Installable skillMIT

SPARC Specification Phase

Run the SPARC Specification phase — gather requirements, define acceptance criteria, identify constraints, and store the spec in memory

BY digitalcrest01UPDATED Oct 28, 2026
INTENDED FOR✳Claude CodeInstall in a listed agent.

instructions.md

---
name: sparc-spec-e43e6ec4eb54
description: Run the SPARC Specification phase — gather requirements, define acceptance criteria, identify constraints, and store the spec in memory
argument-hint: "<feature-description>"
allowed-tools: mcp__claude-flow__memory_store mcp__claude-flow__memory_search mcp__claude-flow__memory_retrieve mcp__claude-flow__task_create mcp__claude-flow__task_update mcp__claude-flow__task_complete mcp__claude-flow__hooks_intelligence_trajectory-start mcp__claude-flow__hooks_intelligence_trajectory-step mcp__claude-flow__neural_predict Bash Read Edit
---

# SPARC Specification Phase

Run Phase 1 of the SPARC methodology: define what must be built and how success is measured.

## When to use

When starting a new feature or project that needs structured requirements gathering before any code is written. This phase produces the foundational specification that all subsequent phases (Pseudocode, Architecture, Refinement, Completion) build upon.

## Steps

1. **Initialize phase tracking** — call `mcp__claude-flow__hooks_intelligence_trajectory-start` with metadata `{ "phase": "specification", "feature": "$ARGUMENTS" }`

2. **Check for prior work** — call `mcp__claude-flow__memory_search` with namespace `sparc-state` and query for the feature to see if a SPARC workflow already exists. If it does, retrieve existing artifacts. If not, initialize state with phase 1.

3. **Search for similar patterns** — call `mcp__claude-flow__neural_predict` with the feature description to find relevant past specifications and learned patterns

4. **Gather requirements** — analyze the feature description and the codebase to identify:
   - **Functional requirements**: what the feature must do (user-facing behaviors)
   - **Non-functional requirements**: performance targets, security constraints, scalability needs
   - **Integration points**: what existing systems or APIs are affected
   - **Data requirements**: what data is created, read, updated, or deleted

5. **Define acceptance criteria** — write at least 3 concrete, testable acceptance criteria in Given/When/Then format:
   ```
   AC-1: Given [precondition], when [action], then [expected result]
   AC-2: Given [precondition], when [action], then [expected result]
   AC-3: Given [precondition], when [action], then [expected result]
   ```

6. **Identify constraints** — document:
   - Performance constraints (latency, throughput, resource limits)
   - Security constraints (authentication, authorization, data sensitivity)
   - Compatibility constraints (browser support, API versions, backward compatibility)
   - Infrastructure constraints (deployment environment, dependencies)

7. **Map edge cases** — list at least 3 edge cases or failure scenarios:
   - What happens with invalid input?
   - What happens under concurrent access?
   - What happens when external dependencies fail?

8. **Store specification** — call `mcp__claude-flow__memory_store` with:
   - Namespace: `sparc-phases`
   - Key: `spec-{feature-slug}`
   - Value: JSON with `{ status: "complete", requirements, acceptanceCriteria, constraints, edgeCases, integrationPoints }`

9. **Update phase state** — call `mcp__claude-flow__memory_store` with:
   - Namespace: `sparc-state`
   - Key: `current-phase-{feature-slug}`
   - Value: updated state with artifacts list including the spec key

10. **Record trajectory step** — call `mcp__claude-flow__hooks_intelligence_trajectory-step` with the specification summary

11. **Present specification** — display the full specification document to the user with a summary table and suggest running `/sparc advance` to pass the gate and move to the Pseudocode phase

## Output format

```
# Specification: {Feature Name}

## Requirements
### Functional
- FR-1: ...
- FR-2: ...

### Non-Functional
- NFR-1: ...

## Acceptance Criteria
- AC-1: Given ..., when ..., then ...
- AC-2: Given ..., when ..., then ...
- AC-3: Given ..., when ..., then ...

## Constraints
- Performance: ...
- Security: ...
- Compatibility: ...

## Edge Cases
- EC-1: ...
- EC-2: ...
- EC-3: ...

## Integration Points
- IP-1: ...

---
Phase 1 complete. Run `/sparc advance` to pass the gate check.
```
COMPATIBILITY & REFERENCES

Platform guidance

Imported from the listed AgentSkill.sh category. Complete supplied package files and notices are retained. Required external tools and credentials are not installed. No bundled code was executed during import. Individual agent task behavior has not been tested.