feat(config): models.yaml lane map + per-project native roles — 0.6.0 #2
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "feat/models-yaml-lane-map"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Two commits, each standing on its own:
feat(configure-codex-roles): per-project native role layer via --project— pre-existing local work:status/apply/remove --project <repo>manage the project.codexrole layer, project leaves overriding only the roles they name. Verified by its own 19 tests.feat(config): models.yaml lane map, shared schema, + bin/isc — 0.6.0— model selection moves from dottedmodel.<lane>.<tier>keys tomodels.yaml, the schema shared verbatim with the Claude Code and Cursor variants.Reviewer defaults are inverted relative to the Claude Code variant on purpose: this harness implements on GPT, so an independent review belongs on a Claude lane. A native role profile is static config Codex itself runs, so it must stay Codex-runnable —
adversarial_reviewerinheritsimplementer_high, and bothisc doctorand the installer refuse to write an Anthropic id into one; cross-family review reaches Claude throughengine.claude.review.Legacy
- model.role.<name>lines still resolve, one step weaker than the yaml in the same scope. The other legacy families are documented in a migration table and named byisc doctorrather than silently ignored.isc syncrecords a digest of the resolved role map, sodoctordetects profiles rendered from a different lane map instead of merely checking that files exist. One effective Codex home (CODEX_HOME) is used by resolution, init and the installer alike, and the installer's parser matchesbin/iscbyte for byte.51 checks in
tests/isc-smoke.shplus 23 Python tests (needs python3 ≥ 3.11 fortomllib, or the codex CLI).