Skip to content

feat(skillopt): ingest blinded live-pairwise packet → canonical feedback (#77b) #511

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
jerryfane merged 2 commits into from
Jun 27, 2026
Merged

feat(skillopt): ingest blinded live-pairwise packet → canonical feedback (#77b) #511

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
fefdda7
feat(skillopt): ingest blinded live-pairwise packet into canonical fe…
jerryfane Jun 27, 2026
32e8be4
fix(skillopt): reject foreign picks run_id on pairwise import (#508)
jerryfane Jun 27, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions internal/cli/skillopt.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -78,6 +78,8 @@ func runSkillOpt(args []string, stdout, stderr io.Writer) int {
return runSkillOptTrain(args[1:], stdout, stderr)
case "ab":
return runSkillOptAB(args[1:], stdout, stderr)
case "pairwise":
return runSkillOptPairwise(args[1:], stdout, stderr)
default:
fmt.Fprintf(stderr, "unknown skillopt command %q\n\n", args[0])
printSkillOptUsage(stderr)
Expand All @@ -102,6 +104,7 @@ func printSkillOptUsage(w io.Writer) {
fmt.Fprintln(w, " gitmoot skillopt feedback github publish --run <run-id> [--repo owner/repo] [--pr <number>]")
fmt.Fprintln(w, " gitmoot skillopt feedback github sync --run <run-id> [--repo owner/repo] (--issue <number>|--pr <number>)")
fmt.Fprintln(w, " gitmoot skillopt ab <agent> \"<prompt>\" [--challenger <versionId>] [--pick a|b] [--seed N] [--judge] [--judge-only] [--home path]")
fmt.Fprintln(w, " gitmoot skillopt pairwise import <packet-dir> [--packet path] [--secret-map path] [--picks path] [--reviewer name] [--home path] [--json]")
fmt.Fprintln(w, " gitmoot skillopt judge-report [--template id]")
fmt.Fprintln(w, " gitmoot skillopt judge promote --template <id> --task-kind <kind> --file <pkg.json> [--home <h>] [--yes] [--json]")
fmt.Fprintln(w, " gitmoot skillopt train init --name <name> --template <id> --review-repo owner/repo --artifact-kind kind --preview kind (--request text|--request-file path)")
Expand Down
24 changes: 12 additions & 12 deletions internal/cli/skillopt_ab.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -561,10 +561,10 @@ func defaultReadSkillOptABLine() (string, bool) {
// (ranking=[winner,loser], source=skillopt-ab) that passes
// store.validateRankedFeedbackEventOptions.
func recordSkillOptABPick(ctx context.Context, store *db.Store, paths config.Paths, runID, pickSourceURL, templateID string, champion, challenger skillOptABVariant, championDelivery, challengerDelivery skillOptABDelivery, winnerLabel, loserLabel, prompt string) error {
if err := ensureSkillOptABRunRows(ctx, store, paths, runID, templateID, champion, challenger, championDelivery, challengerDelivery, prompt); err != nil {
if err := ensureSkillOptABRunRows(ctx, store, paths, runID, skillOptABItemID, templateID, champion, challenger, championDelivery, challengerDelivery, prompt, skillOptABSource, skillOptABFeedbackSource); err != nil {
return err
}
return upsertSkillOptABRankedEvent(ctx, store, runID, winnerLabel, loserLabel, "human", skillOptABSource, pickSourceURL)
return upsertSkillOptABRankedEvent(ctx, store, runID, skillOptABItemID, winnerLabel, loserLabel, "human", skillOptABSource, pickSourceURL)
}

// ensureSkillOptABRunRows idempotently upserts the shared pairwise scaffolding for
Expand All @@ -575,12 +575,12 @@ func recordSkillOptABPick(ctx context.Context, store *db.Store, paths config.Pat
// judge row is self-sufficient when --judge-only skips the human record, and the
// rows are byte-identical when both run (every write here is an idempotent upsert
// keyed by (run_id,item_id,label)).
func ensureSkillOptABRunRows(ctx context.Context, store *db.Store, paths config.Paths, runID, templateID string, champion, challenger skillOptABVariant, championDelivery, challengerDelivery skillOptABDelivery, prompt string) error {
func ensureSkillOptABRunRows(ctx context.Context, store *db.Store, paths config.Paths, runID, itemID, templateID string, champion, challenger skillOptABVariant, championDelivery, challengerDelivery skillOptABDelivery, prompt, source, feedbackSource string) error {
blobStore := artifact.NewStore(paths.ArtifactBlobs)

