chore(config): enable LLM and codex_cli conflict-resolution tiers in sir-merge-a-lot config

[nv-msvoboda]

What

Updates .sir-merge-a-lot.yml to enable the merge bot's LLM and coding-agent conflict-resolution tiers, while keeping grooming strictly on-demand:

• trigger_mode: "on_demand" — unchanged. The bot takes no action on any PR until explicitly invited via @sir-merge-a-lot groom. No autonomous grooming, no autonomous merging.
• llm: enabled: true — makes LLM conflict resolution explicit. Replaces the deprecated commented-out glamr: block, whose comment incorrectly claimed the resolver was disabled (the bot's code default is enabled). The LLM tier fires only when the deterministic ladder (rerere → Mergiraf) cannot resolve a conflict, and every resolution passes the bot's bounded-edit safety check.
• coding_agent_provider: codex_cli — opts into the coding-agent fallback tier: when the raw-LLM resolver fails (prompt too large, or LLM timeout), the conflicted file is dispatched to a sandboxed codex_cli run instead of immediately escalating to a human. Trigger reasons, timeout (600s), and per-MR attempt cap (1) are intentionally omitted — they match the bot's built-in defaults exactly.

Why

Follow-up to the sir-merge-a-lot on-demand onboarding: enables the conflict-resolution ladder's LLM and codex_cli tiers so invited grooms can resolve conflicts the deterministic tools can't, while the bot retains zero autonomous authority.

How tested

• uv run ruff check . clean — N/A, YAML-only config change
• uv run mypy switchyard clean — N/A, YAML-only config change
• uv run pytest tests/ green — N/A, YAML-only config change
• Manual smoke: parsed this exact file with sir-merge-a-lot's load_config and the strict resolver-layer hydrator (coding_agent_config_from_smal_config, which raises on any misconfiguration). Zero warnings; hydrates to trigger_mode=on_demand, llm_enabled=True, and CodingAgentConfig(provider=codex_cli, trigger_reasons={llm_timeout, prompt_too_large_for_llm}, timeout=600s, max_attempts_per_mr=1).

Checklist

• One class per file — N/A, no code changed
• Public symbols exported — N/A, no code changed
• Unit tests — N/A, no code changed
• README / --help — N/A, no customer-facing surface changed
• Commits signed off (Signed-off-by) per the DCO.

Notes for reviewers

The codex_cli tier is a silent no-op until the SMAL worker deployment polling this repo runs the nemo-agent-hub image with the OpenShell contract (AIREAPP-1853) — verify the worker image before expecting coding-agent dispatches. Rollback at any time is deleting the two new blocks (or the file) via a normal PR.

🤖 Generated with Claude Code