sandgardenhq/sgai/stpa-step1-define-purpose

STPA Step 1: Define Purpose of Analysis

Objective

Establish the foundation for the entire STPA analysis by identifying:

1. Losses - What we absolutely cannot afford to happen
2. Hazards - System states that could lead to those losses
3. System-Level Constraints - What the system must do/not do to prevent hazards

Announce at Start

"I'm using the STPA Step 1 skill to define the purpose of our analysis. We'll identify losses, hazards, and constraints through a series of questions."

Phase 1: Identify Losses (L)

Questions to Ask (ONE at a time)

Q1: What are the unacceptable outcomes for this system?

• Examples for software: data loss, unauthorized access, system downtime, financial loss
• Examples for physical: injury, property damage, environmental harm
• Examples for AI: biased decisions, safety violations, unintended actions

Q2: For each loss, how severe is the impact?

• Critical: Immediate danger or catastrophic failure
• Serious: Significant damage or major disruption
• Moderate: Notable impact but recoverable
• Minor: Inconvenience with minimal lasting effect

Recording Losses

#### Losses (L)
-L-1: [Loss of human life or serious injury]
-L-2: [Loss of data integrity]
-L-3: [Loss of service availability]
-L-4: [Financial loss exceeding $X]

Phase 2: Identify Hazards (H)

Definition

A hazard is a system state or set of conditions that, together with a worst-case set of environmental conditions, will lead to a loss.

Questions to Ask

Q1: What system states could lead to [Loss L-1]?

• Think about the system's behavior, not component failures
• Focus on what the SYSTEM does, not what COMPONENTS do

Q2: Under what environmental conditions would this hazard lead to the loss?

• Worst-case scenarios
• Concurrent conditions that compound risk

Hazard Template

H-[number]: [System] [unsafe condition/behavior] [leading to L-X]

Examples

Software System:

• H-1: Authentication service grants access to unauthorized users [→ L-2, L-4]
• H-2: Database writes incomplete transaction records [→ L-2]

Physical System:

• H-1: Vehicle accelerates when driver intends to stop [→ L-1]
• H-2: Pressure vessel exceeds safe operating limits [→ L-1, L-3]

AI System:

• H-1: ML model recommends unsafe medication dosage [→ L-1]
• H-2: Autonomous agent takes actions outside approved scope [→ L-1, L-4]

Recording Hazards

#### Hazards (H)
-H-1: [System] [condition] [→ L-X, L-Y]
-H-2: [System] [condition] [→ L-X]

Phase 3: Define System-Level Constraints (SC)

Definition

System-level constraints specify what the system must do or must NOT do to prevent hazards.

Derivation Rule

For each hazard, derive at least one constraint that, if enforced, prevents the hazard.

Constraint Template

SC-[number]: [System] must [always/never] [behavior] [to prevent H-X]

Examples

From H-1 (auth service):

• SC-1: Authentication service must verify credentials before granting access [→ H-1]
• SC-2: Authentication service must never bypass verification checks [→ H-1]

From H-1 (vehicle):

• SC-1: Vehicle must decelerate when brake is applied [→ H-1]
• SC-2: Vehicle must never accelerate during brake application [→ H-1]

Recording Constraints

#### System-Level Constraints (SC)
-SC-1: [System] must [behavior] [→ H-X]
-SC-2: [System] must never [behavior] [→ H-X]

Validation Questions

Before proceeding to Step 2, verify:

1. Completeness: "Are there any other losses you can think of?"
2. Traceability: "Does each hazard trace to at least one loss?"
3. Coverage: "Does each hazard have at least one constraint?"
4. Clarity: "Are the constraints specific enough to be verifiable?"

Output Template

Record in .sgai/PROJECT_MANAGEMENT.md:

## STPA Analysis
 
### Step 1: Purpose Definition
 
#### Losses (L)
-L-1: [description] - Severity: [Critical/Serious/Moderate/Minor]
-L-2: [description] - Severity: [level]
 
#### Hazards (H)
-H-1: [System] [unsafe condition] [→ L-1, L-2]
-H-2: [System] [unsafe condition] [→ L-1]
 
#### System-Level Constraints (SC)
-SC-1: [System] must [behavior] [→ H-1]
-SC-2: [System] must never [behavior] [→ H-1]

When to Proceed to Step 2

Move to Step 2 when:

• [ ] All significant losses are identified
• [ ] Each loss has at least one associated hazard
• [ ] Each hazard has at least one constraint
• [ ] Human partner confirms the analysis looks complete

Load: skills({"name":"stpa/step2-control-structure"})