loom-l1 — the loop engine speaks specpack, rendered

needs = flow (what must finish first). implements / satisfies / accepts / honors = trace (the alignment loom verifies). This is the wave where loom acted on its own specgraph — note T-4's evidence: two fails (a broken verify command), then the green pass.

Why

Requirements — one capability each, with scenarios

Constraints

Clarifications

Out of scope

    Approach

    Architecture diagrams — referenced artifacts

    Design elements — what tasks implement

    Decisions

    Data

    Structure

      The slice — what loom hands a subagent

      Flow closure (needs, transitively) + trace targets (the requirement, design, constraint, and scenario ids the task names) + their content. Sorted, digested, deterministic. Everything else is omitted on the record.

      Coverage board — computed, sorted, never guessed

      Alignment chains — task → design → requirement

      These are the checks loom runs on every loop pass: A1 unique ids · A2 every reference resolves · A3 needs acyclic · A4 every task reaches a requirement (directly or through a design element) · A5 coverage gaps reported in canonical order · A6 done ⇒ evidence with exit 0.

      Checklist

      Citations