diff options
| author | Peter Stone <thepeterstone@gmail.com> | 2026-07-10 03:30:37 +0000 |
|---|---|---|
| committer | Peter Stone <thepeterstone@gmail.com> | 2026-07-10 03:30:37 +0000 |
| commit | 2ae4f511332f02f8c5c1779f7d097956f3ea31fd (patch) | |
| tree | b1c4c7f25c20136c1411f9674b0579c436907cc8 /docs/superpowers | |
| parent | 2c8a5c17dd43bee329f62a9c662d04571b9cd2d7 (diff) | |
docs: add implementation plan for generalize-finalize-arbitration (piece 4b-2)
Diffstat (limited to 'docs/superpowers')
| -rw-r--r-- | docs/superpowers/plans/2026-07-10-generalize-finalize-arbitration.md | 825 |
1 files changed, 825 insertions, 0 deletions
diff --git a/docs/superpowers/plans/2026-07-10-generalize-finalize-arbitration.md b/docs/superpowers/plans/2026-07-10-generalize-finalize-arbitration.md new file mode 100644 index 0000000..64ea7c9 --- /dev/null +++ b/docs/superpowers/plans/2026-07-10-generalize-finalize-arbitration.md @@ -0,0 +1,825 @@ +# Generalize Finalize Arbitration Implementation Plan + +> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking. + +**Goal:** `finalizeArbitration` works on any builder-role node in a story's tree — root or nested — not just the story root. Idempotency becomes structural (keyed to the specific arbitration task, via the story's own event stream) instead of `story.Status`-based. `story.Status` is only ever touched when the node being finalized IS the actual root position. A rejected nested node gets its fix-attempt spawned directly (no story-level field to poll, unlike the root's existing `NEEDS_FIX`/`ensureFixAttempt` flow, which is untouched). Evaluator and arbitration instructions gain the acceptance-criteria fallback the design always intended (`Task.AcceptanceCriteria`, falling back to the story's own) but which was never actually wired up. + +**Architecture:** `spawnRoleTask` gains a `parentTaskID` parameter (all 4 existing call sites pass `""`, preserving today's top-level-only behavior; a new nested-fix-attempt call site passes the rejected node's own `ParentTaskID`). `ensureArbitration` gains a `node *task.Task` parameter so its instructions can read that node's own `AcceptanceCriteria` (falling back to the story's). `finalizeArbitration` gains a `ctx` parameter, is renamed internally from `root` to `node`, and branches on `node.ParentTaskID == ""` to decide whether to touch `story.Status` and which rejection path to take. A new `spawnNestedFixAttempt` helper is `ensureFixAttempt`'s direct-spawn counterpart for positions with no story-level anchor to poll. + +**Tech Stack:** Go, `internal/scheduler` package only. `processStory` itself is NOT restructured into a tree-walk here — that is a separate, later plan (piece 4b-3). This plan only generalizes the mechanism itself and updates `processStory`'s two existing call sites to the new signatures, so the story root keeps working exactly as it does today, verified by the full existing test suite passing unchanged. + +## Global Constraints + +- This is piece 4b-2 of `docs/superpowers/specs/2026-07-09-recursive-arbitrated-review-design.md`'s piece 4, following piece 4b-1 (which made a nested `builder`-role task go `READY` instead of `COMPLETED` in `internal/executor`). Piece 4b-3 (the tree-walk wiring `processStory` to discover every `READY` builder node at any depth) is a separate, later plan. +- Root vs. nested plumbing asymmetry is intentional, not a forbidden special case: the review mechanism itself (evaluate → arbitrate → approve-or-reject → `COMPLETED` means verified) is 100% identical at every depth, achieved by both root and nested nodes going through the exact same `ensureEvaluators`/`ensureArbitration`/`finalizeArbitration` functions with identical approve/reject logic. Only the notification wiring for a *rejection* differs: the root has a natural, already-existing anchor (`story.Status`) to poll via `ensureFixAttempt`, unchanged; a nested position has no such row to poll, so its fix-attempt is spawned directly instead. Same semantic outcome (a same-role, same-position fix-attempt, capped by `maxFixAttempts`), different plumbing suited to where a story-level anchor genuinely exists. +- No changes to `ensureFixAttempt`, `fixAttemptDepth`, `rejectionReasoning`, `findEvaluators`, `findArbitration`, `taskTree`, `processRetro`'s own logic, `autoAccept`, `maybeEmitVerdict`, `arbitrationVerdict`, `dependsOnAll`, `formatAcceptanceCriteria` — verify by running the full existing test suite and confirming these functions' own tests are unaffected. +- No change to `processStory`'s overall shape or the `NEEDS_FIX` early-return at its top — only its two internal call sites (`ensureArbitration`, `finalizeArbitration`) gain the new parameters this plan adds. + +--- + +### Task 1: Generalize `finalizeArbitration`, add nested fix-attempt spawning, wire in acceptance-criteria fallback + +**Files:** +- Modify: `internal/scheduler/story_orchestrator.go` (`spawnRoleTask`, `ensureEvaluators`, `ensureArbitration`, `finalizeArbitration`, `ensureFixAttempt`, `processRetro`, `processStory`; add `spawnNestedFixAttempt`) +- Modify: `internal/scheduler/story_orchestrator_test.go` (5 new tests) + +**Interfaces:** +- Consumes: `task.CurrentAttempt`, `o.fixAttemptDepth`, `o.rejectionReasoning`, `o.arbitrationVerdict`, `event.KindArbitrationDecided`/`event.KindVerdictReported` (all unchanged). +- Produces: `spawnRoleTask(ctx, name, roleName string, dependsOn []string, parentTaskID string, template *task.Task, instructions string) (*task.Task, error)` — new `parentTaskID` parameter, fifth positional argument. `ensureArbitration(ctx, st, node *task.Task, evaluators []*task.Task) (*task.Task, bool)` — new `node` parameter, third positional argument. `finalizeArbitration(ctx, st, node, arbitration *task.Task)` — new `ctx` parameter, first positional argument (parameter formerly named `root` is now `node`). `spawnNestedFixAttempt(ctx, st, node *task.Task, reasoning string)` — new. + +- [ ] **Step 1: Add `parentTaskID` to `spawnRoleTask`** + +In `internal/scheduler/story_orchestrator.go`, find: + +```go +func (o *StoryOrchestrator) spawnRoleTask(ctx context.Context, name, roleName string, dependsOn []string, template *task.Task, instructions string) (*task.Task, error) { + nt := &task.Task{ + ID: uuid.NewString(), + Name: name, + Project: template.Project, + RepositoryURL: template.RepositoryURL, + Agent: task.AgentConfig{ + Role: roleName, + Instructions: instructions, + }, + Priority: task.PriorityNormal, + Tags: []string{"story-orchestrator"}, + DependsOn: dependsOn, + Retry: task.RetryConfig{MaxAttempts: 1, Backoff: "exponential"}, + State: task.StatePending, + } +``` + +Replace with: + +```go +func (o *StoryOrchestrator) spawnRoleTask(ctx context.Context, name, roleName string, dependsOn []string, parentTaskID string, template *task.Task, instructions string) (*task.Task, error) { + nt := &task.Task{ + ID: uuid.NewString(), + Name: name, + ParentTaskID: parentTaskID, + Project: template.Project, + RepositoryURL: template.RepositoryURL, + Agent: task.AgentConfig{ + Role: roleName, + Instructions: instructions, + }, + Priority: task.PriorityNormal, + Tags: []string{"story-orchestrator"}, + DependsOn: dependsOn, + Retry: task.RetryConfig{MaxAttempts: 1, Backoff: "exponential"}, + State: task.StatePending, + } +``` + +Also update this function's doc comment. Find: + +```go +// spawnRoleTask creates a new role-typed, top-level task (no ParentTaskID — +// these are DAG siblings via DependsOn, not delegated subtasks; see +// internal/executor.Pool.cascadeFail's doc comment for why that distinction +// matters) and submits it to the pool. Agent.Type/Model are left empty so +// Phase 5's role-based dispatch resolves them from the role's escalation +// ladder on first dispatch (internal/executor.Pool.execute()); if no active +// role_configs row exists for the role, that same code path logs a warning +// and dispatches without role resolution — an accepted degraded mode we +// don't special-case here either. +``` + +Replace with: + +```go +// spawnRoleTask creates a new role-typed task and submits it to the pool. +// parentTaskID is "" for the DAG-sibling tasks every caller except +// spawnNestedFixAttempt creates (Evaluators, Arbitration, Retro, and the +// root's own fix-attempts spawned by ensureFixAttempt are always top-level — +// see internal/executor.Pool.cascadeFail's doc comment for why that +// distinction matters); spawnNestedFixAttempt is the one caller that passes +// a non-empty parentTaskID, so a rejected nested node's fix-attempt lands at +// the same tree position as the node it supersedes (required for +// task.CurrentAttempt's ParentTaskID-equality matching to resolve it). +// Agent.Type/Model are left empty so Phase 5's role-based dispatch resolves +// them from the role's escalation ladder on first dispatch +// (internal/executor.Pool.execute()); if no active role_configs row exists +// for the role, that same code path logs a warning and dispatches without +// role resolution — an accepted degraded mode we don't special-case here +// either. +``` + +- [ ] **Step 2: Update `ensureEvaluators`'s call site and add the acceptance-criteria fallback** + +In the same file, find: + +```go + spawnedAny := false + for _, r := range evaluatorRoles { + if _, ok := found[r]; ok { + continue + } + nt, err := o.spawnRoleTask(ctx, fmt.Sprintf("%s: %s", r, st.Name), r, []string{root.ID}, root, + fmt.Sprintf("Evaluate the changes made by task %s against the %q dimension for story %q.\n\nStory spec:\n%s", root.ID, r, st.Name, st.Spec)) + if err != nil { + o.logf("story orchestrator: spawn evaluator", "storyID", st.ID, "role", r, "error", err) + continue + } + found[r] = nt + spawnedAny = true + } +``` + +Replace with: + +```go + criteria := root.AcceptanceCriteria + if len(criteria) == 0 { + criteria = st.AcceptanceCriteria + } + + spawnedAny := false + for _, r := range evaluatorRoles { + if _, ok := found[r]; ok { + continue + } + nt, err := o.spawnRoleTask(ctx, fmt.Sprintf("%s: %s", r, st.Name), r, []string{root.ID}, "", root, + fmt.Sprintf("Evaluate the changes made by task %s against the %q dimension for story %q.