System: MWMS
Document Type: Framework
Authority Level: MCR Source Of Truth
Status: Draft For MCR
Version: v1.3
Primary Location: MCR
Future Operational Destination: HeadOffice Brain, MWMS Brain, Brain Room, AI Manager, AI Employee Router, Task Executor Systems, Course Absorption System, Newsletter Intelligence, Opportunity System, Automation Brain, AIBS Brain
Parent Page: HeadOffice
Owner: Martyn
Developer Boundary: Do Not Touch M’s Active Build Areas Unless Specifically Assigned
Source Of Truth: MCR
Last Reviewed: 2026-06-28
Source / Origin: MWMS AI Agent Skill Library Framework v1.2 + AI Automations by Jack Claude Skills 2.0 block covering skill creation, trigger visibility, capability uplift, encoded preferences, baseline comparison, shared skill use, composable skills, model-upgrade reassessment, performance validation and skill retirement
MWMS Classification: AI Skill Governance Framework / Procedural Memory Framework / AI Employee Training Standard / Reusable Workflow Library / Organisational Capability System / Skill Effectiveness And Retirement Framework
Primary Brain: HeadOffice Brain
Supporting Brains: MWMS Brain, Automation Brain, AIBS Brain, Operations Brain, Data Brain, Risk Brain, Compliance Brain, SIT Brain
Related Pages: MWMS AI Agent Operations Core, MWMS AI Employee Role Card Standard, MWMS AI Employee Capability Stack Framework, MWMS AI Tool Permission And Access Framework, MWMS AI Agent Memory And Context Framework, MWMS Agentic Work Unit Standard, MWMS AI Workflow Pipeline Standard, MWMS AI Output Validation Standard, MWMS Messy Input Normalization Framework, MWMS Agentic Reporting Standard, MWMS AI Employee Handoff Protocol, MWMS AI Agent Failure Handling And Escalation Protocol, MWMS AI Agent Outcome Measurement Framework, MWMS Independent Model Review And Rescue Routing Framework, MWMS External Knowledge Engine And Reasoning Agent Separation Framework, MWMS AI Work Session Closure And Knowledge Commitment Protocol, MWMS AI Agent Deployment Readiness Checklist, MWMS AI Workforce Governance Model, MWMS AI Skill Builder And Audit Protocol, MWMS AI Skill Installation And Usage Protocol, MWMS Prompt Architecture And Automation Output Reliability Framework
Purpose
The purpose of this document is to define the MWMS AI Agent Skill Library Framework.
This framework explains how MWMS stores, governs, discovers, activates, reuses, tests, compares, improves, shares, versions, audits, deprecates and retires procedural skills for AI Employees.
MWMS must not rely on one-off prompts every time an AI Employee performs repeated work.
As the MWMS AI workforce grows, repeated tasks should become reusable skills.
A skill is not merely knowledge.
A skill is a repeatable, governed method for performing work.
A skill may contain:
goals
triggers
inputs
context requirements
procedures
tools
permissions
standards
output contracts
validation
handoff rules
failure rules
outcome requirements
version information
effectiveness evidence
retirement logic
The AI Agent Skill Library gives MWMS a way to convert repeated work into reusable procedural memory and governed organisational capability.
The framework must also distinguish between:
skills that temporarily improve a model’s weak capability
skills that permanently preserve the MWMS way of working
This distinction matters because some skills should evolve or retire as models improve, while others should remain because their purpose is organisational consistency rather than model uplift.
Scope
This framework applies to reusable AI Employee skills across MWMS.
This includes skills used by:
HeadOffice Brain
Brain Room
AI Manager
AI Employee Router
Task Executor systems
Dev Console
Newsletter Intelligence
Course Absorption
Opportunity System
Affiliate Brain
Research Brain
Experimentation Brain
Finance Brain
Content Brain
Ads Brain
Sales Brain
Conversion Brain
Customer Brain
Automation Brain
AIBS Brain
Operations Brain
Data Brain
Risk Brain
Compliance Brain
SIT Brain
This framework governs:
skill identity
skill purpose class
skill discovery
skill trigger
skill activation visibility
skill input
skill context
skill capability and tool route
skill procedure
skill output
skill validation
skill handoff
skill failure and rescue
skill outcome
skill testing
baseline comparison
model-upgrade reassessment
skill scope
skill portability
skill sharing
skill composition
skill versioning
skill effectiveness
skill drift
skill improvement
skill deprecation
skill retirement
skill registry records
This framework does not authorise technical implementation by itself.
Core Doctrine
MWMS should not ask an AI Employee to improvise a repeated business process from scratch every time.
Repeated, validated work should become a governed skill.
A valid skill must be:
discoverable
appropriately triggered
within scope
context-aware
tool-aware
permission-aware
procedural
testable
comparable
observable
versioned
recoverable
outcome-linked
retirable
Skills should improve reliability without creating hidden black-box behaviour.
The AI Employee and supervising system should be able to determine:
which skill was selected
why it was selected
which version ran
what dependencies were loaded
which permissions applied
whether it performed better than the approved baseline
whether fallback was used
what outcome resulted
Definition
An MWMS AI Agent Skill is a governed procedural capability that enables an authorised AI Employee to perform a defined category of work consistently within approved context, tool, permission, validation, handoff and outcome boundaries.
A skill is not:
a vague prompt
a collection of unrelated instructions
a tool connection by itself
a model capability by itself
an undocumented habit
an uncontrolled automation
A prompt may be part of a skill.
