Skip to content

Switchyard lets LLM applications route traffic across models and providers while preserving native OpenAI and Anthropic API compatibility - enabling flexible model selection, benchmarking, and cost/performance optimization.

Rust

chore(config): enable LLM and codex_cli conflict-resolution tiers in sir-merge-a-lot config (#34)

nv-msvoboda08490b4a8805
diff --git a/.sir-merge-a-lot.yml b/.sir-merge-a-lot.yml
index cbb30b6f2..778293e2a 100644
--- a/.sir-merge-a-lot.yml
+++ b/.sir-merge-a-lot.yml
@@ -1,4 +1,4 @@
-# .sir-merge-a-lot.yml — switchyard onboarding (initial)
+# .sir-merge-a-lot.yml — switchyard
#
# Trigger mode: on_demand
# Meaning: the bot does NOTHING on PRs until explicitly invited via
@@ -16,10 +16,22 @@
# grooming), update this file via a normal PR + review.
trigger_mode: "on_demand"
-# Optional: GLAMR LLM resolver. Disabled for the initial onboarding —
-# switchyard team explicitly opts in by flipping this to true.
-# glamr:
-# enabled: false
+# LLM conflict resolver. Fires only when the deterministic ladder
+# (rerere -> Mergiraf) cannot resolve a conflict, and only on PRs the
+# bot was explicitly invited to groom (on_demand mode). Every LLM
+# resolution passes the bot's bounded-edit safety check; resolutions
+# touching lines outside conflict markers are rejected.
+llm:
+ enabled: true
+
+# Coding-agent fallback tier. When the raw-LLM resolver fails
+# (prompt too large for the LLM, or LLM timeout), the bot dispatches
+# the conflicted file to a sandboxed codex_cli run (via OpenShell)
+# instead of immediately escalating to a human. Trigger reasons,
+# timeout (600s), and per-MR attempt cap (1) use the bot's defaults.
+# Requires the SMAL worker deployment to be an image with the
+# OpenShell contract (AIREAPP-1853); on older images this is a no-op.
+coding_agent_provider: codex_cli
# Optional: merge-command allowlist. With on_demand mode, command-
# driven merges still require a commenter on this list (and standard