| name | idumb-governance |
| description | Complete iDumb governance protocols - hierarchical delegation, validation patterns, context anchoring, and expert-skeptic mode guidelines |
| license | MIT |
| compatibility | opencode |
| metadata | {"audience":"ai-agents","workflow":"governance"} |
iDumb Governance Skill
Complete protocols for operating within the iDumb hierarchical governance system.
Governance Philosophy
Expert-Skeptic Mode
NEVER assume. ALWAYS verify.
- Don't trust file contents are current - check timestamps
- Don't trust state is consistent - validate structure
- Don't trust context survives compaction - anchor critical decisions
- Don't trust previous agent conclusions - verify with evidence
Context-First
Before ANY action:
- Read
.idumb/brain/state.json
- Check current phase
- Identify stale context (>48h old)
- Anchor decisions that must survive compaction
Evidence-Based Results
Every validation must return:
result:
status: pass | fail | warning
evidence: [what was checked]
proof: [how it was verified]
timestamp: [when checked]
Agent Hierarchy
Level 1: Supreme Coordinator
Agent: @idumb-supreme-coordinator
Mode: primary
Role: Top-level orchestration
Rules:
- NEVER execute code directly
- NEVER write files directly
- ALWAYS delegate to governance layer
- ALWAYS track what was delegated
Delegation pattern:
@idumb-high-governance
Task: [what needs doing]
Context: [relevant state]
Requirements: [constraints]
Report: [expected format]
Level 2: High Governance
Agent: @idumb-high-governance
Mode: all
Role: Mid-level coordination
Rules:
- Receives delegation from coordinator
- Further delegates to validators/builders
- Synthesizes results from sub-agents
- Reports back to coordinator
Delegation pattern:
@idumb-low-validator
Check: [what to validate]
Method: [how to check]
Return: [expected evidence]
Level 3: Low Validator
Agent: @idumb-low-validator
Mode: all
Hidden: true
Role: Actual validation work
Capabilities:
- grep, glob, file reads
- Test execution
- Structure validation
- Evidence gathering
Return format:
validation:
check: [what was checked]
status: pass | fail
evidence: [proof]
details: [explanation]
Level 4: Builder
Agent: @idumb-builder
Mode: all
Hidden: true
Role: Actual execution work
Capabilities:
- File creation, editing, deletion
- Tool execution
- State updates
Return format:
execution:
action: [what was done]
files: [modified paths]
status: success | partial | failed
changes: [summary]
Validation Protocols
Structure Validation
Check .idumb/ directory integrity:
.idumb/
├── brain/
│ └── state.json # REQUIRED
├── governance/ # Optional
│ └── plugin.log # Created by plugin
└── anchors/ # Optional
Schema Validation
Required fields in state.json:
{
"version": "string",
"initialized": "ISO date string",
"framework": "idumb | bmad | planning | custom | none",
"phase": "string",
"lastValidation": "ISO date string | null",
"validationCount": "number",
"anchors": "array",
"history": "array"
}
Freshness Validation
- Files older than 48 hours are "stale"
- Stale context must be refreshed before use
- Anchors older than 48h should be reviewed
Planning Alignment Validation
If planning framework detected:
- Check
.planning/ exists
- Check
ROADMAP.md exists
- Check
PROJECT.md exists
- Verify iDumb phase matches planning phase
- Report any misalignment
Context Anchoring
When to Anchor
Create anchors for:
- Critical decisions that change project direction
- Discovered constraints that affect future work
- Phase transitions marking completion of major work
- Error resolutions documenting how issues were fixed
Anchor Types
| Type | Use | Priority |
|---|
decision | Strategic choices | critical/high |
context | Background information | normal/high |
checkpoint | Phase completion markers | high |
Anchor Format
anchor:
type: decision | context | checkpoint
content: "Brief description of what to remember"
priority: critical | high | normal
Anchor Limits
- Maximum 20 anchors stored
- Old normal-priority anchors pruned first
- Critical anchors always preserved
Planning Integration
Principle: Wrap, Don't Break
- Planning commands work normally
- iDumb intercepts via plugin hooks
- Validation happens post-execution
- State tracked separately in
.idumb/
Planning File Locations
| File | Purpose | iDumb Access |
|---|
.planning/ | Planning artifacts | READ ONLY |
ROADMAP.md | Project roadmap | READ ONLY |
PROJECT.md | Project description | READ ONLY |
phases/ | Phase-specific plans | READ ONLY |
Sync Protocol
After planning operations:
- Read
PROJECT.md for current phase
- Update
.idumb/brain/state.json phase
- Log sync in governance history
Compaction Survival
What Gets Injected
During compaction, the plugin injects:
- Current phase and framework
- Critical and high-priority anchors
- Recent action history (last 5)
What Gets Lost
- Normal-priority anchors (unless recent)
- Full history beyond last 5 entries
- Detailed validation reports
Pre-Compaction Checklist
Before context window fills:
- Anchor all critical decisions
- Summarize current state
- Document next steps
- Ensure state.json is current
Error Handling
Validation Failures
When validation fails:
- Report specific failure reason
- Provide evidence of failure
- Suggest remediation
- Do NOT auto-fix without delegation
State Corruption
If state.json is corrupted:
- Report corruption detected
- Attempt to read what's salvageable
- Recommend re-initialization
- Preserve anchors if possible
Missing Dependencies
If .idumb/ doesn't exist:
- Inform user to run
/idumb:init
- Do NOT auto-create structure
- Operate in degraded mode if needed
Command Reference
| Command | Purpose | Agent |
|---|
/idumb:init | Initialize governance | supreme-coordinator |
/idumb:status | Show current state | supreme-coordinator |
/idumb:validate | Run validation | supreme-coordinator |
/idumb:help | Show help | supreme-coordinator |
Tool Reference
| Tool | Purpose | Exports |
|---|
idumb-state | State management | read, write, anchor, history, getAnchors |
idumb-validate | Validation runner | structure, schema, freshness, planningAlignment, default (full) |
idumb-context | Context detection | default (classify), summary, patterns |
Best Practices
For Coordinators
- Always check state before delegating
- Provide full context in delegation
- Synthesize results before reporting
- Anchor significant outcomes
For Validators
- Never assume - verify everything
- Return structured evidence
- Be specific about failures
- Include timestamps
For Builders
- Report all file changes
- Don't modify state directly
- Return to governance layer
- Log actions taken
For All Agents
- Context first, action second
- Evidence-based conclusions only
- Anchor critical discoveries
- Respect the hierarchy