engineering-spec— agent skill

You are generating a structured engineering specification suitable for any project management tool (Jira, Linear, GitHub Issues, plain markdown, etc.).

Input

The user will provide one or more of:

1. A feature description in text.
2. An image of a design, spec, or existing ticket.
3. A conversation or mixed input describing what to build.
4. Output from a prior analysis skill (e.g., /mvp) with story candidates.

If the user provides prior analysis output, use the story candidates and feature breakdown as the basis. Do not re-analyze.

Determine Story Type

Based on the input, determine whether this is a backend or frontend story:

• Backend: API endpoints, database changes, business logic, server-side processing, integrations, queues, cron jobs. Prefix with BE:.
• Frontend: UI components, pages, user interactions, client-side logic, styling. Prefix with FE:.
• If the user explicitly states the type, use that.
• If unclear, ask.

Title Format

Start with BE: or FE: followed by a concise feature name (max 8 words after prefix).

Examples:

• BE: User Invitation Flow
• FE: Dashboard Analytics Page
• BE: Webhook Event Processing
• FE: Settings Profile Editor

Required Sections

Description

One paragraph (2-4 sentences). Explain:

• What is being built
• How users interact with it
• The high-level outcome

No filler. No implementation details. Just the what and why.

Acceptance Criteria

Organize criteria into logical groups. Each group has a bold category header as a top-level bullet with sub-bullets for specific, testable requirements.

Format:

- **Category Name:**
  - Requirement sentence.
  - Another requirement sentence.

- **Another Category:**
  - Requirement sentence.

Category rules:

• Group related requirements under a descriptive bold header.
• Every requirement must be standalone and testable.
• Include validation behavior, failure behavior, and edge cases.
• Include idempotency rules when applicable.

For BE stories — always include a Routes category if API endpoints are involved:

• Start with authentication requirements.
• List each endpoint: who calls it, method + path in inline code, what it does.
• Format: FE calls METHOD /path to [description].
• Include request behavior, response shape, and error behavior.

Example:

- **Routes:**
  - All endpoints require user authentication.
  - FE calls `GET /api/invitations` to list pending invitations for the current user.
  - FE calls `POST /api/invitations` to create a new invitation; returns the invitation object with a unique token.
  - FE calls `DELETE /api/invitations/:id` to revoke a pending invitation; returns 404 if already accepted.

For FE stories — include a UI Behavior category:

• Describe component behavior and interaction states (loading, empty, error, success).
• Reference API endpoints the FE will consume using inline code for paths.

Include a Business Rules category when applicable:

• Lifecycle states (draft, active, archived).
• Constraints (e.g., only one active per organization).
• Resolution or aggregation logic.

Dev Notes

Technical implementation guidance for the developer.

For BE stories:

• Schema/Database: State new or modified schemas/tables.
• Tables: List each table with columns:

  **table_name**
  - column_name (TYPE, modifiers) -- description
  - Indexes: ...
  - Foreign keys: ...

• Resolution/Processing Logic: Exact conditions, fallback order, behavior when preconditions are not met.
• Integration Points: Services, methods, or hooks that trigger behavior. Specify blocking vs async, idempotency mechanism.
• Concurrency Protection: Database-level protection, locking strategy, how race conditions are prevented.

For FE stories:

• Components: New components to create and existing ones to modify.
• State Management: What state is needed and where it lives.
• API Integration: Endpoints to consume with request/response shapes.
• Routing: New routes or route changes.

Strict Rules

• Match this format exactly. Do not invent new sections or rename existing ones.
• No vague language. No "handle properly", "etc.", or "as needed".
• No summarization or placeholders.
• Every requirement must be explicit and testable.
• API routes must include the full method and path in inline code backticks.
• Write as if implementation begins immediately after reading.
• If the input is an image, extract all visible text and structure before generating.
• If requirements are ambiguous, ask clarifying questions before writing the spec.

============================================================ SELF-HEALING VALIDATION (max 2 iterations) ============================================================

After producing documentation, validate completeness:

1. Verify all required sections are present and non-empty.
2. Verify internal cross-references and links resolve correctly.
3. Verify no placeholder text remains ("{TODO}", "[TBD]", "...", "etc.").
4. Verify code examples are syntactically valid.

IF VALIDATION FAILS:

• Identify which sections are incomplete or contain placeholders
• Re-generate only the deficient sections
• Repeat up to 2 iterations

Next Steps

After delivering the spec, suggest:

• "Run /arch-review with this spec to get architect-level feedback before implementation."
• "Or run /si to implement this spec directly in the current repo."

============================================================ SELF-EVOLUTION TELEMETRY ============================================================

After producing output, record execution metadata for the /evolve pipeline.

Check if a project memory directory exists:

• Look for the project path in ~/.claude/projects/
• If found, append to skill-telemetry.md in that memory directory

Entry format:

### /backend-spec — {{YYYY-MM-DD}}
- Outcome: {{SUCCESS | PARTIAL | FAILED}}
- Self-healed: {{yes — what was healed | no}}
- Iterations used: {{N}} / {{N max}}
- Bottleneck: {{phase that struggled or "none"}}
- Suggestion: {{one-line improvement idea for /evolve, or "none"}}

Only log if the memory directory exists. Skip silently if not found. Keep entries concise — /evolve will parse these for skill improvement signals.