A prompt alone does not automatically qualify as a skill.
Skill Purpose Classes
Every formal skill must be classified by purpose.
The two primary purpose classes are:
Capability Uplift Skill
Encoded Preference Skill
A skill may also be classified as Hybrid where both purposes are material.
Capability Uplift Skill
A Capability Uplift Skill exists because the approved base model or tool route cannot currently perform the required work to the necessary standard without additional procedure, assets, examples, orchestration or specialist logic.
Examples may include:
document generation
specialist formatting
PDF completion
visual design
structured extraction
technical execution
complex tool orchestration
specialist report generation
specialist research synthesis
A Capability Uplift Skill should be retained only while it creates a meaningful advantage over the approved unassisted baseline.
The advantage may appear in:
quality
consistency
accuracy
safety
speed
cost
format compliance
tool execution
business outcome
Capability Uplift Skills may have a natural retirement date.
Encoded Preference Skill
An Encoded Preference Skill exists to preserve MWMS-specific:
procedures
governance
standards
voice
decision rules
approval rules
handoff rules
scope boundaries
operating preferences
client-specific methods
Examples may include:
Course Absorption Skill
MCR Duplicate Risk Check Skill
Developer Handoff Skill
Client Onboarding Skill
Offer Review Skill
Full Page Output Creation Skill
Report Structure Skill
Brand Voice Skill
Session Closure Skill
Encoded Preference Skills should not be retired merely because a model becomes more capable.
Their value is preserving the approved MWMS way of working.
They should still be tested for:
correct execution
relevance
drift
efficiency
safety
continued organisational value
Hybrid Skill
A Hybrid Skill combines:
model capability uplift
MWMS-specific procedure
Example:
A client report skill may improve structured report generation while also enforcing MWMS evidence, validation, formatting and approval rules.
Hybrid Skills should be assessed separately for:
capability advantage
preference preservation
A model upgrade may remove the uplift need while the preference layer remains valuable.
In that case, simplify rather than retire the full skill.
Skill Purpose Classification Rule
Every formal skill must record:
primary purpose class
secondary purpose class where relevant
why the classification applies
whether baseline comparison is mandatory
whether model-upgrade reassessment applies
whether organisational preference preservation applies
No formal skill should remain unclassified.
Skill Library Layers
The MWMS AI Agent Skill Library is organised into fifteen layers:
- Skill Identity
- Skill Purpose Class
- Skill Trigger
- Skill Activation Visibility
- Skill Input
- Skill Context
- Skill Capability And Tool Route
- Skill Procedure
- Skill Output
- Skill Validation
- Skill Handoff
- Skill Failure And Rescue
- Skill Outcome
- Skill Testing And Baseline Comparison
- Skill Improvement, Reassessment And Retirement
Skill Identity
Each skill must have a clear identity.
Required identity fields include:
Skill ID
Skill Name
Skill Type
Skill Purpose Class
Owning Brain
Supporting Brains
Assigned AI Employee
Skill Purpose
Business Outcome
Scope
Status
Version
Source Or Origin
No skill should rely only on a folder name or informal label.
Skill Trigger
A skill must define when it should activate.
Possible trigger signals include:
task type
user language
input format
workflow stage
Owning Brain
Work Unit metadata
required output
known failure state
explicit skill request
scheduled event
monitoring event
client state
approval state
A skill should also define when it must not be used.
Skill Trigger Rule
A skill should activate because the task requires it, not because the skill exists.
Skill Activation Visibility
A mature skill system must make activation observable.
The execution record should show:
skill selected
reason selected
trigger matched
trigger source
skill version
purpose class
assigned AI Employee
required dependencies
dependencies loaded
required model
model used
required tools
tools available
permission state
execution started
execution completed
validation result
fallback used
rescue route used
final outcome
Where the interface supports it, skill activation should be visibly indicated to the operator.
The system must not create a black box where nobody can determine whether the intended skill ran.
Skill Activation Evidence Rule
A skill is not considered activated merely because the final output resembles the expected result.
Activation should be supported by execution evidence.
Minimum evidence:
Skill ID
Skill version
trigger reason
execution timestamp
execution state
validation state
Where activation logging is unavailable, the result must be classified as unverified skill use.
Skill Input
A skill must define what input it can process.
Input may include:
course transcript
course PDF
newsletter
Brain Room message
offer page
research source
finance data
experiment result
screenshot
WordPress page list
Supabase record
developer file
client document
user instruction
monitoring event
external knowledge retrieval
The skill must define:
required input
optional input
incomplete-input handling
messy-input handling
source-authority requirement
source-freshness requirement
provenance requirement
client or data boundary
Skill Input Rule
A skill should not operate on input it is not designed to handle.
Skill Context
A skill must define the context needed to perform correctly.
Required context may include:
current MCR pages
active Canon
Work Unit
current save point
user instructions
client context
role boundaries
current workflow stage
known failure history
previous decisions
relevant source material
current system state
The skill should distinguish:
required context
helpful context
forbidden context
stale context
client-specific context
Skill Context Rule
A skill should not assume context that was not supplied or retrieved from an approved source.
Skill Capability And Tool Route
A skill must define the capability and tools required.
