Skip to content

v1.19.0 -- model tier economics layer

Choose a tag to compare

View all tags
@juan294 juan294 released this 13 Jun 06:08
· 5 commits to main since this release
v1.19.0
7b9a231

Ported the model-tier + cost-monitoring economics layer from copilot-rpi v1.16.0, adapted to Claude Code.

Added

  • Model tier economics layer. Added a haiku floor tier alongside opus/sonnet and gave every command an explicit Model tier line. /status and /describe-pr now run on the floor; /research, /plan, /pre-launch stay frontier; the rest run mid-tier.
  • methodology/cost-monitoring.md. Four cost pools, model tiers as the primary lever, access tiers, cost-per-outcome measurement, and cost-tied approval gates. Reading order now 12 files.
  • Cost-report scheduled agent (runs on the floor tier) that turns Claude Code / Anthropic API usage exports into cost-per-outcome numbers and flags tier drift.
  • Rule #74 (Pin a model tier to every workflow) and Rule #75 (Measure cost per outcome before betting beyond the floor) under a new "Cost & Models" section. Now 75 rules.
  • GUIDE "Model Tiers at a Glance" table.

Changed

  • context-engineering.md "Model Selection" rewritten -- replaced the generic guidance that contradicted the per-command tier lines with the 3-tier system, a tier-to-Claude-model binding table, subagent tier-inheritance, and an override-upward-never-silently-downward rule.
Assets 2
Loading