Use AWF /reflect for Copilot model discovery in daily-model-inventory

[Copilot]

✨ Enhancement

daily-model-inventory was relying on direct Copilot API calls with COPILOT_GITHUB_TOKEN, which was failing with HTTP 400 and creating an unnecessary secret dependency. This change moves Copilot model discovery to the AWF api-proxy /reflect path used by the runtime.

• What does this improve?

  • Removes brittle direct Copilot model fetch logic from the workflow pre-jobs.
  • Uses live Copilot model data from AWF /reflect for inventory and multiplier analysis.
  • Simplifies workflow setup by removing COPILOT_GITHUB_TOKEN dependency for model discovery.

• Why is this valuable?

  • Aligns model discovery with the existing AWF proxy auth/routing path.
  • Avoids known failure mode (api.githubcopilot.com/models HTTP 400).
  • Reduces operational surface area (one fewer collection job and artifact dependency).

• Implementation approach:

  • Removed collect_copilot_models job from .github/workflows/daily-model-inventory.md.
  • Added/updated bash tools to fetch and inspect /reflect once, then query Copilot models from the saved payload.
  • Updated prompt sections:
    • ## Inputs: Copilot source is /reflect instead of pre-step Copilot artifact.
    • Step 1/2: explicitly query/filter Copilot endpoint from /reflect.
    • Step 3: Copilot multiplier comparison now uses Copilot models sourced from /reflect plus docs-derived multiplier data.
  • Added graceful handling for /reflect unavailability so analysis continues for other providers.

mkdir -p /tmp/gh-aw/model-inventory && \
  (curl -fsS http://api-proxy:10000/reflect > /tmp/gh-aw/model-inventory/reflect.json || \
   printf "%s" "{\"endpoints\":[],\"error\":\"reflect endpoint unavailable\"}" > /tmp/gh-aw/model-inventory/reflect.json)

jq ".endpoints[] | select(.provider == \"copilot\") | .models" /tmp/gh-aw/model-inventory/reflect.json

[Copilot]

Pull request overview

Moves Copilot model discovery in the daily-model-inventory workflow from direct Copilot API calls (requiring COPILOT_GITHUB_TOKEN) to using the AWF api-proxy /reflect endpoint, aligning discovery with the runtime gateway path.

Changes:

• Removes the collect_copilot_models pre-job (and its artifact upload) from the workflow.
• Adds tooling to fetch and query http://api-proxy:10000/reflect for Copilot model IDs.
• Updates the workflow prompt/instructions to treat Copilot inventory as coming from /reflect and to continue gracefully when /reflect is unavailable.

Show a summary per file

File	Description
.github/workflows/daily-model-inventory.md	Removes direct Copilot model fetch job and switches Copilot model discovery instructions/tools to use AWF /reflect.

Copilot's findings

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Comments suppressed due to low confidence (1)

.github/workflows/daily-model-inventory.md:363

• Per the documented /reflect response shape, endpoints[].models is a list of model IDs (or null while discovery runs) and does not include vendor/capabilities/billing metadata. The note that Copilot models from /reflect "may include vendor metadata" appears inaccurate and could mislead the analysis steps. Consider rewording to match /reflect (IDs only), and if vendor attribution is required, describe how it will be derived.

Note: Copilot model data must be read from the AWF `/reflect` endpoint. Copilot serves models from
multiple vendors (Anthropic, OpenAI, Google), and those models may include `vendor` metadata.

• Files reviewed: 1/1 changed files
• Comments generated: 2