Rework os-context audit criteria to catch expensive-model overuse #71

Open
opened 2026-07-16 19:15:26 +00:00 by jared · 0 comments
Owner

Surfaced during os-sdlc design discussion (2026-07-16). Recent sessions burn too many tokens on expensive models where smart delegation and determinism would improve speed and reduce cost with no quality drop. The os-context IRL session audit — especially its audit criteria — should detect this pattern: flag work done at opus/fable tier that needed no main-loop judgment, and missed delegation/determinism opportunities. Relevant: plugins/os-context/ (audit-sessions skill, prompts/session-start/10-orchestration.md).

Surfaced during os-sdlc design discussion (2026-07-16). Recent sessions burn too many tokens on expensive models where smart delegation and determinism would improve speed and reduce cost with no quality drop. The os-context IRL session audit — especially its audit criteria — should detect this pattern: flag work done at opus/fable tier that needed no main-loop judgment, and missed delegation/determinism opportunities. Relevant: plugins/os-context/ (audit-sessions skill, prompts/session-start/10-orchestration.md).
Sign in to join this conversation.
No description provided.