Canonizing Flaws

Purpose

Interview the user about an emergent behavior that has become beloved or expected, and register it in the Canon of Flaws for protection against "optimization."

Philosophy

"Beloved 'bugs' are registered and protected from optimization"

The key insight is that a "perfect" implementation of the original design would often be worse than the imperfect reality users have come to love. This skill protects the emergent soul of products.

Pre-Flight Checks

Sigil Setup: Verify .sigil-setup-complete exists
Canon File: Check sigil-mark/soul-binder/canon-of-flaws.yaml exists
- If missing, create with empty flaws array
Strictness Level: Load from .sigilrc.yaml

Workflow

Step 1: Identify the Behavior

If behavior_name not provided, ask:

question: "What emergent behavior would you like to protect?"
header: "Behavior"

Explain what qualifies:

Was not originally intended
Users have come to expect or enjoy
Removing would cause confusion or complaints

Examples:

A UI quirk that became a feature
An interaction pattern that emerged from user behavior
A "bug" that users now rely on

Step 2: Interview - Intended vs Emergent

Question 2.1: Intended Behavior

question: "What was the INTENDED behavior? What should have happened according to the original design?"
header: "Intended"

Question 2.2: Emergent Behavior

question: "What ACTUALLY happens (that became beloved)? Describe the behavior users have come to expect."
header: "Emergent"

Question 2.3: Discovery

question: "How did you discover this behavior was valued?"
header: "Discovery"
options:
  - label: "User complaints when it changed"
    description: "Users noticed and complained when behavior was modified"
  - label: "Community discussion"
    description: "Users discussed/documented the behavior"
  - label: "Usage analytics"
    description: "Data showed users relying on this pattern"
  - label: "Internal discovery"
    description: "Team noticed users expecting this behavior"
multiSelect: false

Step 3: Interview - Protection Criteria

Question 3.1: Usage

question: "Approximately what percentage of users rely on this behavior?"
header: "Usage"
options:
  - label: "< 5%"
    description: "Small group, may not meet threshold"
  - label: "5-20%"
    description: "Significant minority, meets threshold"
  - label: "20-50%"
    description: "Large segment, strong candidate"
  - label: "> 50%"
    description: "Majority, critical to protect"
multiSelect: false

If < 5%, warn but allow proceeding with UNDER_REVIEW status.

Question 3.2: Community Attachment

question: "How would users react if this behavior was 'fixed'?"
header: "Attachment"
options:
  - label: "Mild confusion"
    description: "Low attachment - some users might notice"
  - label: "Complaints"
    description: "Moderate attachment - expect support tickets"
  - label: "Outrage/backlash"
    description: "High attachment - would damage trust"
multiSelect: false

Question 3.3: Skill Expression (Optional)

question: "Does this behavior reward skill or expertise?"
header: "Skill"
options:
  - label: "Yes - timing/learning based"
    description: "Requires learning, separates novice from expert"
  - label: "No - discovered by accident"
    description: "Random discovery, not skill-based"
  - label: "N/A"
    description: "Not applicable to this product type"
multiSelect: false

Step 4: Define Protection

Question 4.1: Affected Code

question: "What code patterns might accidentally 'fix' this behavior?"
header: "Patterns"

Provide examples:

*submit*handler*
*debounce*click*
*animation*duration*

Accept glob patterns that should trigger protection checks.

Question 4.2: Protection Rule

question: "Complete this sentence: 'Any change that __________ must be BLOCKED.'"
header: "Rule"

Example: "Any change that prevents the double-click animation must be BLOCKED."

Step 5: Generate Entry

Generate the next flaw ID by reading existing entries:

yaml

- id: "FLAW-{next_id}"
  name: "{behavior_name}"
  status: "PROTECTED"  # or UNDER_REVIEW if < 5% usage
  canonized_date: "{today}"
  canonized_by: "{user or 'Taste Owner'}"

  description: |
    {Brief description of the flaw}

  intended_behavior: |
    {From Q2.1}

  emergent_behavior: |
    {From Q2.2}

  why_protected: |
    - Discovery: {From Q2.3}
    - Usage: {From Q3.1}
    - Attachment: {From Q3.2}
    - {Additional context}

  affected_code_patterns:
    - "{pattern_1}"
    - "{pattern_2}"

  protection_rule: |
    {From Q4.2}

  de_canonization:
    requires_threshold: 70  # percent approval
    cooldown_days: 180

Step 6: Confirm and Save

Show the user the generated entry:

Here's the Canon of Flaws entry I've prepared:

{formatted_entry}

Does this accurately capture the behavior to protect?

question: "Confirm this entry?"
header: "Confirm"
options:
  - label: "Save"
    description: "Add to Canon of Flaws"
  - label: "Edit"
    description: "Make changes before saving"
  - label: "Cancel"
    description: "Discard and exit"
multiSelect: false

On confirmation:

Load existing canon-of-flaws.yaml
Append new flaw to flaws array
Update last_updated timestamp
Save file

Step 7: Report

{status_emoji} FLAW-{id} "{name}" has been added to the Canon of Flaws.

Status: {PROTECTED | UNDER_REVIEW}

The agent will now {BLOCK | WARN on} any change that matches:
  {affected_patterns}

Protection Rule:
  "{protection_rule}"

De-canonization process:
  - Requires 70% community approval via /consult
  - Requires Taste Owner sign-off via /approve
  - Update canon-of-flaws.yaml status to DE_CANONIZED

Next steps:
  - /craft will respect this flaw during implementation
  - Changes to affected patterns will be {blocked | flagged}

Strictness Behavior

Strictness	Protected Flaw Violation
discovery	"Consider" - informational only
guiding	Warning with explanation
enforcing	BLOCK with override option
strict	BLOCK with override option

Error Handling

Error	Cause	Resolution
"Setup not complete"	Missing marker	Run `/setup` first
"Usage below threshold"	< 5% usage	Allow with UNDER_REVIEW status
"Similar flaw exists"	Duplicate pattern	Suggest updating existing flaw
"Canon file missing"	No canon-of-flaws.yaml	Create empty file first

Do NOT

Automatically reject flaws with low usage
Judge whether the behavior "should" be a flaw
Require technical justification

DO

Trust user judgment about community attachment
Capture the emotional context, not just technical details
Make protection actionable with specific patterns
Explain the de-canonization process clearly

Search AI Tools

canonizing-flaws

Install this agent skill to your Project

SKILL.md