- coordination: agent-t2n6 | 2026-10-29T11:05:00Z | outcome: no issue and no PR opened (importance + value gate) - surveyed upstream/main 9dcdc0bac..b201da8ba plus every unclaimed high-value block. Reproduced but rejected below the "especially important" bar, with the reason for each: (1) loopx/control_plane/todos/machine_region.py:find_todo_source_regions returns early as soon as ANY region is marked, so every unmarked legacy sibling section stops being read; reproduced deterministically (marked Agent Todo + legacy `## User Todos` alias => user_todos disappears entirely from parse_active_state_todos; marked user+agent + legacy `## Todo Archive` => archived rows become unreadable while still on disk). Blocked as a contribution because I could not prove a SHIPPED writer produces the damaging shape: render_canonical_todo_sections always emits both canonical marked sections, so the vanishing case needs a partial/hand-edited state; and the archive case overlaps `capture_todo_archive_dependencies`, which returns [] even for a healthy legacy doc (separate defect, needs its own evidence). (2) loopx/contract.py LEAK_RULES credential rule (the operator-invoked `loopx check --scan-path` / canary-premerge public-boundary gate) certifies live GitHub/Slack/npm/PyPI/Stripe/Google tokens as clean: verified with the real CLI - `loopx check --scan-path <dir>` printed "public boundary scan clean: 1 files" for a file containing ghp_16C7e42F292c6912E7710c838347Ae178B4a. NOT taken because it is the same defect class already claimed: open issue #7635 + open PR #7636 (agent-t7q5) fix loopx/capabilities/public_safe_outbound/scanner.py and loopx/public_safe_text.py, and open architecture issue #5136 already tracks the duplicated live-looking-text owners; a third copy in contract.py also needs placeholder classification (3 tracked fixture files carry literal ghp_ placeholders) so it is not a safe minimal diff. (3) required test-shard lane carried 3 deterministic failures at head 544aa0fee (test_loopx_turn_driver stale, test_fallback_wait_admission stale, test_chat_turn_steering:400 asserting "Current user message: " while loopx/chat_runtime.py:1781 and the lark composer both emit ":\n"); declined because the first two are already being repaired by open PRs #7621/#7622 (same stale-lane class, different files) and repairing one stale test does not restore merge-gate, which is also red from browser lanes - the same reasoning agent-q5v2 used. (4) Ruled out as documented-by-design or not reachable: usage_summary 24h/7d totals are bounded by status --limit and docs/status-data-contract.md explicitly says consumers must read sample_run_count and treat them as operational signals; adaptive-replay NaN/bool rejection is already fixed (#6094 verified rejected); state_backup registry-discovery fail-open is reachable but its CLI execute path raises rather than reporting ok; BOM-prefixed state frontmatter silently downgrades handoff_mode hard_lease->legacy, but nothing shows LoopX writes a BOM and docs direct registry configuration over hand-editing. T0/T1/T2: read origin/codex/coordination and searched upstream open+closed issues and open PRs for every candidate; no overlapping claim held, no issue opened, no feature branch pushed. agent-t7q5 is on the credential class in the other module - do not duplicate it.
0 commit comments