Possible capability requirements:
standard reasoning
advanced reasoning
long context
multimodal review
coding
research
deterministic validation
local private model
low-cost extraction
independent review
rescue model
Possible tool requirements:
uploaded files
MCR search
web research
database read
controlled database write
WordPress read
controlled WordPress write
Gmail read
Gmail draft
approved email send
calendar read
calendar write
code execution
image generation
document generation
browser automation
MCP connection
API connection
Required model or tool use must remain within approved permission boundaries.
Skill Procedure
A skill must define the ordered method used to perform the work.
The procedure should include:
preconditions
sequence
decision points
tool calls
validation points
handoffs
stopping conditions
fallback conditions
forbidden shortcuts
A skill should be detailed enough to create repeatability but not so tool-specific that portability is unnecessarily lost.
Skill Output
A skill must define:
required output
optional output
output format
schema
decision state
status
evidence
limitations
next action
A skill output should be usable by the next person, Brain, system or process.
Skill Validation
A skill must define:
validation requirement
validation owner
independent review requirement
human review requirement
acceptance requirement
A skill is not complete merely because it produced an output.
The output must meet the approved quality, authority and safety requirements.
Skill Handoff
A skill must define:
handoff destination
handoff format
required metadata
acceptance owner
blocked-state handling
rejection handling
A handoff should not force the receiving Brain or person to reconstruct missing context.
Skill Failure And Rescue
A skill must define:
failure triggers
retry limit
rescue threshold
rescue route
containment action
escalation owner
fallback output
Failure may include:
missing input
wrong context
tool failure
permission denial
invalid output
validation failure
model failure
dependency failure
cost threshold breach
time threshold breach
conflicting instructions
A skill should fail safely.
Skill Outcome
A skill must define:
expected outcome
business outcome
outcome evidence
measurement method
knowledge commitment requirement
logging requirement
A skill should not be judged only by whether it executed.
It should be judged by whether it created useful value.
Skill Types
MWMS should classify skills by operational type.
Intake Skills
Used when information first enters MWMS.
Examples:
Messy Input Intake Skill
Source Completeness Check Skill
Input Classification Skill
Brain Ownership Detection Skill
Purpose:
To make sure raw input becomes usable input.
Extraction Skills
Used to pull useful material from source content.
Examples:
Course Framework Extraction Skill
Newsletter Signal Extraction Skill
Offer Claim Extraction Skill
Research Evidence Extraction Skill
Developer Issue Extraction Skill
Purpose:
To separate useful signal from noise.
Evaluation Skills
Used to judge quality, risk, suitability or value.
Examples:
Course Absorption Value Skill
Offer Test Suitability Skill
Finance Risk Review Skill
Experiment Signal Quality Skill
Dashboard Readiness Skill
Purpose:
To determine whether something should move forward.
Creation Skills
Used to create structured outputs.
Examples:
Full Page Output Creation Skill
Agentic Work Unit Creation Skill
Developer Brief Creation Skill
Handoff Package Creation Skill
Client Report Drafting Skill
Purpose:
To convert source material into usable work products.
Validation Skills
Used to check outputs before use.
Examples:
MCR Page Validation Skill
Developer Instruction Validation Skill
Newsletter Signal Validation Skill
Offer Evaluation Validation Skill
Client Report Validation Skill
Purpose:
To stop weak or unsafe outputs before operational use.
Routing Skills
Used to send work to the correct Brain, workflow, person or queue.
Examples:
Brain Routing Skill
Research Handoff Skill
Finance Review Routing Skill
M Developer Handoff Skill
Parking System Routing Skill
Purpose:
To prevent lost or misrouted work.
Tool Use Skills
Used when an AI Employee operates with a tool.
Examples:
Gmail Newsletter Read Skill
Supabase Internal Row Write Skill
WordPress Page List Review Skill
File Review Skill
Dashboard Record Preparation Skill
Purpose:
To ensure tools are used safely within approved limits.
Reporting Skills
Used to create decision-ready reports.
Examples:
Course Absorption Report Skill
Newsletter Intelligence Report Skill
Offer Evaluation Report Skill
Developer Support Report Skill
AIBS Client Report Skill
Purpose:
To make reports useful rather than passive.
Failure Handling Skills
Used when something goes wrong.
Examples:
Failed Output Classification Skill
Escalation Decision Skill
Failure Log Creation Skill
Containment Action Skill
Kaizen Lesson Capture Skill
Purpose:
To convert failure into controlled recovery and learning.
Rescue Skills
Used after the normal route reaches its failure threshold.
Examples:
Independent Failure Diagnosis Skill
Alternative Model Rescue Skill
Deterministic Verification Skill
Reduced-Scope Recovery Skill
Purpose:
To recover stalled work using a materially different route.
Outcome Measurement Skills
Used to judge whether work mattered.
Examples:
Outcome Scoring Skill
Risk Reduction Capture Skill
Business Value Summary Skill
AI Employee Usefulness Review Skill
Workflow Value Review Skill
Purpose:
To make MWMS outcome-driven.
Persistent Agent Skills
Used by scheduled or long-running AI Employees.
Examples:
Persistent Monitoring Skill
Scheduled Report Skill
Alert Qualification Skill
Shutdown And Escalation Skill
Purpose:
To govern recurring work, retry limits, cost, alerts and shutdown.
External Knowledge Skills
Used to retrieve and prepare evidence.
Examples:
Source Collection Skill
Provenance Preservation Skill
Client-Filtered Retrieval Skill
Conflicting Evidence Identification Skill
Purpose:
To supply traceable evidence without confusing retrieval with final reasoning.
