@cleocode/core 2026.4.4 → 2026.4.6

package/src/validation/protocols/protocols-markdown/decomposition.md

@@ -0,0 +1,421 @@
+---
+id: DCMP
+title: Decomposition Protocol
+version: 2.0.0
+status: active
+type: conditional
+audience: [llm-agent, orchestrator]
+tags: [decomposition, hierarchy, atomicity, planning]
+skillRef: ct-epic-architect
+mcpOperations:
+  - tasks.add
+  - tasks.update
+  - session.start
+lastUpdated: 2026-02-24
+provenanceTask: T3155
+enforcement: strict
+---
+
+# Decomposition Protocol
+
+**Provenance**: @task T3155, @epic T3147
+**Version**: 2.0.0
+**Type**: Conditional Protocol
+**Max Active**: 3 protocols (including base)
+
+---
+
+## Trigger Conditions
+
+This protocol activates when the task involves:
+
+| Trigger | Keywords | Context |
+|---------|----------|---------|
+| Epic Planning | "epic", "project", "initiative" | Large scope work |
+| Breakdown | "decompose", "break down", "split" | Task subdivision |
+| Planning | "plan", "roadmap", "phases" | Work organization |
+| Analysis | "scope", "estimate complexity" | Size assessment |
+
+**Explicit Override**: `--protocol decomposition` flag on task creation.
+
+---
+
+## Hard Invariants vs. Configurable Limits
+
+### Hard Invariants (always enforced regardless of profile)
+- **Parent must exist**: The parent task ID must be present in the task store
+- **No circular references**: Moving a task under its own descendant is always rejected
+- **Dependency DAG integrity**: The blockedBy/blocks relationship must remain acyclic
+
+### Configurable Limits (profile-driven, overridable per project)
+- **maxSiblings** (default: 0 = unlimited for `llm-agent-first` profile; 7 for `human-cognitive`)
+  — Set `hierarchy.maxSiblings` in `.cleo/config.json` to enforce a limit
+- **maxActiveSiblings** (default: 32): Operational concurrency guard — caps the number of `pending|active|blocked` siblings under one parent
+- **maxDepth** (default: 3): Maximum nesting depth (epic → task → subtask)
+
+### Enforcement Profiles
+
+| Profile | maxSiblings | maxActiveSiblings | maxDepth |
+|---------|-------------|-------------------|----------|
+| `llm-agent-first` (default) | 0 (unlimited) | 32 | 3 |
+| `human-cognitive` | 7 | 3 | 3 |
+
+Configure via `.cleo/config.json`:
+```json
+{ "hierarchy": { "enforcementProfile": "human-cognitive" } }
+```
+
+---
+
+## Requirements (RFC 2119)
+
+### MUST
+
+| Requirement | Description |
+|-------------|-------------|
+| DCMP-001 | MUST follow MECE principle (Mutually Exclusive, Collectively Exhaustive) |
+| DCMP-002 | MUST map dependencies between tasks |
+| DCMP-003 | MUST respect the configured hierarchy.maxDepth policy (default: 3) |
+| DCMP-004 | MUST verify atomicity for leaf tasks |
+| DCMP-005 | MUST NOT include time estimates (use size: small/medium/large) |
+| DCMP-006 | MUST include acceptance criteria for each task |
+| DCMP-007 | Siblings MUST respect the configured hierarchy.maxSiblings policy (0 = unlimited) |
+| DCMP-008 | Depth MUST respect the configured hierarchy.maxDepth policy (default: 3) |
+| DCMP-009 | MUST set `agent_type: "analysis"` in manifest |
+
+### SHOULD
+
+| Requirement | Description |
+|-------------|-------------|
+| DCMP-010 | SHOULD identify parallel execution opportunities |
+| DCMP-011 | SHOULD flag unclear requirements for HITL |
+| DCMP-012 | SHOULD consider existing task overlap |
+| DCMP-013 | SHOULD assign phase to each task |
+
+### MAY
+
+| Requirement | Description |
+|-------------|-------------|
+| DCMP-020 | MAY propose multiple decomposition options |
+| DCMP-021 | MAY identify risks and mitigations |
+| DCMP-022 | MAY suggest task consolidation |
+
+---
+
+## Output Format
+
+### Hierarchy Structure
+
+```
+EPIC (T001) - Large initiative
+├── TASK (T002) - Coherent unit of work
+│   ├── SUBTASK (T003) - Atomic task (MUST be atomic)
+│   └── SUBTASK (T004) - Atomic task
+├── TASK (T005) - Another unit
+│   └── SUBTASK (T006)
+└── TASK (T007) - Independent task
+```
+
+### Size Definitions
+
+| Size | Scope | Atomicity |
+|------|-------|-----------|
+| `small` | 1-2 files, single concern | MUST be atomic |
+| `medium` | 3-7 files, related concerns | SHOULD be atomic |
+| `large` | 8+ files, multiple concerns | MUST decompose further |
+
+### Atomicity Criteria (6-Point Test)
+
+| # | Criterion | Pass Condition |
+|---|-----------|----------------|
+| 1 | Single File Scope | Affects ≤3 tightly-coupled files |
+| 2 | Single Cognitive Concern | One bounded concern per task |
+| 3 | Clear Acceptance Criteria | Testable completion condition |
+| 4 | No Context Switching | Completable within a single agent session |
+| 5 | No Hidden Sub-Decisions | All choices made at decomposition |
+| 6 | Programmatic Validation | Result verifiable by code/test |
+
+### Dependency Graph
+
+```json
+{
+  "tasks": [
+    {"id": "T001", "type": "epic", "children": ["T002", "T005", "T007"]},
