feat(judges): Add harness context to judge API#46
Merged
Conversation
dcramer
force-pushed
the
codex/judge-harness-context
branch
2 times, most recently
from
bc202a9 to
7897a6d
Compare
Pass configured harness context into automatic and explicit judge calls so rubric judges can reuse the suite prompt seam without duplicating provider setup. Register fixture run context for matcher assertions, including raw output and session objects, while keeping explicit matcher overrides available for manual values. Make harness prompt configuration required and keep judge prompting on context.harness.prompt(...) so the API does not split judge capabilities across harness and runtime objects. Fixes GH-45 Co-Authored-By: OpenAI Codex <codex@openai.com>
dcramer
force-pushed
the
codex/judge-harness-context
branch
from
7897a6d to
df19255
Compare
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Give every judge the same
JudgeContextpopulated from the configured harness. LLM-backed judges now reuse the suite model seam through the requiredharness.prompt(...)method, while adapter-specificruntimeobjects stay scoped to app execution internals such as tools and events.Single Judge API
Automatic judges, explicit
toSatisfyJudge(...)calls, and built-in deterministic judges all receive the same normalized context. The harness-specific judge context and judge-facing runtime object are gone, so custom judges read run data, metadata, tool calls, and the configured harness from one place.Implicit Matcher Context
Fixture-backed runs register their run, session, and output objects so matcher calls can infer input, metadata, tool calls, and harness without repetitive options. Exact registered objects win over the latest-run fallback, which keeps
expect(result.output).toSatisfyJudge(...)concise without hardcoding one narrow case.Required Harness Prompt
Harness.promptis required across the root type and first-party harness constructors. Rubric and factuality judges callharness.prompt(...);harness.run(...)remains the explicit escape hatch for intentionally running the application again.API Policy
Add a small API design policy for this lesson: prefer one shared contextual API, keep owned capabilities mandatory, put capabilities on the object that owns their configuration, and avoid parallel public objects with overlapping lifecycle names such as
harnessandruntime.Fixes #45