metadataJSON, err := json.Marshal(map[string]any{
"feedback_source": skillOptABFeedbackSource,
"source": skillOptABSource,
"feedback_source": feedbackSource,
"source": source,
"champion": champion.version.ID,
"challenger": challenger.version.ID,
})
Expand All @@ -599,7 +599,7 @@ func ensureSkillOptABRunRows(ctx context.Context, store *db.Store, paths config.
if err := store.UpsertEvalRun(ctx, run); err != nil {
return err
}
if err := store.UpsertEvalReviewItem(ctx, db.EvalReviewItem{RunID: runID, ItemID: skillOptABItemID, Title: prompt}); err != nil {
if err := store.UpsertEvalReviewItem(ctx, db.EvalReviewItem{RunID: runID, ItemID: itemID, Title: prompt}); err != nil {
return err
}

Expand All @@ -612,7 +612,7 @@ func ensureSkillOptABRunRows(ctx context.Context, store *db.Store, paths config.
{label: skillOptABChampionLabel, delivery: championDelivery},
{label: skillOptABChallengerLabel, delivery: challengerDelivery},
} {
evalArtifact, err := prepareReviewItemContentArtifact(blobStore, runID, skillOptABItemID, opt.label, []byte(answerOrPlaceholder(opt.delivery.answer)), "text/plain", "text")
evalArtifact, err := prepareReviewItemContentArtifact(blobStore, runID, itemID, opt.label, []byte(answerOrPlaceholder(opt.delivery.answer)), "text/plain", "text")
if err != nil {
return err
}
Expand All @@ -621,7 +621,7 @@ func ensureSkillOptABRunRows(ctx context.Context, store *db.Store, paths config.
}
if err := store.UpsertEvalReviewOption(ctx, db.EvalReviewOption{
RunID: runID,
ItemID: skillOptABItemID,
ItemID: itemID,
Label: opt.label,
ArtifactID: evalArtifact.ID,
Role: opt.label,
Expand All @@ -640,7 +640,7 @@ func ensureSkillOptABRunRows(ctx context.Context, store *db.Store, paths config.
// (reviewer=skillopt-ab-judge,source=skillopt-ab-judge) as SEPARATE coexisting
// rows on the same run/item, and what makes repeated picks each persist via a
// distinct per-pick source_url.
func upsertSkillOptABRankedEvent(ctx context.Context, store *db.Store, runID, winnerLabel, loserLabel, reviewer, source, sourceURL string) error {
func upsertSkillOptABRankedEvent(ctx context.Context, store *db.Store, runID, itemID, winnerLabel, loserLabel, reviewer, source, sourceURL string) error {
ranking, err := json.Marshal([]string{winnerLabel, loserLabel})
if err != nil {
return err
Expand All @@ -651,7 +651,7 @@ func upsertSkillOptABRankedEvent(ctx context.Context, store *db.Store, runID, wi
}
return store.UpsertRankedFeedbackEvent(ctx, db.RankedFeedbackEvent{
RunID: runID,
ItemID: skillOptABItemID,
ItemID: itemID,
RankingJSON: string(ranking),
TieGroupsJSON: string(tieGroups),
Winner: winnerLabel,
Expand Down Expand Up @@ -766,12 +766,12 @@ func runSkillOptABJudge(ctx context.Context, store *db.Store, paths config.Paths

// Ensure the shared run/item/options exist (idempotent) so the judge row is
// self-sufficient even under --judge-only where the human record path never runs.
if err := ensureSkillOptABRunRows(ctx, store, paths, runID, templateID, champion, challenger, optionAToDelivery(optionA, optionB, champion), optionAToDelivery(optionA, optionB, challenger), options.prompt); err != nil {
if err := ensureSkillOptABRunRows(ctx, store, paths, runID, skillOptABItemID, templateID, champion, challenger, optionAToDelivery(optionA, optionB, champion), optionAToDelivery(optionA, optionB, challenger), options.prompt, skillOptABSource, skillOptABFeedbackSource); err != nil {
log.Printf("skillopt ab judge: ensure run rows: %v", err)
return
}
judgeSourceURL := skillOptABJudgeSourceURL(challenger.version.ID)
if err := upsertSkillOptABRankedEvent(ctx, store, runID, winnerLabel, loserLabel, skillOptABJudgeReviewer, skillOptABJudgeSource, judgeSourceURL); err != nil {
if err := upsertSkillOptABRankedEvent(ctx, store, runID, skillOptABItemID, winnerLabel, loserLabel, skillOptABJudgeReviewer, skillOptABJudgeSource, judgeSourceURL); err != nil {
log.Printf("skillopt ab judge: record judge pick: %v", err)
return
}
Expand Down
Loading
Loading