\n\nStory spec:\n%s\n\nAcceptance criteria:\n%s", root.ID, r, st.Name, st.Spec, formatAcceptanceCriteria(criteria))) + if err != nil { + o.logf("story orchestrator: spawn evaluator", "storyID", st.ID, "role", r, "error", err) + continue + } + found[r] = nt + spawnedAny = true + } +``` + +- [ ] **Step 3: Generalize `ensureArbitration`** + +In the same file, find the full function: + +```go +// ensureArbitration returns the single Arbitration task depending on all 4 +// evaluators, spawning it if it doesn't exist yet. Idempotency is again +// structural: it looks for an existing "planner"-role dependent of the first +// evaluator task that depends on every evaluator ID, rather than relying on +// story.Status (which a human can freely rewrite via PUT /api/stories/{id}). +func (o *StoryOrchestrator) ensureArbitration(ctx context.Context, st *story.Story, evaluators []*task.Task) (*task.Task, bool) { + ids := make([]string, len(evaluators)) + for i, ev := range evaluators { + ids[i] = ev.ID + } + + dependents, err := o.Store.ListDependents(evaluators[0].ID) + if err != nil { + o.logf("story orchestrator: list evaluator dependents", "storyID", st.ID, "error", err) + return nil, false + } + for _, d := range dependents { + if d.Agent.Role == arbitrationRole && dependsOnAll(d, ids) { + return d, true + } + } + + instructions := fmt.Sprintf( + "Arbitrate the 4 evaluator verdicts for story %q (task %s). Read each evaluator task's summary/events "+ + "and decide whether the story is ready to ship. Acceptance criteria:\n%s", + st.Name, st.ID, formatAcceptanceCriteria(st.AcceptanceCriteria)) + nt, err := o.spawnRoleTask(ctx, "Arbitration: "+st.Name, arbitrationRole, ids, evaluators[0], instructions) + if err != nil { + o.logf("story orchestrator: spawn arbitration", "storyID", st.ID, "error", err) + return nil, false + } + return nt, true +} +``` + +Replace with: + +```go +// ensureArbitration returns the single Arbitration task depending on all 4 +// evaluators, spawning it if it doesn't exist yet. Idempotency is again +// structural: it looks for an existing "planner"-role dependent of the first +// evaluator task that depends on every evaluator ID, rather than relying on +// story.Status (which a human can freely rewrite via PUT /api/stories/{id}). +// node is the task being evaluated (root or nested) -- used only to read its +// own AcceptanceCriteria, falling back to the story's if node has none set. +func (o *StoryOrchestrator) ensureArbitration(ctx context.Context, st *story.Story, node *task.Task, evaluators []*task.Task) (*task.Task, bool) { + ids := make([]string, len(evaluators)) + for i, ev := range evaluators { + ids[i] = ev.ID + } + + dependents, err := o.Store.ListDependents(evaluators[0].ID) + if err != nil { + o.logf("story orchestrator: list evaluator dependents", "storyID", st.ID, "error", err) + return nil, false + } + for _, d := range dependents { + if d.Agent.Role == arbitrationRole && dependsOnAll(d, ids) { + return d, true + } + } + + criteria := node.AcceptanceCriteria + if len(criteria) == 0 { + criteria = st.AcceptanceCriteria + } + instructions := fmt.Sprintf( + "Arbitrate the 4 evaluator verdicts for task %q (%s) within story %q. Read each evaluator task's summary/events "+ + "and decide whether this task is ready to ship. Acceptance criteria:\n%s", + node.Name, node.ID, st.Name, formatAcceptanceCriteria(criteria)) + nt, err := o.spawnRoleTask(ctx, "Arbitration: "+st.Name, arbitrationRole, ids, "", evaluators[0], instructions) + if err != nil { + o.logf("story orchestrator: spawn arbitration", "storyID", st.ID, "error", err) + return nil, false + } + return nt, true +} +``` + +- [ ] **Step 4: Generalize `finalizeArbitration` and add `spawnNestedFixAttempt`** + +In the same file, find the full function (including its doc comment): + +```go +// finalizeArbitration handles the Arbitration task reaching COMPLETED: it +// emits KindArbitrationDecided, then either approves or rejects root based on +// whether the arbitration task recorded a structured KindVerdictReported +// event (AgentChannel.ReportVerdict). An arbitration task that never calls +// report_verdict defaults to approval, preserving the prior behavior for +// agents that don't report a structured verdict. +// +// On approval, root is promoted READY -> COMPLETED *here* — not eagerly the +// moment it first reached READY — so COMPLETED means "verified by +// arbitration", uniformly, the same rule this design applies at every depth +// of a builder-role task tree (see +// docs/superpowers/specs/2026-07-09-recursive-arbitrated-review-design.md). +// On rejection, root is left untouched at READY: it has been superseded, and +// task.CurrentAttempt will resolve future lookups of this position forward +// to whatever fix-attempt task ensureFixAttempt spawns once the story lands +// at NEEDS_FIX below — READY already satisfies a dependent's DependsOn (see +// executor.depDoneStates), so the fix attempt is immediately dispatchable +// without root ever becoming COMPLETED. +// +// Gated on st.Status == "VALIDATING" so repeated ticks (or a story a human +// already advanced past REVIEW_READY) don't re-emit the event, re-promote +// root, or re-write the status — this is the one place in the orchestrator +// where the story's own status field, not a structural dependents check, is +// the idempotency guard, because by this stage there's nothing further to +// check structurally: the Arbitration task is the last task in the chain, so +// "does a subsequent task exist" isn't an available signal. +func (o *StoryOrchestrator) finalizeArbitration(st *story.Story, root, arbitration *task.Task) { + if st.Status != "VALIDATING" { + return + } + + approved, _, hasVerdict := o.arbitrationVerdict(st, arbitration) + + payload, _ := json.Marshal(struct { + TaskID string `json:"task_id"` + Summary string `json:"summary"` + Approved *bool `json:"approved,omitempty"` + }{TaskID: arbitration.ID, Summary: arbitration.Summary, Approved: optionalBool(approved, hasVerdict)}) + if err := o.Store.CreateEvent(&event.Event{ + TaskID: st.ID, + Kind: event.KindArbitrationDecided, + Actor: event.ActorSystem, + Payload: payload, + }); err != nil { + o.logf("story orchestrator: emit arbitration_decided", "storyID", st.ID, "error", err) + } + + if hasVerdict && !approved { + st.Status = "NEEDS_FIX" + } else { + // Approved (or no structured verdict reported -- preserves the + // prior unconditional-approve default). root only becomes COMPLETED + // here, after arbitration. + if err := o.Store.UpdateTaskState(root.ID, task.StateCompleted); err != nil { + o.logf("story orchestrator: finalize arbitration: promote root to completed", "storyID", st.ID, "taskID", root.ID, "error", err) + return + } + st.Status = "REVIEW_READY" + } + if err := o.Store.UpdateStory(st); err != nil { + o.logf("story orchestrator: update story status after arbitration", "storyID", st.ID, "status", st.Status, "error", err) + } +} +``` + +Replace with: + +```go +// finalizeArbitration handles an Arbitration task reaching COMPLETED for any +// builder-role node in a story's tree -- root or nested -- and either +// approves or rejects that node based on whether the arbitration task +// recorded a structured KindVerdictReported event +// (AgentChannel.ReportVerdict). An arbitration task that never calls +// report_verdict defaults to approval, preserving the prior behavior for +// agents that don't report a structured verdict. +// +// Idempotency is structural and keyed to this specific arbitration task, not +// story.Status (a story-level field only the root position has any claim +// to): it checks the story's own event stream for a KindArbitrationDecided +// event already referencing arbitration.ID, and does nothing further if +// found. This is what lets this same function run for every builder node in +// a story's tree without a node-specific analog of story.Status to gate on. +// +// On approval, node is promoted READY -> COMPLETED *here* -- not eagerly the +// moment it first reached READY -- so COMPLETED means "verified by +// arbitration", uniformly, at every depth (see +// docs/superpowers/specs/2026-07-09-recursive-arbitrated-review-design.md). +// On rejection, node is left untouched at READY: it has been superseded, and +// task.CurrentAttempt will resolve future lookups of this position forward +// to whatever fix-attempt gets spawned. +// +// st.Status is only ever touched here when node IS the story's root +// (node.ParentTaskID == "", true only for the position anchored at +// st.RootTaskID within this story's own tree): REVIEW_READY on approval, +// NEEDS_FIX on rejection -- exactly as before this function was +// generalized. A rejected ROOT's fix-attempt is still spawned by +// ensureFixAttempt on the next tick, polling st.Status == "NEEDS_FIX" -- +// unchanged, since a Story row is a natural, already-existing anchor for +// that trigger. A rejected NESTED node has no story-level field to poll, so +// its fix-attempt is spawned directly, right here (spawnNestedFixAttempt, +// below), respecting the same maxFixAttempts cap via fixAttemptDepth. Both +// paths produce the identical semantic outcome (a same-role, same-position +// fix-attempt, capped) through different wiring suited to where a +// story-level anchor exists (root) or doesn't (nested) -- not a special case +// in the review mechanism itself, which stays 100% identical at every depth. +func (o *StoryOrchestrator) finalizeArbitration(ctx context.Context, st *story.Story, node, arbitration *task.Task) { + events, err := o.Store.ListEvents(st.ID, 0) + if err != nil { + o.logf("story orchestrator: finalize arbitration: list story events", "storyID", st.ID, "error", err) + return + } + for _, e := range events { + if e.Kind != event.KindArbitrationDecided { + continue + } + var payload struct { + TaskID string `json:"task_id"` + } + if json.Unmarshal(e.Payload, &payload) == nil && payload.TaskID == arbitration.ID { + return // already finalized + } + } + + approved, reasoning, hasVerdict := o.arbitrationVerdict(st, arbitration) + isRoot := node.ParentTaskID == "" + + payload, _ := json.Marshal(struct { + TaskID string `json:"task_id"` + Summary string `json:"summary"` + Approved *bool `json:"approved,omitempty"` + }{TaskID: arbitration.ID, Summary: arbitration.Summary, Approved: optionalBool(approved, hasVerdict)}) + if err := o.Store.CreateEvent(&event.Event{ + TaskID: st.ID, + Kind: event.KindArbitrationDecided, + Actor: event.ActorSystem, + Payload: payload, + }); err != nil { + o.logf("story orchestrator: emit arbitration_decided", "storyID", st.ID, "error", err) + } + + if hasVerdict && !approved { + if isRoot { + st.Status = "NEEDS_FIX" + if err := o.Store.UpdateStory(st); err != nil { + o.logf("story orchestrator: update story status after arbitration", "storyID", st.ID, "status", st.Status, "error", err) + } + return + } + o.spawnNestedFixAttempt(ctx, st, node, reasoning) + return + } + + // Approved (or no structured verdict reported -- preserves the prior + // unconditional-approve default). node only becomes COMPLETED here, + // after arbitration. + if err := o.Store.UpdateTaskState(node.ID, task.StateCompleted); err != nil { + o.logf("story orchestrator: finalize arbitration: promote node to completed", "storyID", st.ID, "taskID", node.ID, "error", err) + return + } + if isRoot { + st.Status = "REVIEW_READY" + if err := o.Store.UpdateStory(st); err != nil { + o.logf("story orchestrator: update story status after arbitration", "storyID", st.ID, "status", st.Status, "error", err) + } + } +} + +// spawnNestedFixAttempt handles a rejected NESTED builder node +// (node.ParentTaskID != "") -- the counterpart to ensureFixAttempt, which +// handles the root position via st.Status == "NEEDS_FIX" polling. A nested +// position has no story-level field to poll, so its fix-attempt is spawned +// directly here, the moment rejection is discovered. Idempotency and the +// maxFixAttempts cap mirror ensureFixAttempt exactly: look for an existing +// role-matched dependent before spawning, and refuse to spawn a new one past +// the cap (leaving node's position permanently unresolved, node's own parent +// stuck BLOCKED forever -- the same "leave it for a human" resolution the +// root's own cap already accepts, per this design's Non-Goals). +func (o *StoryOrchestrator) spawnNestedFixAttempt(ctx context.Context, st *story.Story, node *task.Task, reasoning string) { + dependents, err := o.Store.ListDependents(node.ID) + if err != nil { + o.logf("story orchestrator: nested fix attempt: list node dependents", "storyID", st.ID, "nodeID", node.ID, "error", err) + return + } + for _, d := range dependents { + if d.Agent.Role == node.Agent.Role { + return // fix attempt already spawned + } + } + + if depth := o.fixAttemptDepth(node); depth >= maxFixAttempts { + o.logf("story orchestrator: nested fix attempt: max fix attempts reached, leaving node rejected for a human", "storyID", st.ID, "nodeID", node.ID, "depth", depth) + return + } + + if reasoning == "" { + reasoning = "(no arbitration reasoning found)" + } + instructions := fmt.Sprintf( + "A previous attempt at task %q (%s) within story %q was rejected by arbitration. Fix the issues and resubmit.\n\n"+ + "Acceptance criteria:\n%s\n\nArbitration's rejection reasoning:\n%s", + node.Name, node.ID, st.Name, formatAcceptanceCriteria(node.AcceptanceCriteria), reasoning) + if _, err := o.spawnRoleTask(ctx, "Fix attempt: "+node.Name, node.Agent.Role, []string{node.ID}, node.ParentTaskID, node, instructions); err != nil { + o.logf("story orchestrator: nested fix attempt: spawn", "storyID", st.ID, "nodeID", node.ID, "error", err) + } +} +``` + +- [ ] **Step 5: Update the remaining `spawnRoleTask` call sites** + +In the same file, find (inside `ensureFixAttempt`): + +```go + nt, err := o.spawnRoleTask(ctx, "Fix attempt: "+st.Name, "builder", []string{oldRoot.ID}, oldRoot, instructions) +``` + +Replace with: + +```go + nt, err := o.spawnRoleTask(ctx, "Fix attempt: "+st.Name, "builder", []string{oldRoot.ID}, "", oldRoot, instructions) +``` + +Find (inside `processRetro`): + +```go + nt, err := o.spawnRoleTask(ctx, "Retro: "+st.Name, retroRole, []string{arbitration.ID}, root, instructions) +``` + +Replace with: + +```go + nt, err := o.spawnRoleTask(ctx, "Retro: "+st.Name, retroRole, []string{arbitration.ID}, "", root, instructions) +``` + +- [ ] **Step 6: Update `processStory`'s two call sites** + +In the same file, find (inside `processStory`): + +```go + // Stage 2: Evaluators -> Arbitration. + arbitration, ok := o.ensureArbitration(ctx, st, evaluators) + if !ok { + return + } + arbitration = o.autoAccept(st, arbitration) + + // Stage 3: Arbitration -> approve (root: READY -> COMPLETED, story -> + // REVIEW_READY) or reject (story -> NEEDS_FIX; a fix attempt is spawned + // next tick by ensureFixAttempt, depending on the still-READY root -- + // task.CurrentAttempt resolves future lookups of