the-agency-ai/src/claude/skills/define

<!-- Flag #62/#63: allowed-tools removed. Inherits Bash(*) from .claude/settings.json. Restricting to specific subcommand patterns at the skill level silently blocks agents on permission prompts the agent cannot see — see dispatch #171 for the devex incident that surfaced this trap. -->

Define — PVR Completeness

Drive toward complete Product Vision & Requirements (PVR) using the 1B1 protocol with a completeness checklist.

Usage

• /define — start fresh, auto-detect project
• /define "Project Name" — start fresh for named project
• /define --from <seed-file> — extract initial requirements from a seed document
• /define --continue <pvr-file> — resume work on an existing PVR

Behavior

Step 1: Establish context

If --from specified, read the seed file. If --continue, read the existing PVR. Otherwise, ask what the project is about.

Step 2: Scan for existing work

Glob agency/workstreams/*/pvr-*.md for existing PVR files. If found, offer to continue from the most recent one.

Step 3: Build discussion agenda

Map the completeness checklist to discussion items. Skip items already covered in an existing PVR.

Step 4: Start discussion

Invoke /discuss via the Skill tool, passing the agenda items. The 1B1 protocol handles the resolution cycle for each item.

Step 5: Update PVR progressively

After each item resolves, update the PVR file. Don't batch writes to the end.

PVR location: agency/workstreams/{workstream}/pvr-{workstream}-{slug}-{YYYYMMDD}.md.

Step 6: Present completeness scorecard

After all items are discussed, show the checklist with status:

PVR Completeness:

1. Problem Statement       ✓ Complete
2. Target Users            ✓ Complete
3. Use Cases               ✓ Complete
4. Functional Requirements ✓ Complete
5. Non-Functional Reqs     ~ Partial (performance TBD)
6. Constraints             ✓ Complete
7. Success Criteria        ✓ Complete
8. Non-Goals               ✓ Complete
9. Open Questions          2 remaining

Score: 8/9 complete

Completeness Checklist

1. Problem Statement — What problem does this solve? For whom? Why now?
2. Target Users — Who uses this? What are their characteristics?
3. Use Cases — Primary workflows. What does the user do?
4. Functional Requirements — What must the system do?
5. Non-Functional Requirements — Performance, scalability, reliability, accessibility
6. Constraints — Technical, business, regulatory, timeline
7. Success Criteria — How do we know it worked? Measurable outcomes.
8. Non-Goals — What we explicitly won't do (prevents scope creep)
9. Open Questions — Unresolved items requiring research or decisions