fix(execution-policy): final-stage approval terminates the policy instead of rewinding to stage 1 (#7893) (#7936)
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - Issues can carry an embedded multi-stage `executionPolicy` (e.g. QA → CodeReviewer → CodePusher) driven by `applyIssueExecutionStageTransition` in `server/src/services/issue-execution-policy.ts` > - On approval, the next stage was picked with `nextPendingStage()`, which scans the **whole** stage list from index 0 for the first id not in `completedStageIds` > - Stage ids are regenerated whenever the embedded policy is re-sent or edited mid-flow (a supported operation — the existing "reassigns the active stage when the current participant is removed" test depends on it), so earlier `completedStageIds` can stop matching the current policy; a final-stage approve then "finds" stage 1 pending again and rebuilds a first-stage review (#7893) — an endless re-review loop that can recycle indefinitely against a moving main tip > - This pull request makes approvals advance with a forward-only scan (only stages *after* the one being approved), so approving the last stage always terminates the policy, and adds a guard so an already-completed execution state is terminal for `status=done` > - The benefit is final-stage approvals close the issue as the policy intends, with no behavior change for non-final advancement or reject/changes_requested verdicts ## Linked Issues or Issue Description Fixes #7893 ## What Changed - `server/src/services/issue-execution-policy.ts`: - New `nextPendingStageAfter(policy, completedStage, state)` helper — forward-only scan from the approved stage's index; the approval path uses it instead of `nextPendingStage()`. Approving the final stage therefore always yields `nextStage === null` → completed state → the caller's `done` flows through. - New guard: `requestedStatus === "done"` with an already-`completed` execution state returns without restarting the chain at stage 1 (closes the same loop when a stale completed state lingers). - Reject/`changes_requested` verdicts and intact-state forward advancement are untouched. - `server/src/__tests__/issue-execution-policy.test.ts`: 4 regression tests, including one that reproduces the exact rewind (regenerated stage ids + final-stage approve → previously reassigned QA at `currentStageIndex 0`; now terminal completed) and an explicit final-stage rejection test pinning the unchanged path. ## Verification - `npx vitest run server/src/__tests__/issue-execution-policy.test.ts` → 54 passed (50 pre-existing + 4 new). - `pnpm --filter @paperclipai/server typecheck` → clean. - The rewind was confirmed empirically against unmodified code first (a test asserting the buggy output passed pre-fix and flips post-fix), plus brute-forced realistic operation sequences (checkout dances, status round-trips, interim comments per the agent flow documented around #4889) to verify intact-state flows are unaffected. - Related suites (`issue-execution-policy-routes`, `issue-comment-reopen-routes`, `issues-service`, `issue-thread-interaction-routes`, `issue-agent-mutation-ownership-routes`) also pass locally. ## Risks - Behavior deliberately preserved: non-final approvals (forward scan is identical when state is intact), rejections at any stage, reopen-from-done (state cleared on reopen, fresh chain still starts at stage 1), and explicit `in_review` restarts. - The policy schema has no terminal-state field, so per the issue's Ask the policy simply terminates and the requested `done` status flows through. ## Model Used - Claude Fable 5 (`claude-fable-5`, Anthropic) via Claude Code, agentic mode with tool use (subagent implementation + independent adversarial review subagent), extended thinking enabled. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above (none found for #7893) - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] If this change affects the UI, I have included before/after screenshots (N/A — server-only change) - [x] I have updated relevant documentation to reflect my changes (N/A — internal stage-advance semantics) - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green (will confirm once CI runs on this PR) - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups (pending first review) - [x] I will address all Greptile and reviewer comments before requesting merge Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Harshit Khemani
GitHub
parent
31c59f8822
commit
3e1dc90bf2
|
|
@ -1081,6 +1081,178 @@ describe("issue execution policy transitions", () => {
|
|||
});
|
||||
});
|
||||
|
||||
describe("final stage completion terminates the policy (#7893)", () => {
|
||||
function threeStagePolicy() {
|
||||
return makePolicy([
|
||||
{ type: "review", participants: [{ type: "agent", agentId: qaAgentId }] },
|
||||
{ type: "review", participants: [{ type: "agent", agentId: ctoAgentId }] },
|
||||
{ type: "approval", participants: [{ type: "user", userId: ctoUserId }] },
|
||||
]);
|
||||
}
|
||||
|
||||
it("final-stage approval completes even when earlier completedStageIds are stale", () => {
|
||||
const policy = threeStagePolicy();
|
||||
const approvalStageId = policy.stages[2].id;
|
||||
// completedStageIds reference stage ids from a previous version of the
|
||||
// embedded policy (stage ids regenerate when the policy is re-sent or
|
||||
// edited mid-flow); only the active final stage id still matches.
