses multiple critical skills.

35. Summary

Human skills and AI skills are two connected layers of community capability.

A human skill carries understanding, responsibility, review, onboarding, fallback, and resilience.

An AI skill carries repeatable agentic behavior, output format, boundaries, allowed actions, forbidden actions, and acceleration.

AIFC therefore says:

Human skills anchor capability.
AI skills accelerate capability.
Both must remain connected.

A community with only human skills may not use AI’s potential.

A community with only AI skills may lose the ability to understand itself.

An AI-first, human-managed community needs both.

Human Skills and AI Skills turn knowledge into shared human and agent capability.