Causal Reasoning Skill | Agent Skills

Causal Reasoning

Execute systematic cause-effect reasoning. The logic of process and action.

Relationship to Goals

Threads are the execution layer for goals. Goals define what to achieve; threads define how.

Goal (goal-setter)
  └── Subgoal
        └── Thread (reasoning-causal) ← executes via 6-stage flow
              └── Learning → updates Goal state (goal-tracker)

Thread types:

Goal-linked: Created from subgoals, has goal_id in metadata
Reactive: Created from signals (no goal), may spawn or link to goal

Type Signature

Causal : Input → Hypothesis → Implication → Decision → Action → Learning

Where:
  Input       : Observation × Context → FactualStatement
  Hypothesis  : FactualStatement × CanvasAssumption → TestableHypothesis
  Implication : TestableHypothesis → (Impact × Probability × Timeline)
  Decision    : Implication × Alternatives → Commitment
  Action      : Commitment → [ExecutableTask]
  Learning    : [ExecutedTask] × Outcomes → CanvasUpdate × GoalUpdate

When to Use

Process execution with known steps
Decision with clear cause-effect chain
Operational workflows (sales, marketing, engineering)
Canvas hypothesis testing
Action planning and execution
Executing subgoals (goal-linked threads)

Thread Types

| Type | Location | Use For | |------|----------|---------| | Business | threads/operations/{name}/ | Strategic decisions, product changes | | Sales | threads/sales/{name}/ | Deal pipelines, prospects | | Marketing | threads/marketing/{name}/ | Campaigns, content launches | | Engineering | threads/engineering/{name}/ | Requirements → specifications |

Thread-specific details: See references/threads/{type}.md

6-Stage Flow

Execute stages sequentially. Each stage produces a markdown file in the thread directory.

Stage 1: Input

File: 1-input.md Purpose: Capture factual observation that triggers the flow.

Content:

What happened? (fact, not opinion)
When? Where? Who observed?
Raw data/evidence links
Context (what we believed before)

Rules:

Facts only, no interpretation
No solutions or recommendations
Link to evidence

Detail: references/stages/input.md

Stage 2: Hypothesis

File: 2-hypothesis.md Purpose: Link observation to Canvas assumption being tested.

Content:

Which assumption does this challenge/validate?
What do we believe will happen?
What would prove us wrong?
Testable prediction

Rules:

Must reference strategy/canvas/10.assumptions.md
State falsifiable hypothesis
Define success/failure criteria

Detail: references/stages/hypothesis.md

Stage 3: Implication

File: 3-implication.md Purpose: Analyze business impact with numbers.

Content:

Revenue impact (quantified)
Timeline (short/medium/long)
Resource requirements
Risk assessment
Opportunity cost

Rules:

Include specific numbers
Compare scenarios
Identify dependencies

Detail: references/stages/implication.md

Stage 4: Decision

File: 4-decision.md Purpose: Make official commitment with impact score.

Content:

Decision statement (PROCEED/DEFER/DECLINE)
Alternatives considered
Impact score calculation
Approval status

Impact Scoring:

| Score | Action | |-------|--------| | < 0.8 | Auto-execute | | ≥ 0.8 | Flag for human approval |

Mode-Aware Formulas:

VENTURE: Impact = (Strategic Value × Market Size × Defensibility) / 3 BOOTSTRAP: Impact = (Revenue Impact × Time to Cash × Margin) / 3

Check strategy/canvas/00-business-model-mode.md for mode.

Detail: references/stages/decision.md

Stage 5: Actions

File: 5-actions.md or 5-actions/ directory Purpose: Generate executable tasks.

Content:

Typed actions (sales:, marketing:, engineering:*)
Assigned owners
Deadlines
Success criteria
Dependencies

Action Types by Thread:

| Thread | Action Types | Skills | |--------|--------------|--------| | Sales | lead-intake, qualify, demo, pilot, close | sales-* | | Marketing | research, create, publish, promote, measure | marketing-* | | Engineering | requirements, specification, implementation | engineering-* | | Business | varies by decision | - |

Detail: references/stages/actions.md

Stage 6: Learning

File: 6-learning.md Purpose: Document outcomes and update Canvas + Goal.

