docs: add llms.md — AWF /reflect endpoint guide for LLM tool configuration

[Copilot]

Tools that need OpenAI/Anthropic LLM access inside the AWF agent container had no documented way to discover the api-proxy endpoints, ports, or available model names at runtime.

Changes

• .github/aw/llms.md (new) — documents the AWF api-proxy /reflect endpoint:

  • Response schema (provider, port, configured, models, models_url, models_fetch_complete)
  • Default port table: openai→10000, anthropic→10001, copilot→10002, gemini→10003
  • Shell snippets to query /reflect and enumerate models via models_url
  • Configuration patterns for OpenAI SDK, Anthropic SDK, and any OpenAI-compatible client — including a dynamic pattern that resolves base URL from /reflect at runtime
  • Network note: no extra network.allowed entries needed; api-proxy is in the AWF sandbox network

• .github/aw/github-agentic-workflows.md — added LLM API endpoint discovery row linking to llms.md in the Reference Documentation table

Example: discover and configure

# Discover configured providers
curl -sf http://api-proxy:10000/reflect | jq '.endpoints[] | select(.configured)'

# Dynamically resolve Anthropic base URL
ANTHROPIC_PORT=$(curl -sf http://api-proxy:10000/reflect \
  | jq -r '.endpoints[] | select(.provider == "anthropic" and .configured) | .port')
export ANTHROPIC_BASE_URL="http://api-proxy:${ANTHROPIC_PORT}"

[pelikhan]

@copilot optimize instructions for size / quality . Optimize for agentic reasoning.

[Copilot]

@copilot optimize instructions for size / quality . Optimize for agentic reasoning.

Optimized in commit docs(llms.md): optimize for size and agentic reasoning. The file is now ~60% smaller: verbose field-description tables and prose sections collapsed into a single compact provider port table, the response example condensed to one-liners, redundant SDK examples merged, and the "Shell Helper" and standalone "Network Requirements" sections removed. The core facts (provider → port → base URL → credential env var) are now immediately scannable in one table.

[Copilot]

Pull request overview

Adds internal AWF documentation explaining how tools running inside the agent container can discover LLM gateway endpoints/models at runtime via the api-proxy /reflect route.

Changes:

• Added a new guide (.github/aw/llms.md) describing /reflect, default provider ports, and example configuration snippets for common SDKs/clients.
• Linked the new guide from the Reference Documentation table in .github/aw/github-agentic-workflows.md.

Show a summary per file

File	Description
.github/aw/llms.md	New documentation for discovering and using api-proxy LLM endpoints/models via /reflect.
.github/aw/github-agentic-workflows.md	Adds a reference-table link to the new LLM endpoint discovery doc.

Copilot's findings

Tip

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

Comments suppressed due to low confidence (2)

.github/aw/llms.md:53

• This configuration example contradicts the earlier guidance to not pass raw API keys/tokens to the proxy. Many OpenAI clients only require a non-empty key locally; consider using a placeholder value (or clarifying when it is/isn't safe/necessary to reuse $COPILOT_GITHUB_TOKEN) so readers don't propagate real credentials unnecessarily.

```bash
# Use Copilot as the OpenAI backend
export OPENAI_BASE_URL="http://api-proxy:10002/v1"
export OPENAI_API_KEY="$COPILOT_GITHUB_TOKEN"

**.github/aw/llms.md:81**
* The Gemini model-list command uses `http://api-proxy:10003/v1/models`, but existing AWF reflect examples/tests use a `v1beta/models` route for Gemini (and the doc above recommends using `models_url`). To avoid a likely 404, resolve and call Gemini’s `models_url` from `/reflect` (or update the hard-coded path to match the gateway).

Gemini format → { models: [{name: "models/gemini-..."}] }

curl -sf http://api-proxy:10003/v1/models | jq '[.models[].name | ltrimstr("models/")]'

• Files reviewed: 2/2 changed files
• Comments generated: 4