daily-fix: address-concern --summary-file + --dry-run alias

Workflow gap

Two small helper-CLI gaps each cost a real loss or a wasted turn on 2026-08-05:

1. task.py address-concern silently truncates long summaries. At 21:08:51Z (#1739 session) a 1,785-char disposition summary was cut to 196 chars: "[task.py] WARNING: --summary was 1785 chars (cap 200); truncated at a word boundary… Dropped tail: 'call ONE canonical helper in src/ … byte-exact (41.74958409920785), proving pure re-aggregation not re-measurement…'" — the durable concern-disposition record lost most of its evidence text, and the agent did not re-post the tail anywhere.
2. clean_experiment_downloads.py rejects --dry-run. At 21:38:33Z the same session got "error: unrecognized arguments: --dry-run" (rc=2) — the script's preview mode is the no-flag default and --apply opts in, but CLAUDE.md § Disk hygiene documents the sibling janitor as pod.py cleanup --all --dry-run, so the flag-shape confusion between the two janitors is predictable.

verified-at-filing (2026-08-06T07:2xZ): grep -n 'cap 200\|200)' scripts/task.py | head → the summary cap constant is live in the address-concern path; uv run python scripts/clean_experiment_downloads.py --help 2>&1 | grep -c dry-run → 0 (no such flag). Incident quotes are the miner's probed tool_result readbacks (session 2f4940f0 rows 894, 1040).

Proposed change

1. Add --summary-file <path> to address-concern (matching the post-marker --file convention), and make an over-cap --summary a hard error pointing at it — never a warn-and-truncate that silently drops the record's substance.
2. Accept --dry-run as an explicit alias of the default preview mode in scripts/clean_experiment_downloads.py (no behavior change to --apply).

Provenance

• fingerprint: 1f1f8f5420b3

• workflow_fix_target: scripts/task.py, scripts/clean_experiment_downloads.py

• origin: /daily 2026-08-05 problem sweep — miner 7 P12/P13 (both probed).