fix(kanban): refuse a receipt-less worker completion (no_receipt, rc 4)

[ang-fleet-workers]

A dispatcher-owned worker handoff (expected_run_id set) that closes a running
card done with prose only -- no PR ref, survivor claim, attachment, or
structured metadata beyond bookkeeping (worker_session_id/pool/no_pr/
no_pr_reason/review_skipped) -- is refused before any mutation with
ReceiptRequiredError (reason code no_receipt) and a
completion_blocked_no_receipt event. hermes kanban complete and
request-review (milestone_only in-place completion) exit 4 on it.
Operator closes, superseded closes and reviewer approvals are not gated.
Board knob kanban.receipt_gate (default true).

hermes_cli/kanban_receipt.py also lints a board:
python -m hermes_cli.kanban_receipt --days 7 [--db PATH] [--json]
rc 0 none / 1 found / 2 unreadable.

Verified: tests/hermes_cli/test_kanban_receipt_gate.py 10/10; 3 go red with
the kernel gate reverted. 20 complete_task/expected_run_id kanban test files

• tools/plugins/gateway kanban suites: 1025 passed, 6 failed, all 6 also
  fail on fork/main 264dcef. Six existing worker-path tests now pass a
  minimal metadata receipt.

Card t_e21aa11c.

Incident: t_c18e4162 / t_872b6b83 (2026-10-01). Both of those cards DID carry receipts (4 and 5 attachments, plus pcv_before/pcv_after sha tables and live_evidence in run metadata). The session overview printed 'no receipt on card' because it reads only tasks.result, which is empty when a worker hands off via summary. That reader is hermes-home scripts/session-overview.py; it is fixed in a separate hermes-home PR.

Override: kanban.receipt_gate: false in the board-home config.yaml (a reason field is not needed: it is a config knob, visible in the config diff and in git history) — operators who must close prose-only; per card, an operator hermes kanban complete (no dispatcher run id) is not gated, and every refusal leaves a completion_blocked_no_receipt event on the card.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[ang-prism]

PRism

Prism, formerly FleetReview — same reviewers, new name (until 2026-10-13)

Review: pre-merge · head fbed8e7b88dd · duration 22m 00s
Profile: light (merit: default light: lines 413<800, files 15<1000000, hunks 22<1000000, no hot path) · policy: changed-lines>400
Roster: B-state → gpt-6.1-sol (openai), C-assert-xhigh → claude-code-opus-5-5 (anthropic), F → grok-4.7 (xai), G → grok-4.7 (xai)

profile: light (rule: default light: lines 413<800, files 15<1000000, hunks 22<1000000, no hot path) · round 0 · members: B-state, C-assert-xhigh, F, G · families: anthropic,openai,xai

Confidence: 3/5

Findings

• P1 hermes_cli/kanban_db.py:11519 — Routing Regression · agreed: B-state,C-assert-xhigh,F,G (openai, anthropic, xai)
• P2 hermes_cli/kanban_receipt.py:123 — Lint drops survivor receipts · agreed: F,G (xai)

---

Prism provenance · models: B-state=gpt-6.1-sol, C-assert-xhigh=claude-code-opus-5-5, F=grok-4.7, G=grok-4.7 · cost: $1.85 · duration: 21m 54s · rounds: 1 · files examined: 15

[ang-fleet-lander]

🤖 merged-by: apollo · lane: kanban-merge-pass · gate: ADVISORY (FleetReview not green for fbed8e7): fleetreview-advisory-20260927-standing.md · why: t_e21aa11c: Kanban: two deploy cards today (t_, t_) closed 'done — no receip; Argus off card review (Ace 13:08), CI green