SuperQode v0.2.25 : RLM Code as a First-Class Harness
RLM Code as First Class Harness for AuperQode
SuperQode 0.2.25 integrates RLM Code v0.1.11 as a first-class, optional harness backend.
RLM Code continues to own recursive execution, symbolic context, programmatic subcalls, history management, and native
trajectories. SuperQode adds the portable HarnessSpec, policy enforcement, normalized evidence, cross-harness
evaluation, and guarded optimization around it.
What’s new
-
Native rlm-code HarnessSpec backend.
-
Portable Harness Protocol adapter.
-
Support for reference, repo_evidence, and lid execution profiles.
-
Caller-owned context and configurable repository context selection.
-
Normalized evidence for:
- REPL steps and results
- Root versus submodel usage
- Token attribution
- Root-exposed and hidden context
- Structural actions and history offloading
- Native RLM JSONL trajectories
-
Docker-first safety with repository-owned sandbox and network policies.
-
Explicit protection against silently enabling unsafe in-process execution.
-
Maintained rlm-code-lid example for local models.
-
Complete installation, configuration, architecture, evaluation, optimization, safety, and limitations guide.
Installation
uv tool install "superqode[rlm-code]"
For a development checkout:
uv sync --extra rlm-code
The integration supports rlm-code>=0.1.11,<0.2.0.