|
||||
const staleStageIds = [
|
||||
"99999999-9999-4999-8999-999999999991",
|
||||
"99999999-9999-4999-8999-999999999992",
|
||||
];
|
||||
const result = applyIssueExecutionPolicyTransition({
|
||||
issue: {
|
||||
status: "in_review",
|
||||
assigneeAgentId: null,
|
||||
assigneeUserId: ctoUserId,
|
||||
executionPolicy: policy,
|
||||
executionState: {
|
||||
status: "pending",
|
||||
currentStageId: approvalStageId,
|
||||
currentStageIndex: 2,
|
||||
currentStageType: "approval",
|
||||
currentParticipant: { type: "user", userId: ctoUserId },
|
||||
returnAssignee: { type: "agent", agentId: coderAgentId },
|
||||
completedStageIds: staleStageIds,
|
||||
lastDecisionId: null,
|
||||
lastDecisionOutcome: "approved",
|
||||
},
|
||||
},
|
||||
policy,
|
||||
requestedStatus: "done",
|
||||
requestedAssigneePatch: {},
|
||||
actor: { userId: ctoUserId },
|
||||
commentBody: "Approved, ship it",
|
||||
});
|
||||
|
||||
// Must terminate the policy, not wrap around to the first stage.
|
||||
expect(result.patch.executionState).toMatchObject({
|
||||
status: "completed",
|
||||
completedStageIds: expect.arrayContaining([...staleStageIds, approvalStageId]),
|
||||
lastDecisionOutcome: "approved",
|
||||
});
|
||||
expect(result.patch.status).toBeUndefined();
|
||||
expect(result.patch.assigneeAgentId).toBeUndefined();
|
||||
expect(result.decision).toMatchObject({
|
||||
stageId: approvalStageId,
|
||||
stageType: "approval",
|
||||
outcome: "approved",
|
||||
});
|
||||
});
|
||||
|
||||
it("non-final stage approval still advances forward to the next stage", () => {
|
||||
const policy = threeStagePolicy();
|
||||
const firstStageId = policy.stages[0].id;
|
||||
const result = applyIssueExecutionPolicyTransition({
|
||||
issue: {
|
||||
status: "in_review",
|
||||
assigneeAgentId: qaAgentId,
|
||||
assigneeUserId: null,
|
||||
executionPolicy: policy,
|
||||
executionState: {
|
||||
status: "pending",
|
||||
currentStageId: firstStageId,
|
||||
currentStageIndex: 0,
|
||||
currentStageType: "review",
|
||||
currentParticipant: { type: "agent", agentId: qaAgentId },
|
||||
returnAssignee: { type: "agent", agentId: coderAgentId },
|
||||
completedStageIds: [],
|
||||
lastDecisionId: null,
|
||||
lastDecisionOutcome: null,
|
||||
},
|
||||
},
|
||||
policy,
|
||||
requestedStatus: "done",
|
||||
requestedAssigneePatch: {},
|
||||
actor: { agentId: qaAgentId },
|
||||
commentBody: "QA pass",
|
||||
});
|
||||
|
||||
expect(result.patch.status).toBe("in_review");
|
||||
expect(result.patch.assigneeAgentId).toBe(ctoAgentId);
|
||||
expect(result.patch.executionState).toMatchObject({
|
||||
status: "pending",
|
||||
currentStageId: policy.stages[1].id,
|
||||
currentStageIndex: 1,
|
||||
completedStageIds: [firstStageId],
|
||||
});
|
||||
});
|
||||
|
||||
it("final-stage changes requested still returns to the executor", () => {
|
||||
const policy = threeStagePolicy();
|
||||
const approvalStageId = policy.stages[2].id;
|
||||
const result = applyIssueExecutionPolicyTransition({
|
||||
issue: {
|
||||
status: "in_review",
|
||||
assigneeAgentId: null,
|
||||
assigneeUserId: ctoUserId,
|
||||
executionPolicy: policy,
|
||||
executionState: {
|
||||
status: "pending",
|
||||
currentStageId: approvalStageId,
|
||||
currentStageIndex: 2,
|
||||
currentStageType: "approval",
|
||||
currentParticipant: { type: "user", userId: ctoUserId },
|
||||
returnAssignee: { type: "agent", agentId: coderAgentId },
|
||||
completedStageIds: [policy.stages[0].id, policy.stages[1].id],
|
||||
lastDecisionId: null,
|
||||
lastDecisionOutcome: "approved",
|
||||
},
|
||||
},
|
||||
policy,
|
||||
requestedStatus: "in_progress",
|
||||
requestedAssigneePatch: {},
|
||||
actor: { userId: ctoUserId },
|
||||
commentBody: "Needs rework before release",
|
||||
});
|
||||
|
||||
expect(result.patch.status).toBe("in_progress");
|
||||
expect(result.patch.assigneeAgentId).toBe(coderAgentId);
|
||||
expect(result.patch.executionState).toMatchObject({
|
||||
status: "changes_requested",
|
||||
currentStageId: approvalStageId,
|
||||
lastDecisionOutcome: "changes_requested",
|
||||
});
|
||||
});
|
||||
|
||||
it("a completed execution state does not restart the workflow on done", () => {
|
||||
const policy = threeStagePolicy();
|
||||
// Completed state whose stage ids no longer match the current policy
|
||||
// (e.g. policy re-sent with regenerated ids after the chain finished).
