Product Capability

Transform a product requirement or feature request into a capability specification: a structured technical document that bridges product intent and engineering implementation.

Distinct from create-prd:

• create-prd → what and why (Jobs-to-be-Done, personas, scope). Audience: PMs, stakeholders.
• product-capability → what + how (acceptance criteria, task breakdown, traceability). Audience: engineers building it.

When to Use

• Engineering needs a clear, testable spec to implement from — not a narrative PRD
• Acceptance criteria must be explicit and verifiable before a ticket is created
• Traceability is required: feature request → capability → AC → implementation tasks
• A PRD exists but lacks the implementation detail engineers need to estimate or build

When NOT to Use

Instead of product-capability	Use
Exploring the problem space and options	opportunity-solution-tree
Writing the product story for stakeholders	create-prd
Prioritizing what to build next	feature-prioritization
Planning launch readiness	launch-strategy

Structure

A capability spec consists of five parts:

1. Capability Statement

One sentence: what the system must be able to do.

The system must allow authenticated users to export their activity history
as a CSV file filtered by date range.

Write as system capability, not user story. Avoid "As a user, I want…" — that belongs in the PRD.

2. Context

• Source requirement (link to PRD, issue, or ticket)
• Affected components (which services, modules, APIs)
• Dependencies (what must exist before this can be built)

3. Acceptance Criteria

Each criterion must be:

• Testable: a QA engineer can write a test case for it without ambiguity
• Specific: names exact behavior, not intent
• Binary: pass or fail — no "should generally work"

**AC-01**: Given a user with `EXPORT_HISTORY` permission, when they request
  an export for date range [start, end], the system returns a valid CSV within
  5 seconds containing all activity records in that range.

**AC-02**: Given an empty result set, the system returns an empty CSV with
  only the header row, not an error.

**AC-03**: Given a date range exceeding 1 year, the system rejects the request
  with HTTP 422 and error code `DATE_RANGE_TOO_LARGE`.

**AC-04**: Exported CSV rows include: timestamp (ISO 8601), action_type,
  resource_id, user_id, ip_address.

4. Implementation Task Breakdown

Decompose into discrete, estimable engineering tasks. Each task:

• Belongs to one team/component boundary
• Is completable in ≤ 1 day
• Has a clear definition of done

| Task ID | Task | Component | Estimate | Depends on |
|---------|------|-----------|----------|------------|
| T-01 | Add `EXPORT_HISTORY` permission to auth schema | Auth | 2h | — |
| T-02 | Implement date-range validation (max 365d, ISO 8601 parse) | API | 2h | — |
| T-03 | Implement CSV generation service (streaming for large sets) | API | 4h | T-01 |
| T-04 | Add `/api/activity/export` endpoint with pagination | API | 3h | T-02, T-03 |
| T-05 | Frontend: date range picker + export trigger + download | UI | 4h | T-04 |
| T-06 | Integration tests (AC-01 through AC-04) | QA | 3h | T-04 |

Track tasks in SQL:

CREATE TABLE IF NOT EXISTS capability_tasks (
    id TEXT PRIMARY KEY,
    capability TEXT,
    task TEXT,
    component TEXT,
    estimate_hours REAL,
    status TEXT DEFAULT 'pending',  -- pending | in_progress | done | blocked
    depends_on TEXT
);

5. Traceability Matrix

Links requirement → capability → ACs → tasks:

| Requirement | Capability | ACs | Tasks |
|-------------|-----------|-----|-------|
| PRD §3.2: Data export | CSV activity export | AC-01, AC-02, AC-03, AC-04 | T-01–T-06 |

Quality Gate

Before handing off to engineering, verify:

• Capability statement is one sentence and testable
• Every AC is binary (pass/fail) with no ambiguous "should"
• Every AC has a corresponding task that implements and verifies it
• No task exceeds 1 day estimate (split if larger)
• All dependencies identified (APIs, permissions, schemas)
• Traceability matrix complete

Anti-Patterns

Anti-pattern	Fix
AC that says "the UI should be intuitive"	Replace with specific interaction criterion
Task that says "implement the feature"	Break into component-level subtasks
No dependency map	Add depends_on to every task that has a prerequisite
Mixing AC (behavior) with implementation detail	AC describes what, tasks describe how

See Also

• create-prd — JTBD-driven product requirements document
• feature-prioritization — MoSCoW/RICE scoring before spec
• tdd-workflow — write tests from ACs
• agentic-engineering — agent-friendly task decomposition