[Code-delta-support]

Adds a provider-fallback path to the chatbot so a failed primary call retries against the secondary provider. CodeDelta runs on this PR automatically and posts its report below.

[github-actions]

CodeDelta report

Churn: 96 logical statements (+6 / −0 / Δ90); REP_CHURN 0.06
TRUE_CHURN: 6 authored — the other 90 (94% of the total) came from 1 generated file (lockfiles, minified bundles, codegen output)

_{Dependency manifests changed (1): churn-demo/new/package-lock.json (modified) — a separate, quieter signal.}

• CRITICAL agent-scan-demo/analytics/julia/ops_helper.jl:7 — process launch near AI API call (line 7) — possible execution of AI-generated command
• CRITICAL agent-scan-demo/analytics/r/report_gen.R:5 — process launch near AI API call (line 5) — possible execution of AI-generated command
• CRITICAL agent-scan-demo/analytics/scala/src/main/scala/BatchOps.scala:7 — process launch near AI API call (line 7) — possible execution of AI-generated command
• CRITICAL agent-scan-demo/build-tools/AutoPatch.groovy:8 — process launch near AI API call (line 8) — possible execution of AI-generated command
• CRITICAL agent-scan-demo/edge/nginx/lua/dynamic_route.lua:5 — process launch near AI API call (line 5) — possible execution of AI-generated command
• …and 66 more CRITICAL/HIGH finding(s) in the agent report

Agent scan: 64 CRITICAL · 7 HIGH · 81 ELEVATED · 17 endpoint-level · 61 text-level · 2 binary
Governance: 14 file(s) send data to foreign-hosted models (14 non-allied); 2 file(s) call AI inside a loop (cost risk)
Agent infrastructure: 16 artifact(s) in the tree — 4 tier-3 (rogue residue / credentials): agent-scan-demo/agents/openclaw/gateway.auth.token, safety-demo/new/agents/openclaw/gateway.auth.token, agent-scan-demo/agents/openclaw, safety-demo/new/agents/openclaw
🔑 Committed credentials: 2 key-format match(es) in the tree — agent-scan-demo/agents/openclaw/deploy_key.pem:1, safety-demo/new/agents/openclaw/deploy_key.pem:1 (values redacted; rotate first, then remove)

_{AI signals are pointers for review, not verdicts. Full report attached to this run. CodeDelta 2.3.2.}