compose-brownfield-prd

Maps characterize-brownfield output into the exact HVE-format PRD (17 sections) that HVE agents consume directly — never re-scans the repo, never creates an issue.

When to use

• After characterization, to produce the HVE-format PRD
• Loaded internally by brownfield-analyst (disable-model-invocation), or directly if artifacts already exist
• “write the PRD”, “compose the PRD from characterization”

Entry contract

• Path to characterization artifacts (index.md + siblings)
• Product name (kebab-case filename)
• Modernization goals (optional)

Exit contract

• docs/prds/<kebab-case-name>.md — 17 sections, FR-001/NFR-001 IDs, markers, no YAML frontmatter
• State file: prd-sessions/<name>.state.json (currentPhase: brownfield-extraction)

Invariants

• Reloads the artifacts (B4) — never re-scans the repo, never relies on the agent’s recall
• Traceability — every FR links to a Goal ID from Section 1
• Never fabricates a PASS — NFR with no measured evidence → default CONCERNS (Status/Threshold/Actual/Evidence shape)
• Schema gate (S4) — 17 headers in order, unique IDs, no YAML frontmatter, markers present
• Every Low-confidence claim and every NONE-coverage Core feature → one S14 Open Questions row

Why this shape

The PRD applies an explicit schema gate before writing — 17 ordered sections, unique IDs, every FR traced to a goal — so HVE agents consume it without ambiguity.

« Define explicit review criteria before the review begins. » — Wiegers, K., Peer Reviews in Software, 2002.

Allowed customisation

• Modernization goals folded into Section 1 (otherwise derived from tech-debt.md signals)
• Conditional sections (Data & Analytics, Rollout) per characterization signals

See also

• characterize-brownfield — Produces the artifacts consumed here
• brownfield-analyst — Agent that chains characterization → composition
• Brownfield — Family overview