Edge: reserve GLM reasoning budget for skill outputs

[Franlinozz]

Summary

• Reserve enough completion-token budget for skill invokes so GLM-5 can emit final message.content after its reasoning phase.
• Keeps the response shape unchanged: Edge still returns output.summary, output.translation, etc.; the PR fix(web): render per-skill invoke outputs #13 frontend renderer remains the canonical display path.
• Updates the decision note documenting that GLM can still spend reasoning tokens even with chat_template_kwargs.enable_thinking=false.

Production evidence

After PR #14 deployed, the [object Object] bug was gone but a production invoke with the old small budget returned leaked reasoning text as output.summary because GLM returned content: null after exhausting the skill budget. Raw provider tests showed the same prompt emits clean final content once the request has ~1536 completion tokens available.

Verification

• pnpm -F @apogee/edge typecheck
• pnpm -F @apogee/edge build
• pnpm -F @apogee/edge test

Scope

No changes to receipt minting, ownership checks, policy enforcement, status gates, auto-promote logic, Pilot endpoint, Aurora/Vesper/Helix, or heartbeat runtime.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
apogee	Ready	Preview, Comment	May 19, 2026 11:14am