PR Custody

[codex] fix: plan action identities before action build

@shreysingla11checks n/achecks…codex/slug-identity-planner → next17 files · +876 −98updated 2mo ago

▸Description

Why

The current action-building flow lets the builder invent the final class name, and Mercury derives the slug from that class name. That makes action identity drift likely: finder names, knowledge files, builder output, and final Mercury slugs can diverge, which then leads to cleanup renames later.

What

add an action_namer planner that computes final slugs and class names before code generation
make create_app and build_action use the planned action identity before knowledge creation and action building
require ActionBuilderRunner callers to pass both action_name and class_name
tighten builder validation so check_code enforces the expected class name, expected enum, and explicit enum declaration
add tests for deterministic action planning and the new action identity checks

How to Test

make fmt
make chk
uv run pytest -q cortex/tests/test_common/test_action_namer.py cortex/tests/test_tools/test_check_code.py

Pre-Review Checklist

I have self-reviewed this PR

Notes

PR generated from Codex
this keeps slug planning outside ActionBuilder and removes the old build-action fallback that inferred slugs from the returned class name
Mercury already supports explicit class-level enum, so no base-framework change was needed

loading diff…