this position forward + // to it). + if arbitration.State == task.StateCompleted { + o.finalizeArbitration(st, root, arbitration) + } +} +``` + +Replace with: + +```go + // Stage 2: Evaluators -> Arbitration. + arbitration, ok := o.ensureArbitration(ctx, st, root, evaluators) + if !ok { + return + } + arbitration = o.autoAccept(st, arbitration) + + // Stage 3: Arbitration -> approve (root: READY -> COMPLETED, story -> + // REVIEW_READY) or reject (story -> NEEDS_FIX; a fix attempt is spawned + // next tick by ensureFixAttempt, depending on the still-READY root -- + // task.CurrentAttempt resolves future lookups of this position forward + // to it). + if arbitration.State == task.StateCompleted { + o.finalizeArbitration(ctx, st, root, arbitration) + } +} +``` + +- [ ] **Step 7: Add the 5 new tests** + +In `internal/scheduler/story_orchestrator_test.go`, add these 5 tests anywhere after `seedNeedsFixStory`'s definition (e.g., immediately before `TestStoryOrchestrator_ArbitrationCompletes_EmitsDecisionAndReviewReady`, or at the end of the file — placement doesn't matter, they're self-contained): + +```go +// TestStoryOrchestrator_FinalizeArbitration_NestedNodeApproved_DoesNotTouchStoryStatus +// proves finalizeArbitration, generalized to operate on any builder-role +// node (not just the story root), promotes a NESTED node to COMPLETED on +// approval without touching story.Status at all -- REVIEW_READY is reserved +// for the actual root position (node.ParentTaskID == ""). +func TestStoryOrchestrator_FinalizeArbitration_NestedNodeApproved_DoesNotTouchStoryStatus(t *testing.T) { + store := newFakeStoryStore() + root := builderTask("nested-approve-root", task.StateBlocked) + store.tasks[root.ID] = root + st := newStoryWithRoot("nested-approve-story", root.ID, "IN_PROGRESS") + store.stories[st.ID] = st + + nested := builderTask("nested-approve-node", task.StateReady) + nested.ParentTaskID = root.ID + store.tasks[nested.ID] = nested + + arb := &task.Task{ + ID: "nested-approve-arb", + Name: "Arbitration", + Agent: task.AgentConfig{Role: arbitrationRole}, + State: task.StateCompleted, + Summary: "ship it", + } + store.tasks[arb.ID] = arb + payload, _ := json.Marshal(struct { + Approved bool `json:"approved"` + Reasoning string `json:"reasoning"` + }{Approved: true, Reasoning: "meets criteria"}) + if err := store.CreateEvent(&event.Event{TaskID: arb.ID, Kind: event.KindVerdictReported, Actor: event.ActorAgent, Payload: payload}); err != nil { + t.Fatalf("seed verdict event: %v", err) + } + + pool := &fakePool{} + orch := &StoryOrchestrator{Store: store, Pool: pool} + orch.finalizeArbitration(context.Background(), st, nested, arb) + + got, err := store.GetTask(nested.ID) + if err != nil { + t.Fatal(err) + } + if got.State != task.StateCompleted { + t.Errorf("nested node State = %v, want COMPLETED", got.State) + } + + stories, _ := store.ListStories(storage.StoryFilter{}) + var gotStory *story.Story + for _, s := range stories { + if s.ID == st.ID { + gotStory = s + } + } + if gotStory.Status != "IN_PROGRESS" { + t.Errorf("story Status = %q, want unchanged IN_PROGRESS (only root's own arbitration should touch story.Status)", gotStory.Status) + } +} + +// TestStoryOrchestrator_FinalizeArbitration_NestedNodeRejected_SpawnsFixAttemptDirectly +// proves the nested-rejection path: since a nested position has no +// story-level field to poll (unlike the root's NEEDS_FIX/ensureFixAttempt +// flow), its fix-attempt is spawned immediately, right here -- same role, +// same ParentTaskID as the rejected node (so task.CurrentAttempt resolves +// it correctly), and story.Status is left untouched. +func TestStoryOrchestrator_FinalizeArbitration_NestedNodeRejected_SpawnsFixAttemptDirectly(t *testing.T) { + store := newFakeStoryStore() + root := builderTask("nested-reject-root", task.StateBlocked) + store.tasks[root.ID] = root + st := newStoryWithRoot("nested-reject-story", root.ID, "IN_PROGRESS") + store.stories[st.ID] = st + + nested := builderTask("nested-reject-node", task.StateReady) + nested.ParentTaskID = root.ID + store.tasks[nested.ID] = nested + + arb := &task.Task{ + ID: "nested-reject-arb", + Name: "Arbitration", + Agent: task.AgentConfig{Role: arbitrationRole}, + State: task.StateCompleted, + Summary: "found a real problem", + } + store.tasks[arb.ID] = arb + payload, _ := json.Marshal(struct { + Approved bool `json:"approved"` + Reasoning string `json:"reasoning"` + }{Approved: false, Reasoning: "breaks the widget"}) + if err := store.CreateEvent(&event.Event{TaskID: arb.ID, Kind: event.KindVerdictReported, Actor: event.ActorAgent, Payload: payload}); err != nil { + t.Fatalf("seed verdict event: %v", err) + } + + pool := &fakePool{} + orch := &StoryOrchestrator{Store: store, Pool: pool} + orch.finalizeArbitration(context.Background(), st, nested, arb) + + got, err := store.GetTask(nested.ID) + if err != nil { + t.Fatal(err) + } + if got.State != task.StateReady { + t.Errorf("rejected nested node State = %v, want unchanged READY", got.State) + } + + fixAttempts := store.dependentsWithRole(nested.ID, "builder") + if len(fixAttempts) != 1 { + t.Fatalf("expected exactly 1 fix-attempt task depending on the rejected nested node, got %d", len(fixAttempts)) + } + fix := fixAttempts[0] + if fix.ParentTaskID != nested.ParentTaskID { + t.Errorf("fix attempt ParentTaskID = %q, want %q (same position as the rejected node)", fix.ParentTaskID, nested.ParentTaskID) + } + if fix.State != task.StateQueued { + t.Errorf("fix attempt State = %v, want QUEUED", fix.State) + } + + stories, _ := store.ListStories(storage.StoryFilter{}) + var gotStory *story.Story + for _, s := range stories { + if s.ID == st.ID { + gotStory = s + } + } + if gotStory.Status != "IN_PROGRESS" { + t.Errorf("story Status = %q, want unchanged IN_PROGRESS", gotStory.Status) + } +} + +// TestStoryOrchestrator_FinalizeArbitration_NestedNodeRejected_CapsAtMaxFixAttempts +// mirrors the root's own safety net (TestStoryOrchestrator_NeedsFix_CapsAtMaxFixAttempts) +// for a nested position: once a chain of maxFixAttempts consecutive fix +// attempts already exists at this position, no further fix attempt is +// spawned. +func TestStoryOrchestrator_FinalizeArbitration_NestedNodeRejected_CapsAtMaxFixAttempts(t *testing.T) { + store := newFakeStoryStore() + root := builderTask("nested-cap-root", task.StateBlocked) + store.tasks[root.ID] = root + st := newStoryWithRoot("nested-cap-story", root.ID, "IN_PROGRESS") + store.stories[st.ID] = st + + nested := builderTask("nested-cap-node", task.StateReady) + nested.ParentTaskID = root.ID + store.tasks[nested.ID] = nested + + pool := &fakePool{} + orch := &StoryOrchestrator{Store: store, Pool: pool} + + // Build a chain of maxFixAttempts prior fix-attempt tasks at this same + // nested position, ending at prev -- simulating a position that has + // already exhausted its automatic fix attempts. + prev := nested + for i := 0; i < maxFixAttempts; i++ { + nt, err := orch.spawnRoleTask(context.Background(), fmt.Sprintf("Fix attempt %d", i), "builder", []string{prev.ID}, root.ID, root, "fix it") + if err != nil { + t.Fatalf("seed fix-attempt chain: %v", err) + } + prev = nt + } + + arb := &task.Task{ + ID: "nested-cap-arb", + Name: "Arbitration", + Agent: task.AgentConfig{Role: arbitrationRole}, + State: task.StateCompleted, + Summary: "still broken", + } + store.tasks[arb.ID] = arb + payload, _ := json.Marshal(struct { + Approved bool `json:"approved"` + Reasoning string `json:"reasoning"` + }{Approved: false, Reasoning: "still broken"}) + if err := store.CreateEvent(&event.Event{TaskID: arb.ID, Kind: event.KindVerdictReported, Actor: event.ActorAgent, Payload: payload}); err != nil { + t.Fatalf("seed verdict event: %v", err) + } + + orch.finalizeArbitration(context.Background(), st, prev, arb) + + fixAttempts := store.dependentsWithRole(prev.ID, "builder") + if len(fixAttempts) != 0 { + t.Fatalf("expected no new fix-attempt task once maxFixAttempts is reached, got %d", len(fixAttempts)) + } +} + +// TestStoryOrchestrator_FinalizeArbitration_RootStillSetsStoryStatus proves +// the generalized finalizeArbitration still drives story.Status exactly as +// before when node IS the root position (node.ParentTaskID == "") -- the +// one case where touching story.Status is still correct, now decided by +// node.ParentTaskID rather than being the function's only mode. +func TestStoryOrchestrator_FinalizeArbitration_RootStillSetsStoryStatus(t *testing.T) { + store := newFakeStoryStore() + root := builderTask("root-status-root", task.StateReady) + store.tasks[root.ID] = root + st := newStoryWithRoot("root-status-story", root.ID, "VALIDATING") + store.stories[st.ID] = st + + arb := &task.Task{ + ID: "root-status-arb", + Name: "Arbitration", + Agent: task.AgentConfig{Role: arbitrationRole}, + State: task.StateCompleted, + Summary: "ship it", + } + store.tasks[arb.ID] = arb + payload, _ := json.Marshal(struct { + Approved bool `json:"approved"` + Reasoning string `json:"reasoning"` + }{Approved: true, Reasoning: "meets criteria"}) + if err := store.CreateEvent(&event.Event{TaskID: arb.ID, Kind: event.KindVerdictReported, Actor: event.ActorAgent, Payload: payload}); err != nil { + t.Fatalf("seed verdict event: %v", err) + } + + pool := &fakePool{} + orch := &StoryOrchestrator{Store: store, Pool: pool} + orch.finalizeArbitration(context.Background(), st, root, arb) + + gotRoot, err := store.GetTask(root.ID) + if err != nil { + t.Fatal(err) + } + if gotRoot.State != task.StateCompleted { + t.Errorf("root State = %v, want COMPLETED", gotRoot.State) + } + + stories, _ := store.ListStories(storage.StoryFilter{}) + var gotStory *story.Story + for _, s := range stories { + if s.ID == st.ID { + gotStory = s + } + } + if gotStory.Status != "REVIEW_READY" { + t.Errorf("story Status = %q, want REVIEW_READY (root's own approval must still drive it)", gotStory.Status) + } +} + +// TestStoryOrchestrator_FinalizeArbitration_IdempotentViaArbitrationDecidedEvent +// proves the generalized idempotency check (a KindArbitrationDecided