(MOT-4163) providers: unify identity prompts on the minimal core - #560
(MOT-4163) providers: unify identity prompts on the minimal core#560rohitg00 wants to merge 1 commit into
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
skill-check — worker0 verified, 47 skipped (no docs/).
Four for four. Nicely done. |
|
Warning Review limit reached
Next review available in: 59 minutes Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available. How can I continue?After more reviews become available, a review can be triggered using the To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews. How do review limits work?CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability. For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window. Please refer docs for additional details. Review details⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (7)
✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
rohitg00
changed the title
1 participant
What
The seven provider identity prompts (28 to 32 KB each, ~7,100 to 7,600 tokens live) unify on the minimal core from #558 (~1,800 tokens each), plus the per-provider lines their tests pin:
IMPORTANT: NEVER invent function idsemphasis lineNever invent function idsvariant## Autonomy and persistencesection, kept verbatimAll seven
declarationtest suites pass unchanged: they pinstarts_with,agent_trigger, and the per-provider lines above, and each still ships its embedded prompt viainclude_str!.Why
The seven identities are near-copies of the harness prompt that have already drifted into three divergent variants. Every session on any of these providers pays ~7,100+ system tokens per generation; 63% of that is the orchestration playbook, which #558 moves into on-demand skills. After this PR each provider serves the same 1,800-token core, deep playbooks are pulled from
harness/*skills only when a task needs them, and per-provider tuning is a visible delta instead of a fork of the whole prompt.Live A/B evidence is in #558: minus 70% system tokens per generation with correctness parity across 13 sessions, measured with the anthropic identity as baseline.
Dependencies
Depends on #558 shipping the
harness/*playbook skills the core points to. The router operator override (providers.<name>.system_promptin llm-router config) remains the per-rig escape hatch in both directions.Draft: needs #558 direction agreement first.
Linear
Closes MOT-4163. Part of MOT-4157.
Open question from review
The seven files are three content variants generated from one core plus per-provider inserts; the generator ran offline. If this lands, a committed generator plus a CI drift check (regenerate and diff) would keep the eight copies honest - it can be added here or as a follow-up; the open question is where codegen belongs in this repo.