docs(skills): add the four-stage model evaluation loop - #1032
Closed
jswir wants to merge 2 commits into
Closed
Conversation
Mirror the shared evaluation suite so Publisher agents can measure, diagnose, and improve semantic models with deterministic scoring and gated edits. Co-authored-by: Cursor <cursoragent@cursor.com>
Bump the skills package so the new evaluation workflow is published instead of being skipped as unchanged. Co-authored-by: Cursor <cursoragent@cursor.com>
Collaborator
Author
|
Superseded by #1101, which ports the loop as it stands today (with the scripts, the ledger contract, the golden verifier and the run viewer) on top of the manifest in #1088. This branch was a five-week-old snapshot without the scripts; rebasing it would have meant porting things that had since moved. Closing rather than rebasing. |
jswir
added a commit
that referenced
this pull request
Sep 2, 2026
…al-diagnose, eval-improve The four skills that evaluate a semantic model: a set of questions whose goldens are computed from raw tables in a truth package the answerer cannot reach, a blind answerer over the model holding only the analysis skills, a judge that sees the golden and the re-executed rows and may doubt the golden, a diagnosis of each failure that says which artifact owns it, and one smallest model edit gated by a re-run. With the scripts that make it a loop: a runner that refuses to start on drifted goldens or a dead server, a golden verifier, a truth-package scaffold, a server spawner, a browsable run package, and a shared ledger that every stage writes through. Shared skills, identical to upstream (ms2data/agent-skills#81) and registered in publisher-local.json under the new eval group. The manifest's analysis and modeling groups are what the loop installs for the agents it measures, so the same scripts run from this checkout unchanged. The engine-side evaluation of get_context itself is deliberately not here: it is Credible's question about its hosted engine and stays in an unlisted skill upstream. Supersedes #1032, a port of a five-week-old snapshot that predates the scripts. Bundle regenerated; 18 bundle, 169 manifest tests and check-pack pass. credibledata/malloy-samples#23 is a set anyone can run this on.
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 join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Summary
Test plan
bun run test:skillscd packages/server && bun test src/mcp/skillsMade with Cursor