+    {"id": "T002", "type": "task", "parent": "T001", "children": ["T003", "T004"], "depends": []},
+    {"id": "T003", "type": "subtask", "parent": "T002", "depends": []},
+    {"id": "T004", "type": "subtask", "parent": "T002", "depends": ["T003"]},
+    {"id": "T005", "type": "task", "parent": "T001", "depends": ["T002"]},
+    {"id": "T007", "type": "task", "parent": "T001", "depends": []}
+  ],
+  "waves": [
+    {"wave": 0, "tasks": ["T003", "T007"]},
+    {"wave": 1, "tasks": ["T004"]},
+    {"wave": 2, "tasks": ["T002"]},
+    {"wave": 3, "tasks": ["T005"]}
+  ]
+}
+```
+
+### Wave Computation
+
+Tasks are grouped into waves for parallel execution:
+
+| Wave | Condition | Execution |
+|------|-----------|-----------|
+| 0 | No dependencies | Parallel |
+| 1 | Depends only on wave 0 | Parallel after wave 0 |
+| N | Depends on wave N-1 or earlier | Sequential ordering |
+
+### File Output
+
+```markdown
+# Decomposition: {Epic Title}
+
+**Task**: T####
+**Date**: YYYY-MM-DD
+**Status**: complete|partial|blocked
+**Agent Type**: analysis
+
+---
+
+## Epic Summary
+
+{2-3 sentence summary of the epic scope}
+
+## Task Hierarchy
+
+### T001 - {Epic Title} (epic, large)
+
+**Description**: {Epic description}
+**Phase**: {phase}
+
+#### T002 - {Task Title} (task, medium)
+
+**Description**: {Task description}
+**Depends**: none
+**Phase**: {phase}
+**Acceptance Criteria**:
+- {Criterion 1}
+- {Criterion 2}
+
+##### T003 - {Subtask Title} (subtask, small)
+
+**Description**: {Subtask description}
+**Depends**: none
+**Atomicity**: 6/6 (PASS)
+**Acceptance Criteria**:
+- {Testable criterion}
+
+## Dependency Graph
+
+```mermaid
+graph TD
+    T001[Epic] --> T002[Task A]
+    T001 --> T005[Task B]
+    T002 --> T003[Subtask 1]
+    T002 --> T004[Subtask 2]
+    T003 --> T004
+```
+
+## Wave Analysis
+
+| Wave | Tasks | Parallel Opportunities |
+|------|-------|----------------------|
+| 0 | T003, T007 | 2 tasks parallel |
+| 1 | T004 | Sequential |
+| 2 | T002, T005 | 2 tasks parallel |
+
+## Risks
+
+| Risk | Impact | Mitigation |
+|------|--------|------------|
+| {Risk} | {Impact} | {Mitigation} |
+
+## Open Questions
+
+- {Questions requiring HITL resolution}
+```
+
+### Manifest Entry
+
+@skills/_shared/manifest-operations.md
+
+Use `cleo research add` to create the manifest entry:
+
+```bash
+cleo research add \
+  --title "Decomposition: Epic Title" \
+  --file "YYYY-MM-DD_decomposition.md" \
+  --topics "decomposition,planning,epic" \
+  --findings "12 tasks identified,3 wave parallel plan,2 HITL questions" \
+  --status complete \
+  --task T#### \
+  --actionable \
+  --needs-followup T#### \
+  --agent-type analysis
+```
+
+---
+
+## Integration Points
+
+### Base Protocol
+
+- Inherits task lifecycle (start, execute, complete)
+- Inherits manifest append requirement
+- Inherits error handling patterns
+
+### Protocol Interactions
+
+| Combined With | Behavior |
+|---------------|----------|
+| research | Research informs scope understanding |
+| specification | Spec defines task boundaries |
+| implementation | Decomposition feeds implementation order |
+
+### CLEO Integration
+
+```bash
+# Create epic and tasks from decomposition
+cleo add "Epic Title" --type epic --size large
+
+# Add tasks with hierarchy
+cleo add "Task A" --parent T001 --type task --size medium --phase core
+
+# Add subtasks
+cleo add "Subtask 1" --parent T002 --type subtask --size small --phase core
+
+# Set dependencies
+cleo update T004 --depends T003
+```
+
+---
+
+## Example
+
+**Task**: Decompose CLEO Universal Subagent Architecture
+
+**Manifest Entry Command**:
+```bash
+cleo research add \
+  --title "Decomposition: Universal Subagent Architecture" \
+  --file "2026-01-26_subagent-decomposition.md" \
+  --topics "subagent,architecture,decomposition" \
+  --findings "14 tasks identified,4 parallel waves,Base protocol + 7 conditional protocols" \
+  --status complete \
+  --task T2392 \
+  --actionable \
+  --needs-followup T2393,T2394 \
+  --agent-type analysis
+```
+
+---
+
+## Atomicity Validation Module
+
+The 6-point atomicity test is implemented as a shared TypeScript module at
+`src/core/tasks/atomicity.ts`. The `checkAtomicity(task, threshold?)` function
+scores tasks against these criteria:
+
+1. **single-file-scope** — Task targets a single file or bounded concern
+2. **single-cognitive-concern** — Title uses at most one action verb
+3. **clear-acceptance-criteria** — Description specifies verifiable completion
+4. **no-context-switching** — Task spans at most 2 technical domains
+5. **no-hidden-decisions** — No deferred choices (TBD, figure out, unclear)
+6. **programmatic-validation-possible** — Output can be tested programmatically
+
+Default passing threshold: 4/6. The `ATOMICITY_CRITERIA` constant is exported
+for use in protocol validators and orchestrator pre-spawn gating.
