Skill Knowledge Boundaries

Reference documentation for Skill Knowledge Boundaries in Adaptive Skills.

Goal

State the boundary between skill and knowledge so that skills remain reusable, portable, and safe to ship.

A skill is a procedure: how to think, what to ask, what to produce. A knowledge pack is content: a framework, a policy, a persona, a guideline.

The two must not be merged.

What a skill must carry

the situation it applies to
the questions it asks
the moves it executes
the criteria it uses
the format of its output
the verification it runs
the handoff it leaves behind

All of these can be written without referencing any proprietary content.

What a skill must not carry

the body of a proprietary framework
the text of an internal policy
excerpts from confidential or regulated sources
client, customer, or partner names
examples drawn from real, identifiable engagements
detailed scoring tables that encode an organization’s specific carve-outs

If a skill needs that content, it declares a knowledge dependency (see declaring-knowledge-dependencies.md) and lets AletheIA’s Knowledge Governance Layer resolve it.

Two modes every knowledge-aware skill supports

Generic mode. The skill runs with general criteria when no authorized source is available. The output is marked as generic.
Knowledge-aware mode. The skill runs with sources resolved by the registry. The output cites pack ids and versions and applies any active restrictions.

A skill that cannot run in generic mode must say so loudly in its fallback_behavior. Silent dependence on a specific source is an anti-pattern.

Why the boundary matters

Portability. A skill that embeds one company’s framework cannot move to another.
Safety. A skill that ships with internal content leaks that content into every agent that loads it.
Auditability. Embedded content cannot be versioned independently of the skill.
Governance. A skill cannot be the place where source precedence and sensitivity are decided.

Quick checklist when authoring a skill

Does the skill contain any verbatim text from a non-public source? If yes, remove it.
Does the skill name a specific framework as required? If yes, replace with a knowledge dependency on a type.
Are there examples that include real identifiers? If yes, generalize them.
Does the skill state how it behaves when its knowledge dependencies are not met? If no, add fallback_behavior.
Does the skill require human review for any specific outcomes? If yes, declare them.