Content:

Actual vs expected outcome
Hypothesis validated/invalidated?
Canvas sections to update
Goal metrics to update (if goal-linked)
New threads generated

Rules:

Update strategy/canvas/10.assumptions.md
Link learning to original hypothesis
If goal-linked: Update goal state via goal-tracker
Generate follow-up threads if needed

Goal Integration:

If thread.goal_id exists:
  1. Read goal from strategy/goals/active/{goal_id}.md
  2. Update subgoal status (pending → completed)
  3. Extract metrics from learning for goal state
  4. Check if goal success criteria met
  5. If all subgoals complete → mark goal completed

Detail: references/stages/learning.md

Workflow

Goal-Linked Thread (Primary)

1. Receive subgoal from goal-setter
2. Create thread: threads/{type}/{name}/
3. Set meta.json with goal_id and subgoal
4. Execute stages 1-6 sequentially
5. At Stage 4: Calculate impact, flag if ≥0.8
6. At Stage 6: Update Canvas AND goal state
7. Notify goal-tracker of completion

Reactive Thread (Fallback)

1. Receive signal (feedback, anomaly, opportunity)
2. Create thread: threads/{type}/{name}/
3. Set meta.json without goal_id
4. Execute stages 1-6 sequentially
5. At Stage 4: Calculate impact, flag if ≥0.8
6. At Stage 6: Update Canvas
7. Optionally: Link to existing goal or spawn new goal

Thread Structure

threads/{type}/{name}/
├── meta.json           # Thread metadata (includes goal linkage)
├── 1-input.md          # Factual observation
├── 2-hypothesis.md     # Canvas assumption link
├── 3-implication.md    # Impact analysis
├── 4-decision.md       # Commitment + impact score
├── 5-actions.md        # Executable tasks
└── 6-learning.md       # Outcomes + Canvas/Goal update

Thread Metadata (meta.json)

{
  "id": "thread-{type}-{name}",
  "type": "business | sales | marketing | engineering",
  "status": "active | completed | blocked",
  "created": "YYYY-MM-DD",
  "updated": "YYYY-MM-DD",
  "goal_id": "g-{goal-id}",        // Optional: linked goal
  "subgoal": "SG1",                 // Optional: which subgoal
  "stage": 1-6,
  "impact_score": 0.0-1.0
}

Goal-linked threads:

goal_id references strategy/goals/active/{goal-id}.md
subgoal indicates which subgoal this thread executes
Stage 6 learning updates both Canvas AND goal state

Reactive threads (no goal):

goal_id is null or absent
At completion, may link to existing goal or spawn new goal

Decision Authority

AI Autonomous (Impact <0.8):

Within strategic direction
ROI > 3x, risk low-medium
Cost <$100K, timeline <3 months

Human Review (Impact ≥0.8):

Strategic pivot
ROI <2x, high risk
Cost ≥$100K, timeline ≥3 months
Canvas-altering decisions

References

references/
├── stages/           # Stage execution details
│   ├── input.md
│   ├── hypothesis.md
│   ├── implication.md
│   ├── decision.md
│   ├── actions.md
│   └── learning.md
└── threads/          # Thread type specifics
    ├── operations.md
    ├── sales.md
    ├── marketing.md
    └── engineering.md

Note: Action execution uses flat skills (sales-*, marketing-*, engineering-*) not templates.

Success Criteria

Goal-aligned: Thread serves a goal subgoal (when goal-linked)
Evidence-based: Starts with factual observation
Hypothesis-driven: Links to Canvas assumptions
Impact-analyzed: Quantified cost/benefit
Traceable: Complete 6-stage audit trail
Self-correcting: Canvas AND goal updates from learning
Autonomous: AI executes >95% (impact <0.8)

Remember

Every decision flows through 6 stages. No shortcuts.

Goals are primary. Threads execute goals. Reactive threads are fallback.

This skill:

Executes the 6-stage causal flow
Links threads to goals (when goal-linked)
Reads reference docs for detail
Calculates impact scores
Updates Canvas AND goal state from learning
Flags high-impact items for human review

Agent Skills: Causal Reasoning

Install this agent skill to your local

Skill Files