ll-product-epic-to-feat

name	ll-product-epic-to-feat
description	Governed LL workflow skill for transforming a frozen epic_freeze_package into a lite-native feat_freeze_package between ll-product-src-to-epic and downstream FEAT design and QA derivation flows.

LL Product EPIC to FEAT

This skill is a lite-native governed workflow between ll-product-src-to-epic and downstream FEAT consumers. It runs directly through the local scripts/epic_to_feat.py runtime instead of delegating execution back to the legacy lee run stack.

Canonical Authority

• Workflow template: E:\ai\LEE\spec-global\departments\product\workflows\templates\epic-to-feat\v1\workflow.yaml
• Upstream handoff: ll-product-src-to-epic
• Downstream workflows:
  • workflow.dev.feat_to_tech
  • workflow.qa.feat_to_testset
• Downstream consumer ordering: After FEAT freeze, when design_impact_required=true, consumers MUST run ll-dev-feat-to-surface-map first (design ownership layer), then run ll-dev-feat-to-tech and ll-dev-feat-to-proto in parallel.
• Derived child artifacts expected downstream: TECH, TESTSET
• Preferred runtime command: python scripts/epic_to_feat.py run --input <epic-package-dir> --repo-root <repo-root>

Runtime Boundary Baseline

• Interpret this workflow using E:\ai\LEE-Lite-skill-first\ssot\adr\ADR-038-运行时核心抽象边界与对象分层基线.MD.
• This capability is a governed Skill and Workflow for EPIC -> FEAT derivation. Output FEAT bundles are artifacts; they do not replace runtime task, session, or gate authority.

Required Read Order

1. ll.contract.yaml
2. input/contract.yaml
3. output/contract.yaml
4. resources/upstream-workflow-analysis.md
5. agents/executor.md
6. agents/supervisor.md
7. input/semantic-checklist.md and output/semantic-checklist.md

Execution Protocol

1. Accept only a freeze-ready epic_freeze_package emitted by ll-product-src-to-epic.
2. Validate the package structurally before deriving or editing any FEAT output.
3. Resolve the authoritative EPIC context from epic-freeze.md, epic-freeze.json, inherited source_refs, and the upstream acceptance evidence.
4. Run python scripts/epic_to_feat.py executor-run --input <epic-package-dir> to generate the governed FEAT bundle.
5. Collect execution evidence, then hand the governed FEAT bundle to the supervisor.
6. Run python scripts/epic_to_feat.py supervisor-review --artifacts-dir <feat-package-dir> before freeze.
7. Freeze only after the supervisor records a semantic pass and python scripts/epic_to_feat.py freeze-guard --artifacts-dir <feat-package-dir> returns success.
8. Emit a downstream handoff that preserves epic_freeze_ref, src_root_id, feat_refs, integration_context_ref, and the downstream workflow list for governed TECH and TESTSET derivation.
9. When external gate returns revise or retry, rerun run, executor-run, or supervisor-review with --revision-request <revision-request.json> so the regenerated FEAT bundle carries normalized revision context.

Workflow Boundary

• Input: one epic_freeze_package that is freeze-ready and traceable to ll-product-src-to-epic
• Output: one feat_freeze_package with a FEAT inventory, governance evidence, and downstream handoff metadata
• Out of scope: direct TECH authoring, TASK planning, TESTSET authoring, or bypassing the governed EPIC to FEAT boundary

Non-Negotiable Rules

• Do not accept raw requirements, SRC candidate packages, or informal EPIC markdown directly.
• Do not bypass scripts/epic_to_feat.py by hand-authoring only the final FEAT bundle without execution and supervision evidence.
• Every emitted FEAT must remain an independently acceptable capability slice, not an implementation task, screen TODO list, or architecture-only note.
• Preserve epic_freeze_ref, src_root_id, authoritative source_refs, and ADR-025 acceptance semantics.
• Downstream readiness must stay explicit: FEAT output must be strong enough to seed TECH and TESTSET derivation without re-deriving the parent EPIC.
• Always materialize integration-context.json as a downstream seed for ll-dev-feat-to-tech; do not force TECH derivation to guess integration facts from FEAT prose alone.
• Do not let the executor self-approve the semantic validity of its own FEAT output.