Repository navigation
test: cover skill catalog filtering and checkpoint template restore - #8620
gabemilani merged 2 commits into
Conversation
Strix Security ReviewWarning This pull request has 3 commits after the last Strix review ( ✅ No open security findings. PR summary: Adds two regression tests only: one asserting that `build_skill_catalog` does not list skills already promoted to the INSTRUCTIONS disclosure level, and one asserting that a task restored via `POST /none` (checkpoint load) keeps its original `expected_object` template separate from its description. Review summaryThis PR is test-only: it adds Review 1 on this PR · Reviewed by Strix |
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Co-Authored-By: Claude Opus 5.5 <[email protected]>
1853cc5 to
fb17a10
Compare
Thanks for your first contribution to CrewAI, @Trenyx!We really appreciate the time and care you put into this PR. Your work is now part of CrewAI — welcome to the contributor community. Want to share your contribution? Totally optional: Share on X · Share the PR on LinkedIn If the links don't work, feel free to copy and personalize this: If you share, we'd love to see it — tag @crewAIInc on X and CrewAI on LinkedIn. |
Related issue
Closes #8619
Summary
Two test-only commits, one per behaviour, so either can be kept on its own:
skills/test_loader.py::TestBuildSkillCatalog: an activated (INSTRUCTIONS-level) skill is not listed bybuild_skill_catalogas still loadable.test_checkpoint.py::TestCrewCheckpointTaskTemplates: after a checkpoint andCrew.from_checkpoint, a task's saved original expected output is its own template, not its description.Verification
I planted each bug in a private copy (
<to<=inbuild_skill_catalog;_sync_checkpoint_fieldscopying_original_descriptionintocheckpoint_original_expected_output) and the fulllib/crewai/testssuite stayed green (6,313 passed, main at b7de28e). On this branch, each new test fails with its bug and no other test does; without the bugs, all 6,315 pass.Additional context
Test-only; no product code changed. Written with help from an AI assistant (Claude Code); please apply the
llm-generatedlabel (I don't have permission to add labels). Found with Trenyx (trenyx.io), which plants realistic bugs to see what tests miss.Note
Low Risk
Only new tests; no runtime or API changes.
Overview
Test-only PR that adds regression coverage for two behaviors (refs #8619); no production code changes.
Adds
TestBuildSkillCatalogintest_loader.py, assertingbuild_skill_catalogreturns only skills still below INSTRUCTIONS (metadata-only / “pending”) and omits activated and resource-loaded skills from the loadable catalog.Adds
TestCrewCheckpointTaskTemplatesintest_checkpoint.py, asserting that after checkpoint +Crew.from_checkpoint, a task keeps separate uninterpolated templates fordescriptionandexpected_output(includingcheckpoint_original_*and_original_expected_output), so resume can re-interpolate each field from the right source—not from the description alone.Reviewed by Cursor Bugbot for commit 1853cc5. Bugbot is set up for automated code reviews on this repo. Configure here.