sortie-dogs 0.13.1 → 0.13.3

package/README.md

@@ -14,6 +14,9 @@ implementation, validation, review, and model routing.
   continuation, remediation, and restart.
 - **Adaptive execution**: small work stays small; additional agents and stronger
   models are used only when task shape or risk justifies them.
+- **Clear Worker instructions**: give Luna a concise goal, explicit constraints
+  and completion criteria; keep orchestration bookkeeping in the harness.
+  See the [instruction design principles](docs/worker-instruction-design.md).
 - **Coexistence**: Sortie activates only when selected and preserves normal
   OpenCode agents, settings, and user-owned files.
 - **Cost, time, and proof**: the objective is a verified result at the lowest
@@ -78,6 +81,20 @@ internal children and must not be selected as task entry points.
 model routing. A new session alone does not reload an updated
 plugin process, so restart OpenCode after installation or upgrade.
 
+## v0.13.3 runtime updates
+
+The current release integrates native background Mission execution, concise authoritative Worker
+handoffs, and same-context Reviewer corrections from PRs #148/#149 and their Ubuntu remediation.
+After finding defects, the original Reviewer can correct and explicitly self-recheck in the same
+native session, retaining formal checks, current-source evidence, Git delivery and cumulative budget.
+Author self-recheck is recorded as non-independent; a different Reviewer is conditional on a concrete
+reachable residual Major risk. Unresolved Major or Medium findings still block acceptance.
+
+The installed native correction fixture succeeded, but the original Anko measurements remain
+unaccepted, including the latest 25-minute run. This release does not claim general speedup,
+original-task completion or a new SWE-bench score. See the [release notes](docs/release-v0.13.3.md)
+and [retained implementation and measurement history](docs/anko-pr149-ubuntu-handoff.md).
+
 ## v0.12.0 workflow
 
 v0.12.0 keeps Operator → Coordinator → Worker, with independent review:
@@ -111,6 +128,8 @@ Official SWE-bench Lite `dev` results on the same 23 public instances:
 | v0.12.16 (`9b05a34` release; `b1a6c0e` runner) | 4 / 23 (17.4%) | 5 | Fresh 23-task run; eight slots; 40-minute timeout | [Campaign](docs/swebench-v01216-dev23-2026-09-27.md) |
 | v0.12.19 (`24f5386` release; matched rerun) | 8 / 23 (34.8%) | 0 | Eight slots; effective $2/instance; 40-minute timeout; one inference timeout | [Official result and provenance](docs/benchmarks/swebench-v01220-operation-observability-2026-09-28.md) |
 | v0.12.20 (`628eb81` release) | 7 / 23 (30.4%) | 0 | Eight slots; effective $2/instance; 40-minute timeout | [Official result and caveat](docs/benchmarks/swebench-v01220-operation-observability-2026-09-28.md) |
+| v0.12.25 (`49eb1e4` release) | 7 / 23 (30.4%) | 0 | Eight slots; $2/instance; $30 total cap; 20-minute progress check / 40-minute hard maximum | [Comparison baseline](#v0131-dev23-2026-09-30) |
+| v0.13.1 (`d19e8be` release; 2026-09-30) | **8 / 23 (34.8%)** | 0 | Eight slots; $2/instance; $46 total cap; 20-minute progress check / 40-minute hard maximum; GPT-6.1 Sol + Luna Fast | [Run summary](#v0131-dev23-2026-09-30) |
 
 Every row has 23 submitted official predictions; an empty patch counts against
 the score, not as a missing evaluation. The v0.10.14 report does not separately
@@ -132,6 +151,39 @@ single run-to-run difference does not establish causation.
 
 Historical qualification references remain in [benchmark reference](docs/benchmark-reference.md).
 
