fix(model-inventory): enrich /reflect null models via models_url fallback

[Copilot]

The daily model inventory workflow reported Copilot as unavailable because the /reflect endpoint returns models: null for Copilot when the api-proxy hasn't completed its startup model fetch — there was no fallback to retrieve them.

Changes

Pre-step enrichment (Fetch Copilot reflect inventory)

• After fetching /reflect, iterates over configured endpoints where models == null and fetches model IDs directly from each endpoint's models_url — mirrors the enrichReflectModels() pattern in awf_reflect.cjs as documented in .github/aw/llms.md
• Adds diagnostic logging so fallback fetch failures surface in step output rather than silently producing empty results

while IFS= read -r entry; do
  provider=$(printf '%s' "$entry" | jq -r '.provider')
  models_url=$(printf '%s' "$entry" | jq -r '.models_url')
  if raw=$(curl -fsS "$models_url" 2>&1); then
    ids=$(printf '%s' "$raw" | jq -c '[.data[].id] // empty' 2>&1) || { echo "Warning: ..."; ids=""; }
    # inject ids back into reflect.json for the provider
  fi
done < <(jq -c '.endpoints[]? | select(.configured == true and .models == null and .models_url != null)' "$OUT")

Agent instruction consistency

• Step 1, Step 2, and Step 3 all previously told the agent to query http://api-proxy:10000/reflect directly — redundant since the pre-step already fetches and enriches the data. Updated all three to read from the pre-fetched /tmp/gh-aw/model-inventory/reflect.json
• Updated the Copilot data note to explain the pre-fetch + models_url enrichment flow and reference llms.md

Lock file recompiled.

[Copilot]

Pull request overview

Fixes the daily model inventory workflow’s handling of Copilot /reflect responses where models: null by adding a best-effort enrichment step that fetches model IDs from each endpoint’s models_url, and updates agent instructions to consume the pre-fetched reflect payload from disk.

Changes:

• Add a /reflect enrichment loop that fetches model IDs from models_url for configured endpoints with models == null, with warning-level logging on failures.
• Update Steps 1–3 instructions to read Copilot reflect data from /tmp/gh-aw/model-inventory/reflect.json instead of querying the api-proxy directly.
• Recompile the workflow lockfile.

Show a summary per file

File	Description
.github/workflows/daily-model-inventory.md	Adds models_url fallback enrichment + updates instructions to use the enriched reflect.json artifact.
.github/workflows/daily-model-inventory.lock.yml	Regenerates the locked workflow with the updated pre-step script content.

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:315

• if [ -n "$ids" ] treats [] as a successful enrichment, so this can overwrite models: null with an empty array even when the upstream returned no models (or parsing produced an empty list). awf_reflect.cjs keeps models as null when the extracted ID list is empty. Consider checking length > 0 before updating reflect.json (otherwise leave models null and warn) to preserve the “unavailable vs empty” distinction.

          if [ -n "$ids" ]; then
            jq --arg p "$provider" --argjson m "$ids" \
              '(.endpoints[] | select(.provider == $p) | .models) |= $m' \
              "$OUT" > "${OUT}.tmp" && mv "${OUT}.tmp" "$OUT"
            echo "Enriched $provider with $(printf '%s' "$ids" | jq 'length') model(s)"

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