evals: harness for capturing flow runs and qualitative reviews#21
Open
jyliang wants to merge 2 commits into
Open
evals: harness for capturing flow runs and qualitative reviews#21jyliang wants to merge 2 commits into
jyliang wants to merge 2 commits into
Conversation
Adds make eval-task-new / eval-record / eval-review / eval-compare / eval-list to capture a flow run as a versioned record (kernel SHA + cell SHA + thread + diff + manifest), open a long-form review template in $EDITOR, and compare two runs side by side. Reviews are markdown so they're diffable and committed alongside the run for the next iteration of /flow:reflect to read. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Greenfield code-pipeline target — single-file Python TODO CLI in a fresh repo at ~/Workspace/jyliang/mini-todo (init commit 9f05045). Prompt is intentionally tight so divergences across runs reflect the cell, not prompt ambiguity. Watching for spec stage pinning the implicit edge cases, plan stage avoiding unnecessary structure, and implement stage staying under 150 LOC. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
make eval-{task-new,record,review,compare,list}targets for capturing flow runs as versioned, reviewable artifacts.evals/runs/<task>/<run-id>/reviews/<reviewer>.mdand are committed alongside the run, so/flow:reflect(and humans iterating on flow) can read them next time.eval-compareprints version pins and metric deltas side by side and points at shared reviewers' files for prose-level diffing.Layout
run-id=<utc-iso>-<kernel-sha7>— sortable, scannable.Test plan
make eval-task-new TASK=<id>scaffolds tasks//{task.json, prompt.md, README.md}make eval-record TASK=<id> PROJECT=<path>discovers the most recent thread, copies it, generates the diff, writes a manifest pinned to current kernel + cell SHAsmake eval-review TASK=<id> RUN=<run-id>opens$EDITORon a pre-filled review template with an artifact index appendedmake eval-compare TASK=<id> A=<a> B=<b>prints version + metric deltas and lists reviewsmake eval-listsummarises tasks and runs🤖 Generated with Claude Code