Strengthen /squad research into a real spec artifact #1756

Why this matters

Research is load-bearing and already the strongest step in the SDLC chain. In a live end-to-end run against a real .NET repo it independently caught:

• EntryItem has no ID field — Link is the de facto key
• Dead, unwired EF Core / SQL package references in Frontend.csproj
• That reusing the existing FeedSource blob provider avoids introducing a new AppHost storage resource
• A self-contradiction in the user's own epic text

That is genuine analytical value, not summarization.

The problem

/squad plan consumes the latest research artifact as PRIMARY context. But the only quality floor on research output today is a ≥200 character length check. Weak research therefore silently degrades every downstream step — plan, validate, implement — with nothing to catch it.

Gaps toward being a real spec

The current output is prose. To function as a spec that later steps can cite, it needs:

• Evidence/source table — every claim tied to a file, line, or command output
• Explicit goals and non-goals
• Load-bearing assumptions section — what invalidates the analysis if untrue
• Open decisions list — things a human must resolve
• Acceptance-test framing — how we would know the work succeeded
• Traceability IDs so /squad plan and /squad plan validate can reference specific findings

Audience

This is aimed squarely at the SpecKit / spec-driven-development crowd. The value proposition is that Squad produces a real, reviewable spec artifact — not that it merely dumps context into the next prompt.

Where the code lives

workflows/squad.md, the ## skill: squad-research block (~L569).

⚠️ Never put a ## heading inside a ## skill: block. The block closes implicitly at the next H2, and everything up to the next skill marker is silently discarded — this dropped ~29 KB once and still compiled clean. Skills use #####. Run npx vitest run test/gh-aw-quality.test.ts before pushing; it has a zero-discarded-bytes assertion that catches this.

Origin

User directive, 2026-08-19 — recorded in .squad/decisions/inbox/coordinator-research-is-load-bearing.md.

Scope Narrowed — Structural Contract Only (2026-08-20 triage)

Procedures' triage verdict: ship the structural contract pre-E2E; defer insight tuning until E2E results exist.

In scope: Required sections, Rn traceability IDs, citation token per evidence row, replacing the ≥200-char floor at workflows/squad.md:605.

Out of scope (→ #1757, deferred): The judgment half — "a well-formatted bad plan should FAIL." That requires a good/bad golden corpus we don't have until after the E2E series. Do not conflate these axes.

Goal and Success Criteria

Goal: Replace the ≥200-char floor at workflows/squad.md:605 with a structural schema that asserts the research artifact's required sections are present, so downstream /squad plan can reference specific evidence entries by traceability ID.

SC-1 — Character-count floor removed.
workflows/squad.md:605 (or its replacement) no longer enforces ≥200 chars as the sole quality check on research output. Fails if the only quality gate on research artifacts is still a character count.

SC-2 — Required sections declared in the skill spec.
The squad-research skill block explicitly requires these five elements: (a) an evidence table with at least one row, each row containing a source citation token; (b) a goals-and-non-goals section; (c) a load-bearing assumptions section; (d) an open decisions section; (e) at least one Rn traceability ID. Fails if any of those five elements is not named as required in the skill definition.

SC-3 — Test asserts section presence structurally.
test/gh-aw-quality.test.ts contains a test that parses a research artifact fixture and asserts each of the five required sections is present as a named heading — one assertion per section. Fails if the only coverage is a substring check or a single combined assertion. A substring check for R1 does not satisfy this criterion.