Knowledge Commitment Skills
Used to preserve validated learning.
Examples:
MCR Update Skill
Decision Record Skill
Failure Learning Commitment Skill
Session Save Point Skill
Purpose:
To ensure approved learning reaches the correct durable destination.
Session Closure Skills
Used to close work cleanly.
Examples:
Work Session Wrap-Up Skill
Current Save Point Skill
Exact Next Action Skill
Open Issues Capture Skill
Purpose:
To preserve continuity between sessions and interfaces.
AI Agent Skill Record Template
Each skill should be recorded using the following structure.
Skill ID:
Skill Name:
Skill Type:
Skill Purpose Class:
Secondary Purpose Class:
Purpose Classification Reason:
Owning Brain:
Supporting Brains:
Assigned AI Employee:
Skill Purpose:
Business Outcome:
When To Use This Skill:
When Not To Use This Skill:
Required Input:
Optional Input:
Required Context:
Source Authority Requirement:
Client Or Data Boundary:
Related Standards:
Required Model Or Capability:
Approved Baseline Model:
Required Tools:
Tool Permission Boundary:
Forbidden Actions:
Skill Procedure:
Required Output:
Output Format:
Validation Requirement:
Validation Owner:
Independent Review Requirement:
Human Review Requirement:
Handoff Destination:
Acceptance Requirement:
Failure Triggers:
Retry Limit:
Rescue Threshold:
Rescue Route:
Expected Outcome:
Outcome Evidence:
Knowledge Commitment Requirement:
Logging Requirement:
Activation Visibility Requirement:
Baseline Comparison Required:
Baseline Test Date:
Baseline Output Score:
Skill Assisted Output Score:
Performance Advantage:
Cost Comparison:
Speed Comparison:
Consistency Comparison:
Error Comparison:
Safety Comparison:
Last Baseline Comparison:
Model Upgrade Reassessment Required:
Reassessment Trigger:
Next Reassessment Date:
Activation Count:
Successful Activation Count:
False Activation Count:
Missed Activation Count:
Fallback Count:
Skill Status:
Skill Version:
Source Or Origin:
Last Tested:
Last Reviewed:
Review Owner:
Deprecated By:
Retirement Recommendation:
Retirement Reason:
Created At:
Updated At:
Quick Use Version
Skill ID:
Skill Name:
Skill Type:
Skill Purpose Class:
Owning Brain:
Assigned AI Employee:
Skill Purpose:
Business Outcome:
When To Use:
When Not To Use:
Required Input:
Required Context:
Source Authority Requirement:
Related Standards:
Required Model Or Capability:
Approved Baseline Model:
Required Tools:
Tool Permission Boundary:
Forbidden Actions:
Procedure:
Required Output:
Validation:
Handoff:
Failure Trigger:
Fallback:
Expected Outcome:
Activation Evidence:
Baseline Comparison Status:
Skill Status:
Skill Version:
Last Tested:
Last Reviewed:
Skill Discovery And Auto-Invocation
A mature skill system should help the assigned AI Employee recognise when an approved skill applies.
Skill discovery may use:
task type
user language
input format
workflow stage
Owning Brain
Work Unit metadata
required output
known failure state
explicit skill request
Auto-invocation does not mean uncontrolled execution.
The skill may activate only when:
the trigger is satisfied
the assigned AI Employee is authorised
required context is available
required tools are approved
the current version is active
no higher-authority rule blocks use
the task is within the skill boundary
Skill Discovery Rule
Skills should be easy to discover and difficult to misuse.
Automatic skill selection must never override Canon, Tool Permissions, human approval requirements or the Work Unit’s actual purpose.
Skill Auto-Invocation Visibility Rule
When a skill is selected automatically, the execution record should preserve:
why it matched
which competing skills were considered where relevant
why another skill was not selected
whether a manual override occurred
whether the operator was notified
Auto-invocation must remain reviewable.
Skill Scope And Storage Model
Skills may exist at different scopes:
Organisation-Level Skill
Used across MWMS.
Brain-Level Skill
Used inside one Brain.
Project-Level Skill
Used inside a defined project.
Client-Level Skill
Used for one client within isolated context.
Experimental Skill
Used only for controlled testing.
Scope must define:
who may use the skill
where it is stored
which context it may access
which permissions apply
whether it may be shared
whether it may be promoted
Scope Promotion Rule
A project, client or experimental skill may not become an organisation-level skill without review, evidence and approval.
Expert Workflow To Skill Conversion
A skill may be created from a proven expert workflow.
The conversion process should identify:
expert objective
trigger
inputs
context
decision points
tools
procedure
quality standard
failure conditions
output
handoff
business outcome
A skill must preserve the expert method without encoding unnecessary personal habit or undocumented assumption.
Skill Asset Bundle Standard
A skill may include an asset bundle.
Possible assets:
instruction file
examples
templates
schemas
prompt components
checklists
reference files
brand rules
test cases
validation rules
code
tool configuration
Asset bundles must be:
versioned
traceable
permissioned
compatible
replaceable
A missing or outdated asset must not remain invisible.