|
||||
const result = applyIssueExecutionPolicyTransition({
|
||||
issue: {
|
||||
status: "in_review",
|
||||
assigneeAgentId: null,
|
||||
assigneeUserId: ctoUserId,
|
||||
executionPolicy: policy,
|
||||
executionState: {
|
||||
status: "completed",
|
||||
currentStageId: null,
|
||||
currentStageIndex: null,
|
||||
currentStageType: null,
|
||||
currentParticipant: null,
|
||||
returnAssignee: { type: "agent", agentId: coderAgentId },
|
||||
completedStageIds: [
|
||||
"99999999-9999-4999-8999-999999999991",
|
||||
"99999999-9999-4999-8999-999999999992",
|
||||
"99999999-9999-4999-8999-999999999993",
|
||||
],
|
||||
lastDecisionId: null,
|
||||
lastDecisionOutcome: "approved",
|
||||
},
|
||||
},
|
||||
policy,
|
||||
requestedStatus: "done",
|
||||
requestedAssigneePatch: {},
|
||||
actor: { userId: ctoUserId },
|
||||
commentBody: "Closing out",
|
||||
});
|
||||
|
||||
// No rewind to the first stage — the caller's done is allowed through.
|
||||
expect(result.patch).toEqual({});
|
||||
});
|
||||
});
|
||||
|
||||
describe("changes requested with no return assignee", () => {
|
||||
it("throws when requesting changes with no return assignee", () => {
|
||||
const policy = twoStagePolicy();
|
||||
|
|
|
|||
|
|
@ -441,6 +441,16 @@ function nextPendingStage(policy: IssueExecutionPolicy, state: IssueExecutionSta
|
|||
return policy.stages.find((stage) => !completed.has(stage.id)) ?? null;
|
||||
}
|
||||
|
||||
function nextPendingStageAfter(
|
||||
policy: IssueExecutionPolicy,
|
||||
completedStage: IssueExecutionStage,
|
||||
state: IssueExecutionState | null,
|
||||
) {
|
||||
const completed = new Set(state?.completedStageIds ?? []);
|
||||
const completedIndex = policy.stages.findIndex((stage) => stage.id === completedStage.id);
|
||||
return policy.stages.find((stage, index) => index > completedIndex && !completed.has(stage.id)) ?? null;
|
||||
}
|
||||
|
||||
function selectStageParticipant(
|
||||
stage: IssueExecutionStage,
|
||||
opts?: {
|
||||
|
|
@ -701,10 +711,12 @@ function applyIssueExecutionStageTransition(input: TransitionInput): TransitionR
|
|||
throw unprocessable("Approving a review or approval stage requires a comment");
|
||||
}
|
||||
const approvedState = buildCompletedState(existingState, activeStage);
|
||||
const nextStage = nextPendingStage(
|
||||
input.policy,
|
||||
{ ...approvedState, completedStageIds: approvedState.completedStageIds },
|
||||
);
|
||||
// Only stages after the stage being approved are advance candidates.
|
||||
// Scanning the whole policy could wrap back to the first stage when
|
||||
// earlier completedStageIds no longer match the policy (e.g. stage ids
|
||||
// were regenerated by a mid-flow policy edit), turning a final-stage
|
||||
// approval into an endless re-review loop (#7893).
|
||||
const nextStage = nextPendingStageAfter(input.policy, activeStage, approvedState);
|
||||
|
||||
if (!nextStage) {
|
||||
patch.executionState = approvedState;
|
||||
|
|
@ -810,6 +822,12 @@ function applyIssueExecutionStageTransition(input: TransitionInput): TransitionR
|
|||
return { patch };
|
||||
}
|
||||
|
||||
// A workflow whose execution already completed is terminal for approve/done:
|
||||
// closing the issue must not restart the chain at the first stage (#7893).
|
||||
if (requestedStatus === "done" && existingState?.status === COMPLETED_STATUS) {
|
||||
return { patch };
|
||||
}
|
||||
|
||||
let pendingStage =
|
||||
existingState?.status === CHANGES_REQUESTED_STATUS && currentStage
|
||||
? currentStage
|
||||
|
|
|
|||
Loading…
Reference in New Issue