Allow configurable fallback chains for model elevation

Open
#1,555 0 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
4/5
Estimated time
3-5 days
Newbie friendliness
35/100
Issue type
Feature
Clarity
Mostly clear
Activity status
Quiet
Tech stack
typescript

Research direction

Start by tracing the model-elevation workflow and the existing plan_model configuration, then identify where provider availability failures trigger the current inline fallback. Locate the configuration parsing, retry/fallback orchestration, and model-selection reporting paths; the issue names no specific files or tests. Done means an explicitly ordered chain with per-model reasoning effort and fresh-context options is supported, exhaustion behavior is configurable, each attempt and reason is reported, and single-model behavior remains compatible.

Written by the indexing model from the issue text.

Description

Feature request

Allow model elevation to use an explicitly configured, ordered fallback chain when the preferred model is unavailable.

Current behavior

In Compound Engineering 3.23.4, CE Plan accepts a single model through plan_model. If that model cannot complete the elevated step, the workflow eventually falls back to authoring inline with the session model.

For example, I encountered this sequence:

  1. plan_model: fable selected Fable.
  2. The provider reported that the weekly Fable usage limit had been reached.
  3. CE Plan followed its current recovery behavior and continued inline.

That fallback keeps the workflow moving, but it removes the independent model judgment that model elevation was intended to provide.

Desired behavior

It would be useful to configure an ordered list of acceptable alternatives, including the reasoning effort for each one.

For example:

  1. Fable at high
  2. Opus at xhigh
  3. Sol at xhigh, launched as a fresh-context subagent
  4. Stop if none of the configured models is available

The model names should remain aliases such as fable, opus, and sol, allowing the provider or host to resolve the current model generation rather than requiring version-specific IDs.

The exact configuration format is up to the maintainers.

Expected properties

  • Fall back only to models the user explicitly configured.
  • Advance to the next model for clearly identified availability or capacity failures, such as an exhausted usage allowance.
  • Preserve the same read-only evidence and decision bundle across attempts.
  • Allow a fallback candidate to require fresh context so its judgment remains independent from the orchestrating session.
  • Report which models were attempted, why each fallback occurred, and which model and effort ultimately served the request.
  • Allow the user to choose whether total exhaustion falls back inline or stops the workflow.
  • Preserve the existing single-model behavior for users who do not configure a fallback chain.

Why this would help

In unattended or pipeline-driven runs, silently dropping from an independent planning model to the orchestrating session model can materially change the strength of the planning gate.

An explicit fallback chain would preserve the current transparent, user-controlled model-selection philosophy while making temporary provider limits less disruptive.

Dominant language
TypeScript
Stars
25.6k
Forks
2.1k
Avg merge
3h 57m
Merged PRs (30d)
170

Getting set up

This project ships no dev container, Dockerfile or contributing guide, so setting up is up to you: start from its README, and see our first-contribution guide for the general steps.

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from EveryInc/compound-engineering-plugin

All issues in EveryInc/compound-engineering-plugin

Similar issues

More TypeScript issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.