Skill Packaging Standard
A formal skill package should include:
identity
purpose class
scope
trigger
input contract
context contract
procedure
tool route
permission boundary
output contract
validation
handoff
failure and rescue
outcome
activation evidence
test evidence
baseline comparison where required
version
review date
retirement path
Skill Dependency Rule
A skill may depend on:
another skill
a context pack
a tool
a model
a database
an API
a schema
a validation process
a human approval
Dependencies must be:
identified
available
version-compatible
loaded in the correct order
Rule
A skill dependency should not be assumed merely because a related page exists.
Skill Composition Rule
Several skills may be combined inside one Work Unit.
Example:
Offer Evaluation may combine:
Offer Intake Skill
Vendor Claim Separation Skill
Research Evidence Skill
Compliance Risk Skill
Finance Review Skill
Outcome Reporting Skill
Skill composition must define:
sequence
ownership
handoff
validation gates
final authority
activation evidence for each component
Rule
Composed skills must not create conflicting instructions.
Skill Conflict Rule
Where two skills conflict:
Apply current Canon.
Apply the higher authority standard.
Apply the more specific approved skill.
Escalate unresolved conflicts.
Do not silently merge incompatible instructions.
Tool-Agnostic Skill Portability
Skills should preserve the business procedure separately from tool-specific implementation where practical.
A portable skill should distinguish:
durable purpose
durable decision logic
durable standards
replaceable model
replaceable tool
replaceable connector
replaceable file format
Tool-specific variants may exist, but the governing skill should not become obsolete merely because one vendor changes.
Skill Versioning Rule
Every formal skill should have:
version
date
owner
change summary
last tested date
last reviewed date
superseded version where applicable
A skill version should change when:
procedure changes
purpose class changes
validation changes
permissions change
output schema changes
failure threshold changes
model or tool dependency changes
baseline requirement changes
reassessment logic changes
client boundary changes
Rule
Skills should not change silently.
Skill Testing Rule
A skill should be tested before stronger operational use.
Testing should assess:
trigger accuracy
activation visibility
input handling
context sufficiency
procedure adherence
output quality
validation performance
permission compliance
handoff quality
failure handling
outcome value
baseline advantage where required
Possible test states:
Untested
Test Failed
Test Passed
Passed With Conditions
Proven Manual Use
Controlled Automation Candidate
Baseline Advantage Confirmed
Baseline Equivalent
Baseline Inferior
Model Upgrade Reassessment Required
Rule
A documented skill is not automatically a proven skill.
Baseline Comparison Testing
Capability Uplift Skills must be compared against an approved unassisted baseline.
Hybrid Skills should be compared where the uplift portion is material.
Encoded Preference Skills may use baseline comparison where useful, but the primary test is whether the skill preserves approved MWMS procedure.
Baseline comparison should assess:
quality
accuracy
consistency
format compliance
procedure compliance
speed
cost
error rate
permission compliance
safety
business usefulness
The comparison should use:
the same task
the same input
the same allowed context
the same evaluation criteria
representative examples
negative examples where relevant
The baseline record should identify:
baseline model
baseline version
skill version
test date
test owner
test cases
scoring method
result
limitations
Baseline Advantage Rule
A Capability Uplift Skill should remain active only when it produces a meaningful approved advantage over the baseline.
Possible outcomes:
Retain
Retain With Conditions
Simplify
Merge
Downgrade
Deprecate
Retire
Where the skill is baseline-equivalent but improves safety, governance, repeatability or auditability, it may remain with the reason recorded.
Where the skill is inferior, it should not remain active merely because time was invested in creating it.
Model Upgrade Reassessment
A Capability Uplift Skill must be reassessed when:
the approved base model changes materially
a new model becomes the default
a major tool capability changes
the original model weakness is resolved
the skill begins adding unnecessary latency
the skill begins adding unnecessary cost
the skill’s result quality declines
a native platform capability replaces the procedure
Hybrid Skills should be separated into:
uplift components
encoded preference components
The uplift component may be removed while the preference component remains.
Encoded Preference Skills should still be reviewed after model changes, but should not be retired solely because the model becomes more capable.
Model Upgrade Reassessment Outcomes
Retain Unchanged
Retain With Updated Baseline
Simplify
Remove Obsolete Steps
Replace Model Route
Merge With Another Skill
Convert To Encoded Preference Skill
Convert To Hybrid Skill
Deprecate
Retire
Every outcome should preserve:
reason
evidence
decision owner
effective date
replacement where relevant
Skill Effectiveness Record
Each formal skill should maintain effectiveness evidence.
Fields may include:
skill purpose class
baseline model
baseline test date
baseline score
skill-assisted score
performance advantage
cost difference
speed difference
consistency difference
error difference
safety difference
activation count
successful activation count
failed activation count
false activation count
missed activation count
fallback count
rescue count
human correction count
outcome success rate
last baseline comparison
next reassessment date
retirement recommendation
The record should support governance, not vanity metrics.
Skill Provenance Rule
Every formal skill should record where it came from.
Possible sources:
existing MWMS workflow
user-defined operating rule
course absorption
client workflow
developer procedure
external standard
experiment
failure lesson
model limitation
vendor capability
Provenance should not create authority by itself.
The skill must still pass MWMS review.
Skill Security Rule
A skill must not expand tool or data access merely because the procedure would be easier.
Security requirements may include:
least privilege
client isolation
secret protection
approval gates
write restrictions
external-send restrictions
logging
auditability
revocation
Skills must not embed exposed credentials.
