Route hourly LLM-calling workflow through contextual-orchestrator (orchestrator/free) in `.github/workflows/hourly-product-development.yml`

[seonghobae]

What

.github/workflows/hourly-product-development.yml in this repository makes its LLM call directly against a provider
(NVIDIA NIM), bypassing the org's shared contextual-orchestrator gateway
and its orchestrator/free fail-closed zero-cost pool entirely.

OpenCode's generated opencode.json hardcodes "model":"nvidia/nemotron-3-super-120b-a12b" with "provider":{"nvidia":{"options":{"baseURL":"https://integrate.api.nvidia.com/v1","apiKey":"{env:NVIDIA_API_KEY}"}}} — a direct NVIDIA NIM call with zero reference to orchestrator/, a sidecar, or any contextual-orchestrator gateway.

Why this is a violation, not a judgment call

• ADR-0003 in ContextualWisdomLab/.github scopes itself explicitly to
  that repo's three central review pipelines (OpenCode autofix/dispatch,
  Noema, Strix) — it does not authorize this repository's own workflow to
  call a provider directly, and does not mention this repository at all.
• The org owner has directed that all GitHub Actions workflows that make
  an LLM call route through orchestrator/free.
• ContextualWisdomLab/contextual-orchestrator's own hourly loop
  (opencode-hourly-loop.yml) already demonstrates the pattern this
  repository's loop should follow: vendor the pinned
  contextual-orchestrator sidecar, register the five provider credentials
  (BYTEZ_API_KEY, NVIDIA_NIM_API_KEY, NVIDIA_NIM_API_KEY_SUB,
  OPENROUTER_API_KEY, OPENAI_API_KEY) into its KV, serve the loopback
  gateway, and point OpenCode's model at
  contextual_orchestrator_gateway/orchestrator/free instead of a direct
  provider id/base URL. See
  ContextualWisdomLab/contextual-orchestrator/pull/1013 for a worked
  example of the pool-string half of that pattern (that PR does not add the
  gateway to a repo that lacks it; this repo's workflow doesn't have the
  gateway wired in at all yet, which is the larger piece of work here).

What "fixed" looks like

1. Vendor/provision the contextual-orchestrator sidecar in this workflow
   (mirroring .github's scripts/ci/contextual_orchestrator_review_sidecar.sh
   or contextual-orchestrator's own opencode-hourly-loop.yml job).
2. Register the five provider secrets into the sidecar's KV instead of
   passing NVIDIA_API_KEY/NVIDIA_NIM_API_KEY straight to the direct
   provider call.
3. Point the OpenCode config's model (and any OPENCODE_MODEL/
   OPENCODE_MODEL_CANDIDATES/enabled_providers override) at the gateway's
   orchestrator/free virtual id instead of a direct nvidia/nvidia-nim
   provider id or integrate.api.nvidia.com base URL.
4. Retire the repo-local NIM broker/proxy/bridge process this workflow
   currently starts, if one exists, once the gateway sidecar replaces it.
5. Add or update a workflow-contract test (shell-syntax or string-pinning,
   matching this org's usual pattern) covering the changed file, and run it.

This is intentionally left as a tracking issue rather than a same-pass fix:
each of these hourly loops has its own broker/proxy wiring that a blind
string substitution would not safely replace, and getting the sidecar
provisioning right needs repo-specific verification this issue's audit pass
did not have time to do safely.

---

Filed by the org-wide orchestrator/free routing audit (2026-09-02). See
ContextualWisdomLab/.github's docs/adr/0003-contextual-orchestrator-vendored-free-zdr.md
for the governed-gateway pattern (vendored sidecar, 5-credential KV, free-only
route admission) and docs/product-technical-gap-baseline.md's 2026-09-02
entry for the full audit disposition table.

This is a real, unauthorized violation (case b in the audit), not a half-fix:
routing a hardcoded direct-provider call through the shared gateway is a
bigger change than a model-string swap (own broker/proxy code, secret wiring,
and gateway startup all need to move), so it is being tracked here rather
than attempted blind in the audit pass.

🤖 Filed with Claude Code

[seonghobae]

Migrated: #235

.github/workflows/hourly-product-development.yml's develop job now vendors
ContextualWisdomLab/.github's scripts/ci/contextual_orchestrator_review_sidecar.sh
at a pinned commit (the same script .github's pr-review-autofix.yml and
strix.yml already run in production) instead of calling
https://integrate.api.nvidia.com/v1 directly. The five bootstrap-only
provider secrets (BYTEZ_API_KEY, NVIDIA_NIM_API_KEY,
NVIDIA_NIM_API_KEY_SUB, OPENROUTER_API_KEY, OPENAI_API_KEY) are
registered once into the sidecar's own in-memory KV; OpenCode's model is now
contextual-orchestrator/orchestrator/free, and OpenCode itself never
receives any of the five real provider secrets — only an ephemeral, per-run,
loopback-scoped bearer token.

One deliberate, disclosed tradeoff worth flagging here: the job's
step-security/harden-runner egress policy moves from block-mode with a
static host allowlist to audit-mode, because the sidecar's live
multi-provider discovery has no fixed host set to pin and no other consumer
of this exact sidecar in the org (.github's own pr-review-autofix.yml,
strix.yml, noema-review.yml, opencode-review-dispatch.yml, or
contextual-orchestrator's own opencode-hourly-loop.yml) runs it under
block-mode today — this matches the org's own existing posture for this
component rather than inventing a new one. The deny-by-default OpenCode
permission boundary (no webfetch/websearch/task/skill/lsp, three exact
read-only bash commands only) is unaffected. Full reasoning and everything
verified before shipping is in the PR description.

Also found and fixed a repo-specific landmine along the way: the vendored
sidecar hardcodes its own evidence directory to
$GITHUB_WORKSPACE/strix_runs/, which this repo's patch-capture guard
(scripts/ci/hourly_product_guard.py capture) would otherwise have swept
into the autonomous maintainer's own patch every run. Added strix_runs/ to
.gitignore to close that gap.