Skip to content
Adaptive Skills Docs
Esc
navigateopen⌘Jpreview
On this page

Skill Catalog Governance

Reference documentation for Skill Catalog Governance in Adaptive Skills.

Purpose

Define how Adaptive Skills reviews new skills, optional modules, merges, splits and deprecations without turning the catalog into a runtime, policy engine or AletheIA replacement.

This document operationalizes the Lean Skill Doctrine.

Scope

Applies to:

  • new generic skills;
  • optional modules in existing skills;
  • sidecars such as templates, examples, checklists and references;
  • merge, split and deprecation proposals;
  • catalog category changes;
  • S22 lean-implementation review before the skill is created.

Non-goals

  • No runtime policy engine.
  • No automatic skill routing.
  • No blocking mechanism inside SKILL.md.
  • No rewrite of existing skills in S20.
  • No skill-catalog-review skill in this phase.
  • No provider-specific rules.
  • No AletheIA governance embedded into Adaptive Skills methods.

Lifecycle

candidate -> draft -> published -> observed -> reinforced | revise | merge | deprecate
State Meaning Required evidence
candidate A possible skill/module/change exists as an issue, pack, observation or proposal source ref and problem statement
draft A bounded PR proposes the change quality gate and validation plan
published The change has passed repository validation and review merge evidence
observed Real or synthetic use has produced a source-backed observation observation or validation case
reinforced Evidence supports keeping the current design review record
revise Evidence supports a bounded adjustment proposal and review
merge Two surfaces overlap enough to combine overlap evidence and migration note
deprecate A surface should stop being recommended deprecation rationale and alternative

New skill intake

A new skill proposal must include:

  • source refs;
  • dominant task shape;
  • adjacent skills reviewed;
  • reason an optional module is insufficient;
  • expected output and verification;
  • boundary statement showing it declares rather than enforces;
  • completed Skill Quality Gate.

Module proposal flow

Use an optional module when an existing skill owns the dominant task shape and the new behavior is triggered only by context.

A module proposal must declare:

  • activation trigger;
  • added output/evidence;
  • when to skip;
  • why it does not deserve standalone skill status.

Merge, split and deprecation

Merge when two skills repeatedly compete for the same dominant need and a single boundary would be clearer.

Split when one skill has two recurring dominant needs that create different outputs, verification or handoff signals.

Deprecate when a skill is no longer useful, overlaps a stronger surface, or carries unsafe/unclear boundaries that cannot be repaired proportionally.

Protected surfaces

These surfaces require explicit review before change:

  • name and description;
  • When to Use / When NOT to Use;
  • Core Moves;
  • category and domain boundaries;
  • merge, split or deprecation decisions;
  • harness requirements and governance boundary statements.

Proposal-safe surfaces

These surfaces can evolve through smaller proposals when evidence supports the change:

  • Activation Triggers;
  • Optional Modules;
  • Verification;
  • Anti-patterns;
  • templates, examples, checklists and references.

Decision outcomes

Quality-gate decisions use this vocabulary:

accept_as_new_skill
convert_to_optional_module
merge_into_existing_skill
split_existing_skill
deprecate
defer_until_recurrence_is_proven
reject

The decision must name evidence refs and why lighter options were not enough.

Relationship to AletheIA

Adaptive Skills owns catalog doctrine and skill content. AletheIA owns macro-governance over how a consumer task selects, observes, rejects, overrides or proposes skill use.

The matching AletheIA contract is expected to govern:

  • skill selection decisions;
  • skill proposal decisions;
  • over-selection and skill-sprawl signals;
  • narrow blocking posture for authority/safety violations;
  • observation records with provenance.

Adaptive Skills remains useful without AletheIA. The catalog governance here is repository-native; it does not require a runtime or external orchestrator.

Required validation

For S20 docs-only changes:

  • python3 scripts/validate_system_state.py
  • python3 scripts/validate_capabilities.py
  • python3 scripts/validate_evolution.py
  • python3 scripts/validate_skills.py
  • git diff --check

For future skill creation, also run projection status and any skill-specific validation required by the changed files.

Was this page helpful?