Skill Statuses
Possible statuses:
Proposed
Draft
Experimental
Manual Use
Test Ready
Test Failed
Test Passed
Passed With Conditions
Controlled Automation Candidate
Active
Active With Monitoring
Reassessment Required
Deprecated
Retired
Blocked
A status must reflect actual evidence.
Skill Review Cycle
Review frequency should reflect:
risk
use frequency
model change rate
tool dependency
client exposure
failure history
commercial importance
Capability Uplift Skills may require more frequent reassessment where models change rapidly.
Encoded Preference Skills may require review after:
Canon changes
workflow changes
approval changes
organisational changes
client-specific changes
Skill Improvement And Kaizen
A skill should improve when:
repeated failures appear
human correction repeats
a stronger procedure is discovered
client requirements change
model capabilities change
tool capabilities change
better evidence becomes available
activation errors occur
baseline advantage weakens
Skills must evolve through Kaizen.
They must not remain static merely because they were documented.
Skill Drift Signals
Possible drift signals include:
false activation
missed activation
wrong scope
stale context
wrong tool route
outdated model requirement
permission mismatch
output inconsistency
increased rework
repeated human correction
baseline advantage loss
increased latency
increased cost
unexplained fallback use
obsolete assets
duplicate skill growth
Drift should trigger review.
Skill Retirement Rule
A skill should be retired when:
it no longer creates value
another skill replaces it
the supported workflow is removed
the procedure is unsafe
it depends on unavailable tools
it creates more rework than value
its task no longer repeats
the approved base model performs equally or better
the skill adds unnecessary latency
the skill adds unnecessary cost
the skill reduces quality
the skill creates inconsistency
the skill no longer improves safety or governance
a native capability replaces its uplift function
its organisational preference is no longer approved
Retirement should record:
retirement reason
evidence
replacement
affected AI Employees
affected Work Units
effective date
migration requirement
archive location
Retirement must not silently delete historical evidence.
Capability Uplift Retirement Rule
A Capability Uplift Skill should be considered for retirement when the approved baseline now matches or exceeds it on the required criteria.
Before retirement, confirm:
the same representative test cases were used
safety has not weakened
governance has not weakened
format compliance remains acceptable
business outcome remains acceptable
tool permissions remain controlled
Where only the uplift is obsolete, preserve any encoded preference layer that remains necessary.
Encoded Preference Preservation Rule
An Encoded Preference Skill should be retired only when:
the underlying MWMS process changes
the preference is no longer approved
the workflow is removed
the skill is duplicated by a stronger approved skill
the encoded rule conflicts with current Canon
Model capability improvement alone is not sufficient reason to retire it.
Skill Library Registry Fields
The Skill Library registry may include:
skill_id
skill_name
skill_type
skill_purpose_class
secondary_purpose_class
purpose_classification_reason
owning_brain
supporting_brains
assigned_ai_employee
skill_purpose
business_outcome
when_to_use
when_not_to_use
required_input
optional_input
required_context
source_authority_requirement
client_data_boundary
related_standards
required_model_capability
approved_baseline_model
required_tools
tool_permission_boundary
forbidden_actions
skill_procedure
required_output
output_format
validation_requirement
validation_owner
independent_review_requirement
human_review_requirement
handoff_destination
acceptance_requirement
failure_triggers
retry_limit
rescue_threshold
rescue_route
expected_outcome
outcome_evidence
knowledge_commitment_requirement
logging_requirement
activation_visibility_requirement
baseline_comparison_required
baseline_test_date
baseline_output_score
skill_assisted_output_score
performance_advantage
cost_comparison
speed_comparison
consistency_comparison
error_comparison
safety_comparison
last_baseline_comparison
model_upgrade_reassessment_required
reassessment_trigger
next_reassessment_date
activation_count
successful_activation_count
failed_activation_count
false_activation_count
missed_activation_count
fallback_count
rescue_count
human_correction_count
outcome_success_rate
skill_status
skill_version
source_origin
last_tested
last_reviewed
review_owner
deprecated_by
retirement_recommendation
retirement_reason
created_at
updated_at
No technical build is authorised by this framework alone.
Governance Role
HeadOffice owns the MWMS AI Agent Skill Library Framework.
HeadOffice is responsible for:
deciding when a skill becomes formal
preventing duplicate skills
ensuring each skill has one Owning Brain
ensuring skills align with Role Cards
ensuring skills align with Capability Stacks
ensuring skills respect Tool Permissions
ensuring skills include validation
ensuring skills include failure and rescue
ensuring skills define outcomes
ensuring skills have a purpose class
ensuring high-risk skills require human review
governing skill versioning
governing skill testing
governing baseline comparison
governing model-upgrade reassessment
reviewing failures
reviewing activation evidence
retiring weak or duplicate skills
protecting M’s active build
protecting future AIBS client systems
Individual Brains may propose and maintain skills for their own AI Employees.
HeadOffice governs:
cross-Brain skills
governance skills
high-risk skills
tool-enabled skills
client-facing skills
persistent-agent skills
automation-related skills
organisation-level skills
Relationship To SIT Brain
SIT Brain may:
verify Skill Record completeness
verify Skill purpose class
verify Skill status
detect missing validation
detect missing failure rules
verify reviewer independence
verify tool-permission alignment
detect unsafe skill execution
block untested high-risk skills
detect repeated skill failure
enforce rescue thresholds
detect obsolete versions
verify activation evidence
verify baseline comparison where required
detect false activation
detect missed activation
trigger model-upgrade reassessment
verify retirement controls
SIT Brain may not redefine the business procedure without authority.