+#### v0.13.1 dev23 (2026-09-30)
+
+One fresh pass@1 run and one official SWE-bench harness evaluation resolved
+**8/23**, versus **7/23** for v0.12.25. The new resolution was
+`pylint-dev__astroid-1333`; all seven previously resolved IDs were retained.
+Resolved by repository: marshmallow **2/2**, pvlib **0/5**, pydicom **2/5**,
+astroid **3/5**, pyvista **0/1**, sqlfluff **1/5**.
+
+- The scored row is the user-requested fresh run after a host restart. The
+  interrupted initial run is excluded from this score; the fresh run made one
+  attempt per instance with no inference retry.
+- The dataset revision (`6ec7bb89b9342f664a54a6e0a6ea6501d3437cc2`), public rows,
+  and all 23 official evaluation image IDs match the v0.12.25 run. Both used
+  `official-image-testbed` and sequential official scoring.
+- Operator/Coordinator/Reviewer/Advisor defaults changed to
+  `openai/gpt-6.1-sol#xhigh`. Actual task Workers remained
+  `openai/gpt-6-luna-fast#max`, observed across all 23 instances. The harness and
+  total budget also changed, so the extra resolution cannot be attributed to
+  the model change alone.
+- Inference ended with 20 normal completions, two timeouts
+  (`pvlib__pvlib-python-1154`, `sqlfluff__sqlfluff-1763`) and one agent failure
+  (`pvlib__pvlib-python-1854`). All patches, including stopped attempts, were
+  officially scored: 23 completed evaluations, zero empty patches and zero
+  official evaluation errors or infrastructure failures.
+- Known estimated inference cost: **$17.73**; separate unknown-usage hold:
+  **$1.98**, not counted as known expense. Inference wall time was about
+  **81 minutes**, followed by **7.2 minutes** of official scoring.
+- Fixed release commit: `d19e8be0d21180cc23ad2ae4b853d846a18e77bc`;
+  package SHA-256: `99300ceec0c3eee4fa1f984fed50d15d58b2ff4455df0041850ecd63514b0a51`.
+  OpenCode **2.0.20**, official harness **5.0.2**. Local evidence is retained in
+  `_testenv/swebench-v0131-dev23-20260930-r2/result-summary.json`; generated
+  predictions, databases and raw logs are not committed.
+
 ## Mission tools
 
 1. `start_mission`: Operator supplies concise requirements; the host saves original messages and returns a Coordinator task.

package/dist/asset-version.d.ts

@@ -3,5 +3,5 @@
  * installed project marker without importing every asset body.
  */
 export declare const RUNTIME_ASSET_VERSION = "0.3.89-completion-proof-v1";
-export declare const V010_RUNTIME_ASSET_VERSION = "0.13.1-quality-first-review-v1";
+export declare const V010_RUNTIME_ASSET_VERSION = "0.13.3-reviewer-context-v1";
 export type RuntimeAssetVersion = typeof RUNTIME_ASSET_VERSION | typeof V010_RUNTIME_ASSET_VERSION;

package/dist/asset-version.js

@@ -3,4 +3,4 @@
  * installed project marker without importing every asset body.
  */
 export const RUNTIME_ASSET_VERSION = "0.3.89-completion-proof-v1";
-export const V010_RUNTIME_ASSET_VERSION = "0.13.1-quality-first-review-v1";
+export const V010_RUNTIME_ASSET_VERSION = "0.13.3-reviewer-context-v1";

package/dist/core/contract-limits.d.ts

@@ -1,8 +1,13 @@
 /** Common text bounds shared by handoff, manifest and goal evidence validation. */
 export declare const CONTRACT_TEXT_LIMITS: Readonly<{
     title: 160;
-    objective: 2000;
+    objective: 32768;
     statement: 1000;
     command: 8192;
     path: 512;
 }>;
+/** New Mission task generation only; persisted/legacy contracts retain their original bounds. */
+export declare const MISSION_OBJECTIVE_LIMITS: Readonly<{
+    target: 2000;
+    maximum: 3000;
+}>;

package/dist/core/contract-limits.js

@@ -1,2 +1,4 @@
 /** Common text bounds shared by handoff, manifest and goal evidence validation. */
-export const CONTRACT_TEXT_LIMITS = Object.freeze({ title: 160, objective: 2000, statement: 1000, command: 8192, path: 512 });
+export const CONTRACT_TEXT_LIMITS = Object.freeze({ title: 160, objective: 32768, statement: 1000, command: 8192, path: 512 });
+/** New Mission task generation only; persisted/legacy contracts retain their original bounds. */
+export const MISSION_OBJECTIVE_LIMITS = Object.freeze({ target: 2000, maximum: 3000 });

package/dist/core/goal-bound.d.ts

@@ -39,6 +39,13 @@ export interface GoalEvidence {
         readonly candidate_paths: readonly string[];
         /** Missing on legacy evidence: retain its original all-paths snapshot recipe. */
         readonly source_policy?: "project-files-v1" | "declared-paths-v1";
+        /** Fixed when validation starts. Full manifest_hash still identifies the historical execution contract. */
+        readonly freshness?: {
+            readonly contract_hash: string;
+            readonly scratch_paths: readonly string[];
+            readonly protected_paths: readonly string[];
+            readonly environment: Readonly<Record<string, string | null>>;
+        };
     };
     readonly execution: {
         readonly command: readonly string[];

