Bundled dev pack: the test-first skill - #534
Merged
Merged
Conversation
The published versions of this subject all reduce the red step to an instruction to look at something and a checkbox confirming the model looked. This one is organised around what the red run has to produce instead: the code contradicting the test, in values, in output that names the assertion. That partitions the three ways a red run comes back decisive and says nothing, which is where the measured failures sit. A generated test is likelier to pass on the unchanged code than to reproduce the behaviour it was written for, and where it does fail, the output frequently identifies neither the assertion nor the values. The last section is the part no other harness can carry. A ledger item's identity is a hash of its text and its declared check, and the ledger is append-and-mark only, so the check written before the work cannot be narrowed afterwards. Running it records a verdict with the exit code, a hash of the output, and a provenance stamp no model can write. Red before green is then a pair of ordered events a reader who was not there can check, rather than a claim the run makes about itself. Adds the skill and its seven rows in the pack's retrieval table. Signed-off-by: Ion Alpha <contact@ionalpha.io>
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Adds
test-firstto the bundled dev pack, and the seven objectives it must be retrieved for toskill/bundled/skills/retrieval.txt. Text only: no Go changes.Why
The published treatments of this subject reduce the red step to an instruction to watch the test fail and a box confirming the model watched it. The confirmation is what gets asked for, so the confirmation is what gets produced, and a red run that was never decisive passes anyway.
This skill is organised around what the red run has to produce instead: the code contradicting the test, in values, in output that names the assertion. That separates the three ways a run comes back red and still proves nothing, which is where the failures actually sit: the test that was green all along, the failure on a missing import rather than on an assertion, and the failure whose output names neither the assertion nor the values.
The closing section is the half no other harness can carry. A ledger item's identity is a hash of its text and its declared check, and the ledger is append-and-mark only, so a check written before the work cannot be narrowed afterwards. Running it records a verdict carrying the exit code, a hash of the output, and a provenance stamp no model can write. Red before green is then a pair of ordered events a reader who was not there can check, rather than a claim the run makes about itself.
How to verify
TestPackIsRetrievableruns the runtime's own ranker over the real pack for every row in the table, with no model and no tokens, so a description that never reaches its own objectives fails the build rather than failing silently at recall time.TestEveryPackSkillStatesItsTriggersrefuses a skill that ships without a row.Notes for reviewers
The rows were also measured against the pack as it will stand once the three sibling dev-pack skills land: every row in the table still passes with all thirteen skills present, so this skill neither displaces them nor is displaced by them.
dev/checkis green here apart from three pre-existingcmd/flynnfailures that reproduce identically on unmodifiedmain, all three from a localcodexlauncher with no vendored binary on this machine.