plan-feature

Plan Feature Skill

[!IMPORTANT] Plan a feature from BRD-lite brief or clear intent into PRD (What), decisions, implementation plan, and task slices.

Optional args: slug=, ticket=<id/url>, mode=interactive|autonomous|channel, channel=, auto_continue=true|false, profile=business|hybrid|technical.

Instructions

When the user asks to perform this workflow, execute the following steps:

Feature Planning Workflow (PRD / What)

Goal: Produce a PM-owned decision-complete PRD, delivery plan, and IT Department handoff before code starts.

Steps

1. Load context:
   • Load baseline PRD section, task-linked evidence, and canonical ledger (docs/product/<slug>/evidence.json) with stable slug; when intent is unclear, ask which matching ticket/brief is authoritative rather than selecting an unrelated recent document; link strategy, OPP, DEC, critical assumption, and MET refs without duplicating records or minting product IDs as requirement IDs.
   • Explicitly load product-management/product-requirements and its on-demand references/requirements-modeling.md; report truthfully if unavailable.
   • Load product-management/product-business-requirements when BRD-lite context applies; also load ticket text, existing specs, repo patterns, common-operator-profile, common-decision-discipline, quality-engineering-business-analysis, and matched framework skills.
   • Carry forward operator_profile, lane, snc_tier, approval, BRD objective IDs, SMART metric, scope fence, assumptions, evidence ledger, glossary, risks, delivery context. Equivalent approved evidence suffices without separate BRD/PRD/SRS filenames; reused approved context skips rediscovery.
   • When a PRD relies on captured stakeholder meaning, load the applicable BRD confirmation procedure at product-management/product-business-requirements/references/analysis-planning-and-elicitation.md; use requirements-modeling.md as the sole owner of conflict, quality-versus-need, and transition semantics.
2. Interview:
   • Draft provisional PRD direction from current context before asking; ask only for business logic, scope, constraints, and acceptance criteria that cannot be inferred.
   • Ask max 3 blocking decisions at a time with recommended default and 2-3 options; treat non-critical unknowns as explicit assumptions.
   • Confirm problem statement, assumptions, target users, JTBD/use cases, platforms, data, analytics, security, performance, rollout, and non-goals.
   • Confirm each requirement has owner, priority, and status; confirm success metrics and guardrails that must not regress.
   • Carry legitimate no-build, defer, buy, substitute options, negative evidence, and permission boundaries; brief approval alone confers NO budget, exposure, pilot, or rollout consent.
   • Stop when requirements are actionable.
3. Draft PRD:
   • Save to docs/prd/prd-[slug].md when file writes are allowed.
   • Keep "what" separate from "how"; consequential architecture/contracts route to design-solution.
   • Add stable REQ-* and AC-* IDs; link each requirement back to BRD-lite objective. One observable behavior per AC with necessary preconditions; truth tables for interacting conditions.
   • For substantial scope, build capability inventory (verb-object, in-scope/deferred/excluded); one containment parent per node, shared rules referenced without duplicate definitions.
   • Use Given/When/Then AC when behavior can be misread; check user stories for persona, value, INVEST, happy/edge/negative paths.
   • Select views by question (process flow=order, decision table=rules, state=lifecycle) per requirements-modeling.md; prose/tables allowed, no mandatory UML.
   • Include risk categories, rollout/ops, decision log, analytics/telemetry, changelog. Mark unresolved blocking product decisions as blockers (name slice, consequence, owner, input).
   • Include RACI table for BA, PM, architect, backend, frontend, mobile, QA, release, and business/UAT approver when more than one delivery role is involved.
4. Create implementation plan:
   • Define product outcomes, ACs, dependencies, delivery risks, and provisional owner-aligned slices.
   • Keep unresolved architecture/contracts explicit for design-solution; do not author technical design twice in PM plan.
   • Map each provisional task slice to requirement IDs, AC IDs, likely owner role, repo/module, expected artifact, and verification lane.
   • Estimate each slice (t-shirt size + confidence); roll up a delivery-window range in time/cost terms for operator_profile=business, points otherwise. Identify whether design-solution is required before coding.
5. Route:
   • Continue when assumptions are non-critical; return BLOCKED for missing owner, untestable AC, approval, or release constraint.
   • Empirical hypothesis testing needed -> validate-product (when selected/installed; else record capability gap, keep hypothesis as unvalidated assumption, and route to implementation-readiness); architecture/contracts unclear -> design-solution; approved plan with trace and testable ACs -> implementation-readiness.