event +// already referencing this arbitration.ID) correctly no-ops a repeated call +// for the same already-decided arbitration -- the mechanism every node in a +// story's tree now shares, replacing the root-only story.Status == +// "VALIDATING" gate. +func TestStoryOrchestrator_FinalizeArbitration_IdempotentViaArbitrationDecidedEvent(t *testing.T) { + store := newFakeStoryStore() + root := builderTask("idempotent-root", task.StateReady) + store.tasks[root.ID] = root + st := newStoryWithRoot("idempotent-story", root.ID, "VALIDATING") + store.stories[st.ID] = st + + arb := &task.Task{ + ID: "idempotent-arb", + Name: "Arbitration", + Agent: task.AgentConfig{Role: arbitrationRole}, + State: task.StateCompleted, + Summary: "ship it", + } + store.tasks[arb.ID] = arb + payload, _ := json.Marshal(struct { + Approved bool `json:"approved"` + Reasoning string `json:"reasoning"` + }{Approved: true, Reasoning: "meets criteria"}) + if err := store.CreateEvent(&event.Event{TaskID: arb.ID, Kind: event.KindVerdictReported, Actor: event.ActorAgent, Payload: payload}); err != nil { + t.Fatalf("seed verdict event: %v", err) + } + + pool := &fakePool{} + orch := &StoryOrchestrator{Store: store, Pool: pool} + orch.finalizeArbitration(context.Background(), st, root, arb) + orch.finalizeArbitration(context.Background(), st, root, arb) + orch.finalizeArbitration(context.Background(), st, root, arb) + + decided := store.eventsOfKind(event.KindArbitrationDecided) + if len(decided) != 1 { + t.Fatalf("expected exactly 1 arbitration_decided event after 3 calls, got %d", len(decided)) + } +} +``` + +- [ ] **Step 8: Commit locally NOW (before further verification)** + +Run these exact commands, in order: + +```bash +git add -A +git commit -m "refactor(scheduler): generalize finalizeArbitration to any builder-role node, add nested fix-attempt spawning and acceptance-criteria fallback" +git status +git log --oneline -1 +``` + +Do NOT run `git push`. This repo's own bare git remote is not mounted inside dispatch containers. Per this repo's own CLAUDE.md ("After a successful run, new commits are pushed from the workspace"), claudomator's ContainerRunner pushes your commits externally once your run finishes successfully with a clean working tree — your only job is a clean local commit and a clean working tree at teardown. Do this as soon as all edits (Steps 1-7) are done, before spending time on the verification steps below. + +- [ ] **Step 9: Run the full scheduler package test suite** + +Run: `go test ./internal/scheduler/... -v -run 'TestStoryOrchestrator'` +Expected: PASS for every test in the file, including the 5 new ones above. Every pre-existing test must ALSO pass completely unchanged: `TestStoryOrchestrator_SpawnsEvaluators_WhenBuilderReady`, `TestStoryOrchestrator_DoesNotDuplicateEvaluators`, `TestStoryOrchestrator_SpawnsArbitration_WhenAllEvaluatorsComplete`, `TestStoryOrchestrator_DoesNotSpawnArbitration_UntilAllEvaluatorsComplete`, `TestStoryOrchestrator_EmitsEvalVerdict_OncePerEvaluator`, `TestStoryOrchestrator_ArbitrationCompletes_EmitsDecisionAndReviewReady`, `TestStoryOrchestrator_ArbitrationRejects_SetsStoryNeedsFix`, `TestStoryOrchestrator_ArbitrationApproves_SetsReviewReady`, the 3 fix-loop tests (`TestStoryOrchestrator_NeedsFix_SpawnsFixAttempt`, `TestStoryOrchestrator_EnsureFixAttempt_FindsExistingSpawnedAttempt`, `TestStoryOrchestrator_NeedsFix_CapsAtMaxFixAttempts`), `TestStoryOrchestrator_ArbitrationRejects_LeavesRootReadyNotCompleted`, `TestStoryOrchestrator_ReadyBuilder_SpawnsEvaluators_StaysReadyUntilApproved`, `TestStoryOrchestrator_AutoAccept_DoesNotTouchUnrelatedReadyTask`, `TestStoryOrchestrator_SkipsTerminalStories`, `TestStoryOrchestrator_EndToEnd`, and the full retro test cluster. Paste the actual output. If anything fails, fix it, then repeat Step 8's commit (a new commit is fine, don't amend). + +- [ ] **Step 10: Run the full repo test suite** + +Run: `go build ./...` then `go test ./...` (the entire repo). Both must pass — paste the actual output. If `internal/api` flakes under the race detector on a one-off "sql: database is closed" teardown error unrelated to this task's actual changes, rerun once before treating it as real. + +Also run `gofmt -l internal/scheduler/story_orchestrator.go internal/scheduler/story_orchestrator_test.go` and `gofmt -w` any file it flags that wasn't already gofmt-dirty before your changes (check with `git show HEAD~1:<path> | gofmt -l -` if unsure). If you make any further fixes here, commit them too (again, no push). + +## Mandatory verification disclosure + +When you call report_summary, paste the actual terminal output of every command above — literal pass/fail counts, not a claim of success — including the full-repo `go test ./...` run and the final `git status`/`git log` confirmation showing a clean tree with your commit(s) present. If anything here is ambiguous or conflicts with what you find in the actual repo, call ask_user and describe the specific conflict — don't guess or silently decide. |