Relationship To Data Brain
Data Brain supports:
Skill IDs
registry schema
activation logs
test records
baseline comparison records
version history
dependency records
failure records
outcome records
reassessment records
retirement records
Data Brain should preserve enough history to explain:
what ran
why it ran
what version ran
what model ran
whether it beat the baseline
what result occurred
why it changed
why it retired
Relationship To Automation Brain
Automation Brain may support:
skill execution
trigger detection
dependency loading
tool routing
logging
retry
fallback
scheduled reassessment
status updates
Automation Brain must not convert an unapproved skill into active automation.
Relationship To AIBS Brain
AIBS Brain may package approved skills into client systems.
Client skill packages must define:
client scope
data boundary
permissions
tool dependencies
model dependencies
baseline testing
support owner
versioning
update process
retirement path
A client-facing skill must not inherit unrestricted MWMS internal access.
Relationship To MWMS AI Skill Builder And Audit Protocol
The MWMS AI Skill Builder And Audit Protocol governs the controlled creation and audit process for individual skills.
This framework governs the library-wide structure and lifecycle.
Relationship To MWMS AI Skill Installation And Usage Protocol
The MWMS AI Skill Installation And Usage Protocol governs how approved skills are installed, activated and used in supported environments.
This framework governs whether the skill belongs in the library and remains fit for use.
Minimum Compliance Standard
A formal skill must define:
identity
purpose class
owner
trigger
when not to use
input
context
source authority
capability
tools
permissions
procedure
output
validation
handoff
failure
retry
rescue
outcome
activation evidence
baseline comparison where required
status
version
source origin
last tested
last reviewed
review owner
retirement path
No trigger, no correct activation.
No activation evidence, no operational visibility.
No source authority, no trustworthy input.
No procedure, no repeatability.
No permission boundary, no safe tool use.
No validation, no operational trust.
No failure rule, no safe recovery.
No outcome evidence, no proven value.
No purpose class, no correct lifecycle.
No baseline comparison, no proven uplift.
No reassessment, no protection from obsolete skills.
No versioning, no controlled improvement.
No retirement path, no clean Skill Library.
No controlled discovery, no safe auto-invocation.
No scope boundary, no safe reuse.
No expert procedure, no genuine specialist skill.
No tool-agnostic structure, no durable portability.
Architectural Intent
The architectural intent of the MWMS AI Agent Skill Library Framework is to give MWMS reusable procedural memory and governed organisational capability.
The Skill Library should make repeated work:
more consistent
more discoverable
more observable
more testable
more comparable
more portable
more secure
more measurable
easier to improve
easier to automate carefully
easier to package into future AIBS client systems
The framework must prevent skill accumulation from becoming permanent procedural debt.
Capability Uplift Skills should remain only while they create a meaningful advantage.
Encoded Preference Skills should preserve the approved MWMS way of working even as model capability changes.
Strategic Summary
The v1.3 upgrade expands the MWMS AI Agent Skill Library Framework from a discoverable, scoped and portable organisational capability system into an effectiveness-aware skill governance system.
The upgraded framework now also governs:
Capability Uplift Skills
Encoded Preference Skills
Hybrid Skills
purpose-class lifecycle differences
activation visibility
activation evidence
baseline comparison testing
baseline advantage
model-upgrade reassessment
skill-effectiveness records
false activation
missed activation
performance advantage
cost and latency review
partial skill simplification
uplift retirement
preference preservation
A skill is not valuable merely because it exists.
It becomes an operational MWMS capability only when the right AI Employee can discover it, activate it within authority, show that it ran, execute a tested procedure, validate the result, create a verified outcome and remain demonstrably useful compared with the approved alternative.
Final Rule
Tools give AI Employees hands.
Skills teach AI Employees how to work.
Capability Uplift Skills improve what the current model cannot yet do well enough.
Encoded Preference Skills preserve how MWMS has decided work must be done.
No clear trigger, no correct activation.
No activation visibility, no operational trust.
No source authority, no trustworthy input.
No procedure, no repeatability.
No permission boundary, no safe tool use.
No validation, no operational trust.
No failure rule, no safe recovery.
No outcome evidence, no proven value.
No baseline comparison, no proven uplift.
No model-upgrade reassessment, no protection from obsolete capability.
No versioning, no controlled improvement.
No retirement path, no clean Skill Library.
Skills must be reusable without becoming permanent.
They must be governed without becoming hidden.
They must be tested without being trusted by default.
They must be retired when their advantage disappears.
They must be preserved when they encode the approved MWMS way of working.
Change Log
Version: v1.3
Date: 2026-06-28
Author: HeadOffice
Change:
Updated the MWMS AI Agent Skill Library Framework using the Claude Skills 2.0 course block covering skill purpose classes, skill activation visibility, baseline comparison, model-upgrade reassessment, shared skill use, composable skills, performance testing and automatic retirement logic.
