task: distill local-dream Android diffusion contracts with AREX repo-to-skill #234

Goal and acceptance

Deliver a bounded English repository-owned knowledge skill for xororz/local-dream using the portable AREX repo-to-skill workflow, preserving source evidence and negative boundaries rather than presenting a generic repository summary.

• Inspect immutable upstream source and record readsets, license constraints, mechanisms and unknowns.
• Deliver source-derived procedures for Android generation lifecycle, backend/model compatibility, native packaging and failure/evidence diagnosis; no native qualification claims.
• Integrate canonical owner/resources, stable registry, generated catalog/package access, discovery and the two existing matrices.
• Include offline negative/contract tests and validate relevant lint, tests, generator consistency and noneditable wheel access.
• Obtain independent revision-scoped review and applicable current-head checks before Ready/integration.

Scope and constraints

In scope: selected local-dream source contracts and their reusable audit/diagnostic procedures; source-pinned references and license notice; portable AREX-style repository analysis. Out of scope: six separately discovered Android/iOS repositories, native Android/Qualcomm execution, GPU/device runs, weights, paid services, quality/performance/fit/service qualification, new runtime targets, CI lanes or policy changes. Source/helper checks cannot promote a candidate: required native/model evidence remains not_run / INCONCLUSIVE.

Upstream LICENSE is CC BY-NC 4.0; preserve the noncommercial restriction and attribution, do not imply project-code or model/artifact redistribution compatibility. Write original procedural analysis with bounded source citations; do not vendor upstream code or assets. Referenced external model/native repositories are not automatically in the inspected readset.

Ownership and authorization

• Actual requester: repository owner via explicit 2026-10-29 instruction to use AREX-Skill repo-to-skill, create Issue then linked PR, then merge sequentially.
• Status: Accepted within that request; this record precedes implementation.
• Accountable parent: authorized Codex parent agent under @david6666666.
• Implementer: Codex / gpt-6.1-sol [effort=xhigh], bounded local-dream slice.
• Goal acceptance/integration: authorized parent after independent review and current-head checks; no author self-approval.

Inputs and deliverables

• InferRSI assessed main: c38a75b915ea2bd2a93a5f88782fa547139bd27e.
• local-dream master: db162df4d9bb6efc6595250423d8df26241f4c27.
• VectorSpaceLab/AREX-Skill workflow: ac3fe1afa80fb9a09775ecfb2b6cc3ba850a2db6.
• Expected artifacts: one canonical local-dream owner with skill/metadata/references, source inventory and contract tests; registry/catalog/discovery/matrix updates. Generated catalog uses an actual published content revision, not an invented future SHA.

Completion and closure

PR will use non-closing Refs linkage. The accountable parent assesses the whole goal, publishes the required closure summary and may close within the user's delegated sequential integration request. A merged knowledge PR is not a campaign PASS. Remaining uncertainty: required native/model/quality/memory/performance evidence remains unrun; upstream and artifact licensing need their own applicability review.