Remora CC

remora
Run Claude Code with a cost-aware OpenAI agent fleet for one session.

remora launches Claude Code with session-scoped OpenAI model routing,
role agents, and orchestration. Sol handles the main session, planning, and
critical review, while Luna handles lower-cost exploration and implementation.
Exiting the child session removes every override.

繁體中文

Contents

- What remora changes
- Architecture and model map
- Requirements
- Install
- Configure
- Use
- Isolation and security
- Troubleshooting
- Uninstall
- Support remora
- Further reading

What remora changes
Core guarantee: plain claude keeps its original credentials, settings,
agents, and model routing. remora changes only the child process it launches.

Large Plans use a program envelope followed by independently approvable
execution slices. Concrete security, irreversible/external, data, release, or
cross-component acceptance risk triggers independent review; “non-trivial”
alone does not. After two automatic REVISE verdicts, the main session stops
automatic resubmission and dispositions each blocker as FIX, DEFER, or
REJECT. One materially changed unit may receive one final fresh review;
another REVISE pauses or escalates it. The complete contract belongs in
the architecture document.

Intent routing uses execute for clear bounded work, explore_then_plan for
broad or high-impact work, and co_discover for open-ended ideas. Discovery
has a bounded grounding floor and stopping ceiling; pre-approval discovery is
read-only and next_gate=approval stops before execution. Turn-scoped
review_intent accepts only clear explicit fast, default, or strict
requests; ambiguous, quoted, negated, and conflicting cues use default.
Mandatory security, risk, approval, and external-action gates always win. This
is an…