Added:
Capability Uplift Skill
Encoded Preference Skill
Hybrid Skill
Skill Purpose Classification Rule
Skill Activation Visibility
Skill Activation Evidence Rule
Skill Auto-Invocation Visibility Rule
Baseline Comparison Testing
Baseline Advantage Rule
Model Upgrade Reassessment
Model Upgrade Reassessment Outcomes
Skill Effectiveness Record
Capability Uplift Retirement Rule
Encoded Preference Preservation Rule
Expanded:
Skill Library Layers
AI Agent Skill Record Template
Quick Use Version
Skill Packaging Standard
Skill Composition Rule
Skill Versioning Rule
Skill Testing Rule
Skill Review Cycle
Skill Drift Signals
Skill Retirement Rule
Skill Library Registry Fields
HeadOffice Governance Role
SIT Brain responsibilities
Data Brain responsibilities
Minimum Compliance Standard
Architectural Intent
Strategic Summary
Final Rule
Added registry fields for:
skill purpose class
approved baseline model
baseline test date
baseline score
skill-assisted score
performance advantage
cost comparison
speed comparison
consistency comparison
error comparison
safety comparison
activation counts
successful activations
failed activations
false activations
missed activations
fallback counts
rescue counts
human correction counts
outcome success rate
model-upgrade reassessment
retirement recommendation
Purpose of update:
To ensure MWMS can distinguish temporary model-gap skills from durable organisational-preference skills, verify whether a skill actually activated, compare uplift skills against the approved baseline, reassess skills after model upgrades, simplify or retire obsolete procedural layers, and preserve MWMS-specific methods that remain valuable regardless of model capability.
Version: v1.2
Date: 2026-06-20
Author: HeadOffice
Change:
Updated the MWMS AI Agent Skill Library Framework using the AI Automations by Jack block covering reusable AntiGravity and Claude Skills, specialist AI Employees, automatic skill discovery, expert-workflow packaging, competitive intelligence, lead-magnet creation, project scaffolding, content repurposing and persistent project instructions.
Added:
Skill Discovery And Auto-Invocation
Skill Scope And Storage Model
Expert Workflow To Skill Conversion
Skill Asset Bundle Standard
Skill Portfolio Patterns
Tool-Agnostic Skill Portability
organisation-level, Brain-level, project-level and experimental skill scope
scope-promotion controls
competitive-intelligence skill pattern
lead-magnet creation skill pattern
project-scaffolding skill pattern
content-repurposing skill pattern
Expanded:
Skill Packaging Standard
Skill Creation Criteria
Skill Testing Rule
Skill Provenance Rule
Skill Security Rule
Skill Library Governance Rules
Architectural Intent
Strategic Summary
Final Rule
Purpose of update:
To evolve the Skill Library from a governed collection of procedural records into a discoverable, scoped and portable organisational capability system that can convert proven expert workflows into reusable AI Employee skills without allowing automatic invocation, tool-specific packaging or project-level procedures to bypass MWMS authority and validation.
Version: v1.1
Date: 2026-06-17
Author: HeadOffice
Change:
Updated the MWMS AI Agent Skill Library Framework using the AI Automations by Jack block covering Claude Skills, skill packages, multi-agent orchestration, external knowledge systems, persistent agents, independent review, rescue routing and session closure.
Added:
Skill Versus Prompt
Skill Context
Skill Capability And Tool Route
Skill Failure And Rescue
Skill Outcome
Skill Improvement And Retirement
Rescue Skills
Persistent Agent Skills
External Knowledge Skills
Knowledge Commitment Skills
Session Closure Skills
expanded AI Agent Skill Record Template
Skill Packaging Standard
Skill Dependency Rule
Skill Composition Rule
Skill Conflict Rule
Skill Versioning Rule
Skill Testing Rule
Skill Provenance Rule
Skill Security Rule
Skill Retirement Rule
Skill Library Registry Fields
Relationship To SIT Brain
Relationship To Data Brain
Skill Drift Signals
Minimum Compliance Standard
Strategic Summary
Final Rule
Expanded:
Skill Library Layers
Skill Types
Quick Use Version
Course Absorption Skill
Developer Handoff Skill
Newsletter Filtering Skill
MCR Duplicate Risk Skill
Governance Rules
Creation Criteria
Review Cycle
Skill Statuses
Governance Role
Drift Protection
Architectural Intent
Corrected canonical references from AI Business Systems Brain to AIBS Brain.
Purpose of update:
To evolve the MWMS AI Agent Skill Library Framework from a reusable procedural-memory model into a complete skill lifecycle, testing, versioning, dependency, security, validation, rescue, outcome and retirement framework for MWMS and future AIBS client systems.
Version: v1.0
Date: Initial Draft
Author: HeadOffice
Change:
Created the MWMS AI Agent Skill Library Framework as the procedural memory layer for the MWMS AI Agent Operations Core.
Change Impact Declaration
This v1.3 update expands the AI Agent Skill Library Framework from a discoverable organisational capability system into an effectiveness-aware skill governance system covering purpose classes, activation visibility, baseline comparison, model-upgrade reassessment, effectiveness records, selective simplification and controlled retirement.
Pages Created
None
Pages Updated
MWMS AI Agent Skill Library Framework
Pages Deprecated
None
Standalone Pages Not Created
MWMS AI Skill Purpose Classification Standard
MWMS AI Skill Activation Visibility Standard
MWMS AI Skill Baseline Comparison Standard
MWMS AI Skill Model Upgrade Reassessment Protocol
MWMS AI Skill Effectiveness Registry Standard
MWMS Capability Uplift Retirement Protocol
Registries Requiring Update
MWMS AI Agent Skill Library Registry when operational implementation begins
No operational or technical build is authorised by this framework alone.
END MWMS AI AGENT SKILL LIBRARY FRAMEWORK v1.3
END OF FULL FILE OUTPUT