improve daily-model-inventory: richer endpoint data and token multiplier analysis

[Copilot]

The workflow only cached model IDs and had no mechanism to validate or infer the ET multipliers in model_multipliers.json against live API data.

Data collection

Each provider's fetch job now extracts enriched metadata and uploads both models.json (structured) and raw.json (full API response) as artifacts:

Provider	New fields captured
OpenAI	owned_by, created
Anthropic	display_name, created_at, type
Gemini	display_name, description, input_token_limit, output_token_limit, supported_generation_methods, version
Copilot	name, vendor, version, capabilities (limits + supports flags), billing

Agent prompt (6 steps, was 4)

• Step 2 — Explore Raw API Fields: agent inspects each provider's raw.json, identifies which fields carry context limits, capability flags, or billing/pricing data, and summarizes what's worth caching going forward.
• Step 3 — Infer Token Multipliers: agent cross-references live data with model_multipliers.json using provider-specific strategies (Copilot billing.multiplier directly; Gemini inputTokenLimit/outputTokenLimit as a complexity heuristic; OpenAI/Anthropic naming conventions) and produces a gap table: missing entries, stale entries, and inferred-vs-stored discrepancies.

Issue template

Adds Raw API Fields Discovered and Token Multiplier Analysis sections (with sub-tables for missing, stale, and discrepant multipliers) to the generated issue body.

Tools allowlist

Added bash allowlist entries so the agent can query raw.json artifacts and model_multipliers.json.

[Copilot]

Pull request overview

Enhances the daily model inventory workflow to cache richer per-provider model metadata, preserve raw API responses as artifacts, and expand the agent prompt to analyze/validate token multipliers against live data.

Changes:

• Enrich provider models.json with additional metadata fields and upload raw.json alongside it for OpenAI/Anthropic/Gemini/Copilot.
• Expand the agent prompt to explore raw API fields and infer/validate ET token multipliers vs pkg/cli/data/model_multipliers.json.
• Extend the tool allowlist to enable querying raw.json artifacts and the multiplier table.

Show a summary per file

File	Description
.github/workflows/daily-model-inventory.md	Updates collection steps to produce enriched model metadata + raw artifacts; expands prompt steps and issue template; extends bash tool allowlist.
.github/workflows/daily-model-inventory.lock.yml	Regenerates the locked workflow to reflect the new artifact outputs and expanded tool allowlist.

Copilot's findings

Tip

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

Comments suppressed due to low confidence (3)

.github/workflows/daily-model-inventory.md:205

• The Copilot fetch uses curl -sf -o "$OUT/raw.json" ... || true; with -f, HTTP errors can prevent raw.json from being written, but the upload step requires it and will fail. Ensure raw.json is always produced (either capture error bodies by dropping -f or write a fallback JSON error response when curl fails).

          HTTP_STATUS=$(curl -sf -o "$OUT/raw.json" -w "%{http_code}" \
            -H "Authorization: Bearer $COPILOT_GITHUB_TOKEN" \
            -H "Copilot-Integration-Id: copilot-chat" \
            https://api.githubcopilot.com/models) || true

.github/workflows/daily-model-inventory.md:96

• Same curl -sf -o "$OUT/raw.json" ... || true issue here: with -f, raw.json may not be written for non-200 responses, but the artifact upload expects it and will error. Ensure raw.json is always present (capture error bodies by dropping -f or write a fallback JSON error payload on failure) and avoid an empty HTTP_STATUS in the error message.

          HTTP_STATUS=$(curl -sf -o "$OUT/raw.json" -w "%{http_code}" \
            -H "x-api-key: $ANTHROPIC_API_KEY" \
            -H "anthropic-version: 2023-06-01" \
            https://api.anthropic.com/v1/models) || true

.github/workflows/daily-model-inventory.md:148

• With curl -sf -o "$OUT/raw.json" ... || true, a non-200 response (or network failure) can result in raw.json not being created because -f suppresses output on HTTP errors. The subsequent artifact upload includes raw.json and will fail if it’s missing. Ensure raw.json is always created and consider capturing error bodies by removing -f.

          HTTP_STATUS=$(curl -sf -o "$OUT/raw.json" -w "%{http_code}" \
            "https://generativelanguage.googleapis.com/v1beta/models?key=${GEMINI_API_KEY}") || true

• Files reviewed: 2/2 changed files
• Comments generated: 3