[scan-fix] Phase 8: LLM providers / usage / billing

[rennf93]

Phase 8 of the 9-phase serial codebase-scan fix program. Stacks on #321 (Phase 7). CEO merges bottom-up — do not merge until #321 lands.

Findings fixed (8)

• H14 — _apply_grok enables the seeded GROK provider row so resolve_for_agent routes to GrokCliProvider (SuperGrok auth), not just sets the global default. roboco/services/llm.py
• M31 — _resolve_active_tokens routes GROK to usage.json so the live USAGE_SNAPSHOT sweep reflects grok agents mid-run, not only at finalize. orchestrator.py
• M32 — usage sweep passes cache read/write tokens to calculate_cost and tracks them in _usage_by_agent, so live cost reflects Anthropic cache spend (matches the finalize path). orchestrator.py
• M33 — Ollama-Cloud rate-limit parking via a _RATE_LIMIT_MARKERS_BY_PROVIDER / _OVERLOAD_MARKERS_BY_PROVIDER marker map; a glm-5.2:cloud ollama.com 429 parks the provider instead of crash-respawning into the weekly limit. Grok deliberately absent (exit-75 detector). orchestrator.py
• M34 — _reconcile_orphan_spawn_sessions at startup closes crash-orphaned agent_spawn_sessions rows so their tokens roll into usage/cost summaries. orchestrator.py
• L12 — revisit_resets column (migration 067, additive) + persist/restore in the respawn tracker, mirroring tracing_resets, so the PM-respawn breaker's revisit counter survives a restart. tables.py + orchestrator.py
• L18 — Sonnet-5 promo date-gate: _sonnet5_prices() returns the promo tuple <= 2026-08-31, list rates after; Sonnet-5 row removed from _PRICING. roboco/billing/pricing.py
• L20 — grok_cli_usage.main() warns on run-log parse failure (lazy %-formatting) instead of silent zero-attribution; swallow preserved so finalize can't crash on one bad log. roboco/llm/providers/grok_cli_usage.py

Gate (green)

• uv run ruff format . — 2 files reformatted (trailing hygiene commit fbfcf20b)
• uv run ruff check . — clean
• uv run mypy roboco/ — 358 files, no issues
• ROBOCO_TEST_DB_PORT=55432 ROBOCO_TEST_DB_USER=renzof uv run pytest — 12025 passed, 418 skipped, 95% coverage (11 new tests over Phase 7)

Review

Per-task sonnet reviews all Approved (Tasks 1–9). Opus whole-branch review: SHIP, no fix wave. All binding constraints hold (signed commits, no new # type:ignore/# noqa, AGPL-3.0, migration 067 additive with down_revision=066_tasks_source_status_idx).

Notes for the CEO

• Migration 067 (revisit_resets) ships here — run alembic upgrade head after merge.
• Sonnet-5 promo time bomb (follow-up, not a blocker): the pre-existing TestSonnet5PromoTier class asserts promo rates with no date monkeypatch and will fail on/after 2026-09-01. Plan-mandated; the two new tests pin both sides of the gate. Worth a separate follow-up to monkeypatch or delete those 7 tests before the promo expires.
• Revision-id shortening (Task 6): 067_respawn_tracker_revisit_resets (34 chars) → 067_respawn_revisit_resets (26 chars) because alembic_version.version_num is VARCHAR(32). down_revision unchanged.

🤖 Generated with Claude Code

[github-actions]

Thanks for opening your first pull request on RoboCo!

Quick checklist before review (most of these are enforced by CI, but worth a glance):

• make quality — ruff format check, ruff check, mypy, pytest (≥80% coverage), and the rest of the gate
• Panel changes pass pnpm lint and pnpm exec tsc --noEmit (run from panel/)
• No # noqa / # type: ignore shortcuts; pre-existing violations in touched files are fixed
• Added an entry under ## [Unreleased] in CHANGELOG.md
• Signed the CLA (the bot will prompt you on this PR)
• Signed your commits — master requires verified signatures (SSH signing setup)
• Updated any affected docs under docs/

See CONTRIBUTING.md for the full workflow and the Code of Conduct for the community standards we follow.

Welcome aboard — a maintainer will review shortly.