[DIC-28] crux-garden --write: persist gardening report to docs/status/generated/gardening_reports/

[an0mium]

Slice

Adds --write and --output-dir flags to the aragora crux-garden CLI so the GardeningReport is persisted as a timestamped JSON file under docs/status/generated/gardening_reports/ (or a caller-supplied directory), satisfying the acceptance criterion in issue #6222 that was unmet by the stdout-only implementation. Mirrors the DIC-26 --write slice pattern (DIC-26 PR #8288 explicitly deferred DIC-28 persistence as a separate slice in its "Out of scope" section).

Gating

• Default: OFF (flag: ARAGORA_CRUX_GARDENING_ENABLED=1 required; --write must be passed explicitly)
• Live queue effect: none — purely additive file-write on the read-only report path; no debates started, no queue writes
• Advances issue: [DIC-28] Proactive Crux Gardening — scheduled re-examination of resolved and outstanding cruxes #6222 (DIC-28 Proactive Crux Gardening)

Tests

New in tests/cli/test_dic28_crux_garden.py (4 new; 14 pre-existing = 18 passed total):

Test	What it verifies
test_write_flag_creates_report_file	--write produces exactly one gardening_report_*.json in the output dir
test_write_flag_report_json_is_valid	written file is valid JSON with generated_at, resolved_results, outstanding_results, schema_version
test_write_flag_creates_output_dir_if_absent	--write creates the full directory tree when absent (3-level nested path)
test_write_flag_off_leaves_no_file	without --write, no report file is created; output dir untouched

Also adds a yaml stub at the top of the test file (same pattern as DIC-21/23/25) so tests run cleanly in the hermetic uv-tool pytest venv where pyyaml is absent.

Validation

• pytest tests/cli/test_dic28_crux_garden.py --noconftest — 18 passed in 0.16 s
• ruff check aragora/cli/commands/dic28_crux_garden.py tests/cli/test_dic28_crux_garden.py — clean
• mypy aragora/cli/commands/dic28_crux_garden.py --ignore-missing-imports — Success: no issues found
• Net diff: 119 LOC across 3 files (21 command, 16 parser, 82 test)

Out of scope

• Writing to the default docs/status/generated/gardening_reports/ path on CI (operator trigger decision, not a code gate)
• DIC-17 follow-up bridge wiring from the --write output path (tracked separately)
• Automated scheduling of periodic gardening runs (DIC-28 scheduling concern, separate slice)
• DIC-25 fragility-delta integration into gardening results (the GardeningConfig.fragility_shift_threshold hook exists; wiring it to a live DIC-25 catalog requires the Foreman gate to open)

Gate status

Per docs/status/NEXT_STEPS_CANONICAL.md, DIC-13..22 (and by extension DIC-23..28) are in the Delay track. The proof-first Foreman gate has not opened. This PR is planning truth and flag-gated operator tooling only; no boss-ready label applied. Kept as draft.

---

Generated by Claude Code

[scarmani]

Queue-drain close: closing this stale ownerless vision-incubator draft to reduce backlog pressure.

Live checks before closing:

• PR [DIC-28] crux-garden --write: persist gardening report to docs/status/generated/gardening_reports/ #8518 is still a draft at exact head b530c7b on branch vision-incubator/dic-28-crux-garden-write.
• No comments/reviews/dissent were present, and no operator-steering messages were pending for the claimed close lane.
• No active prior owner matched the PR or branch before the queue-drain lane claimed it.
• The branch adds a DIC-28 CLI feature path, so this loop did not repair, evidence-route, mark-ready, or merge it.

This does not delete the branch or discard the commits; the branch is preserved for deliberate revival if the crux-garden write mode should continue.