package/dist/core/goal-bound.js

@@ -64,7 +64,13 @@ export function validGoalEvidence(value, state) {
         protectedBinding.source_paths.every(text) && Array.isArray(protectedBinding.candidate_paths) &&
         protectedBinding.candidate_paths.every(text) &&
         (protectedBinding.source_policy === undefined || protectedBinding.source_policy === "project-files-v1" ||
-            protectedBinding.source_policy === "declared-paths-v1");
+            protectedBinding.source_policy === "declared-paths-v1") &&
+        (protectedBinding.freshness === undefined || (protectedBinding.freshness !== null && typeof protectedBinding.freshness === "object" &&
+            HASH.test(protectedBinding.freshness.contract_hash) &&
+            Array.isArray(protectedBinding.freshness.scratch_paths) && protectedBinding.freshness.scratch_paths.every(text) &&
+            Array.isArray(protectedBinding.freshness.protected_paths) && protectedBinding.freshness.protected_paths.every(text) &&
+            protectedBinding.freshness.environment !== null && typeof protectedBinding.freshness.environment === "object" &&
+            !Array.isArray(protectedBinding.freshness.environment) && Object.values(protectedBinding.freshness.environment).every(value => value === null || typeof value === "string")));
     const matches = criteria.length > 0 && criteria.length === value.measurement.criterion_ids.length &&
         criteria.every((criterion) => criterion.target === value.measurement.target &&
             criterion.entrypoint === value.measurement.entrypoint && criterion.workload === value.measurement.workload &&