+
+---
+
+## Atomicity Validation
+
+### Atomicity Checklist
+
+For each leaf task, verify:
+
+- [ ] **Single Responsibility**: One clear objective
+- [ ] **Completable in 1 agent session**: Scoped to a single execution context
+- [ ] **Testable**: Has clear acceptance criteria
+- [ ] **Mergeable**: Can be PR'd independently
+- [ ] **Reversible**: Can be rolled back if needed
+- [ ] **Dependencies Clear**: Blocking tasks explicit
+
+### Validation Command
+
+**Command** (planned): `cleo atomicity check --epic EPIC_ID`
+
+**Current State**: Available via `validate_decomposition_protocol()` in `lib/protocol-validation.sh`
+
+```bash
+# Programmatic validation
+source lib/protocol-validation.sh
+child_tasks=$(cleo list --parent T2000 --format json | jq -c '.tasks')
+result=$(validate_decomposition_protocol "T2000" "T2000" "$child_tasks" "false")
+```
+
+**Checks**:
+- DCMP-002: No circular dependencies (via hierarchy validation)
+- DCMP-003: Exceeds configured maxDepth (default: 3, epic→task→subtask)
+- DCMP-004: Atomicity test (6 criteria above)
+- DCMP-007: Exceeds configured maxSiblings (default: unlimited for llm-agent-first profile)
+
+### Scoring System
+
+Each leaf task scored on 6 criteria (pass/fail per criterion):
+
+| Score | Interpretation | Action |
+|-------|---------------|--------|
+| 6/6 | Atomic | Proceed |
+| 4-5/6 | Mostly atomic | Review failing criteria |
+| 2-3/6 | Needs refinement | Split or clarify |
+| 0-1/6 | Not atomic | Reject, re-decompose |
+
+**Aggregate Epic Score**: Average of leaf task scores
+
+### Orchestrator Integration
+
+Hierarchy enforcement is implemented in `src/core/tasks/hierarchy-policy.ts`.
+The `resolveHierarchyPolicy(config)` function returns the active `HierarchyPolicy`
+for the project, incorporating the configured enforcement profile and any explicit overrides.
+
+**Pre-Decomposition**:
+- Epic architect validates parent epic exists
+- Checks current depth against the configured hierarchy.maxDepth policy (default: 3)
+- Verifies sibling count against the configured hierarchy.maxSiblings policy (default: unlimited for llm-agent-first profile)
+
+**Post-Decomposition**:
+- Validates all new tasks meet atomicity criteria
+- Blocks spawn of non-atomic tasks
+- Returns refinement suggestions
+
+### Exit Codes
+
+- `EXIT_PROTOCOL_DECOMPOSITION` (63) - Decomposition protocol violation
+- Common violations: Configured maxSiblings exceeded, configured maxDepth exceeded, non-atomic tasks
+
+### Auto-Refinement (Future)
+
+**Planned**: If atomicity score <4/6:
+1. Analyze failing criteria
+2. Suggest task splits
+3. Generate refined decomposition
+4. Re-validate until score ≥4/6
+
+---
+
+## Anti-Patterns
+
+| Pattern | Why Avoid |
+|---------|-----------|
+| Time estimates | Cannot predict duration accurately |
+| Non-atomic leaf tasks | Create hidden work |
+| Missing dependencies | Incorrect execution order |
+| Exceeding configured maxDepth | Complexity explosion |
+| Overlapping scopes (not MECE) | Duplicate work |
+| Skipping atomicity check | Quality regression |
+
+---
+
+*Protocol Version 2.0.0 - Decomposition Protocol*