package/dist/core/operator-mission.d.ts

@@ -1,4 +1,4 @@
-import { type OperatorPlan, type OperatorState, type OperatorTask } from "./operator-runtime.js";
+import { type OperatorPlan, type OperatorState, type OperatorTask, type OperatorRuntime } from "./operator-runtime.js";
 import { type RuntimeProfile } from "./runtime-profile.js";
 export declare const MISSION_REFERENCE = "SORTIE_MISSION_REF ";
 export declare const MISSION_REVIEW_REFERENCE = "SORTIE_MISSION_REVIEW_REF ";
@@ -19,6 +19,7 @@ export interface MissionEvidenceExcerpt {
 export interface MissionReviewScope {
     read: string[];
     write: string[];
+    validationBindings?: NonNullable<import("./goal-bound.js").GoalEvidence["protected_binding"]>[];
 }
 export interface MissionConsultation {
     id: string;
@@ -43,11 +44,13 @@ export interface MissionAttempt {
     predecessorAttemptID?: string | null;
     /** Fingerprint of the settled scoped candidate against which a later Rescue is proposed. */
     candidateID?: string;
-    kind: "implementation" | "normal_remediation" | "astra_rescue";
+    kind: "implementation" | "normal_remediation" | "astra_rescue" | "reviewer_correction";
     status: "pending" | "dispatched" | "succeeded" | "failed" | "cancelled" | "unconfirmed";
     callID?: string;
     childSessionID?: string;
+    dispatchFingerprint?: string;
     nativeOutcome?: "completed" | "failed" | "unknown";
+    terminal?: import("../plugin/runtime-bridge.js").MissionWorkerTerminalRecord;
     observedModel?: string;
     observedVariant?: string;
     failure?: {
@@ -86,6 +89,35 @@ export interface MissionExecution {
         result?: Record<string, unknown>;
     }[];
 }
+export interface MissionLaunchConditions {
+    entrypoint?: string;
+    inputs?: string[];
+    timeout_seconds?: number;
+    cost_limit_usd?: number;
+    benchmark_attempts?: number;
+    grading?: "none" | "official";
+    source: string;
+    applies_to: string;
+    recordedAt?: string;
+}
+/** Native terminal self-recheck by the correction author, explicitly not independent approval. */
+export interface MissionSelfRecheck {
+    runID: string;
+    source: string;
+    /** Candidate source/check identity, excluding optional excerpt presentation. */
+    candidateSource?: string;
+    author: string;
+    callID: string;
+    promptID: string;
+    messageID: string;
+    nativeOutcome: "completed";
+    result: string;
+    unresolvedFindings: string[];
+    residualMajor?: {
+        reachable_path: string;
+        consequence: string;
+    };
+}
 export interface OperatorMission {
     version: "0.12";
     id: string;
@@ -93,13 +125,23 @@ export interface OperatorMission {
     requests: MissionRequest[];
     /** Prior public conversation context, not additional immutable requirements. */
     context?: MissionContext[];
+    /** Explicit continue deliveries, keyed by the original real turn. */
+    steering?: {
+        requestID: string;
+        child: string;
+        status: "pending" | "queued";
+    }[];
     kind?: "implementation" | "operation";
     /** Native shell observations of the requested operation, separate from auxiliary checks. */
     execution?: MissionExecution;
+    /** Fixed benchmark conditions and their provenance, separate from internal Worker counters. */
+    launchConditions?: MissionLaunchConditions[];
     requirements: {
         id: string;
         text: string;
     }[];
+    /** Explicit user path prohibitions, unlike the Coordinator's estimated write list. */
+    prohibitedWrite?: string[];
     /** The current user intentionally replaced the predecessor's requirements. */
     requirementsReplaced?: boolean;
     phase: "open" | "running" | "submitted" | "completed" | "cancelled";
@@ -134,17 +176,36 @@ export interface OperatorMission {
         runID: string;
         risk: string[];
         source: string;
+        candidateSource?: string;
         task: OperatorTask | null;
+        callID?: string;
         evidence?: MissionEvidenceExcerpt[];
         requestFingerprint?: string;
-        verdict: "pending" | "PASS" | "findings" | "evidence-gaps" | "skipped-low-risk";
+        verdict: "pending" | "PASS" | "findings" | "evidence-gaps" | "skipped-low-risk" | "self-rechecked";
         result?: string;
         child?: string;
+        mode?: "independent" | "self-recheck";
+        admittedAt?: number;
+        promptID?: string;
+        selfRecheck?: MissionSelfRecheck;
         /** Completed, independent initial review for this mission, not merely an inherited child ID. */
         initialPrompt?: string;
         /** Observed evidence-only reviews; reporting only, never an acceptance threshold. */
         evidenceGapReviews?: number;
     };
+    /** Retained independently of later review generations; authors never become independent reviewers. */
+    corrections?: {
+        author: string;
+        reviewIdentity: string;
+        priorRunID: string;
+        runID: string;
+        priorSource: string;
+        findings: string;
+        initialPrompt: string;
+        baseline?: string;
+        status: "prepared" | "running" | "ready" | "failed" | "cancelled";
+        selfRecheck?: MissionSelfRecheck;
+    }[];
 }
 export declare const MISSION_CONSULTATION_LIMIT = 32;
 /** Review coverage survives a narrower replan; it is not a Worker write grant. */
@@ -153,6 +214,8 @@ export declare function missionReviewScope(previous: MissionReviewScope | undefi
 export declare function missionReviewVerdict(text: string): "PASS" | "evidence-gaps" | "findings";
 /** Whether the recorded review permits submission and acceptance of the current candidate. */
 export declare function missionReviewAccepted(review: NonNullable<OperatorMission["review"]>): boolean;
+/** A short native report, not a tag/hash-based second-review policy or an approval checklist. */
+export declare function missionSelfRecheckReport(text: string, source: string, hostBound?: boolean): Pick<MissionSelfRecheck, "unresolvedFindings" | "residualMajor"> | undefined;
 export declare function missionExecutionStatus(mission: OperatorMission): "not-required" | "not-started" | "running" | "execution-failed" | "executed";
 /** Observe the command's own terminal summary, never Worker/Reviewer prose. Scores are result data,
  * not process success. Unstructured commands retain their native exit semantics. */
@@ -169,6 +232,7 @@ export declare class OperatorMissionRuntime {
     private readonly writes;
     constructor(projectRoot: string, profile: RuntimeProfile);
     private file;
+    correctionReference(root: string, reviewIdentity: string): string;
     private load;
     /** Recover non-replacement continuations only; an explicit replacement must link the current cancelled run. */
     private loadMission;
@@ -179,6 +243,7 @@ export declare class OperatorMissionRuntime {
     read(root: string): Promise<OperatorMission | undefined>;
     required(root: string): Promise<OperatorMission>;
     capture(root: string, request: MissionRequest): Promise<void>;
+    recordLaunchConditions(root: string, raw: unknown): Promise<OperatorMission>;
     start(root: string, requirements: unknown, replaceRequirements?: boolean, options?: {
         kind?: OperatorMission["kind"];
         context?: MissionContext[];
@@ -209,7 +274,10 @@ export declare class OperatorMissionRuntime {
     brief(state: OperatorMission): string;
 }
 /** The model supplies only useful unit facts; IDs, proof projection and control documents are generated here. */
-export declare function missionPlan(mission: OperatorMission, raw: unknown): OperatorPlan;
+export declare function missionPlan(mission: OperatorMission, raw: unknown, projectRoot?: string): OperatorPlan;
+/** Compact provenance, not a new acceptance verdict or a substitute for original-request comparison. */
+export declare function missionAcceptanceSummary(mission: OperatorMission, run: OperatorState | undefined, operators: OperatorRuntime, observe?: (validation: readonly string[], child: string | null, notBefore?: number) => Promise<unknown>): Promise<Record<string, unknown>>;
+export declare function missionReviewIndependent(mission: OperatorMission, child: string | undefined): boolean;
 export declare function missionPacket(mission: OperatorMission, run?: OperatorState): Record<string, unknown>;
 /** Models forward a short capability, never recopy the host's evidence hashes and source packet. */
 export declare function missionReviewTask(mission: OperatorMission): OperatorTask;