sortie-dogs 0.12.18 → 0.12.19

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 = "0.12.18-review-references-retry-v1";
+export declare const V010_RUNTIME_ASSET_VERSION = "0.12.19-focused-behavior-review-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 = "0.12.18-review-references-retry-v1";
+export const V010_RUNTIME_ASSET_VERSION = "0.12.19-focused-behavior-review-v1";

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

@@ -3,6 +3,6 @@ export declare const CONTRACT_TEXT_LIMITS: Readonly<{
     title: 160;
     objective: 2000;
     statement: 1000;
-    command: 1000;
+    command: 8192;
     path: 512;
 }>;

package/dist/core/contract-limits.js

@@ -1,2 +1,2 @@
 /** 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: 1000, path: 512 });
+export const CONTRACT_TEXT_LIMITS = Object.freeze({ title: 160, objective: 2000, statement: 1000, command: 8192, path: 512 });

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

@@ -130,6 +130,7 @@ export declare class OperatorMissionRuntime {
     start(root: string, requirements: unknown, replaceRequirements?: boolean, options?: {
         kind?: OperatorMission["kind"];
         context?: MissionContext[];
+        cancelledRunID?: string;
     }): Promise<OperatorMission>;
     update(root: string, change: (state: OperatorMission) => void): Promise<OperatorMission>;
     /** Keep host-accepted criteria ahead of new text, including an exactly linked settled predecessor. */

package/dist/core/operator-mission.js

@@ -240,16 +240,19 @@ export class OperatorMissionRuntime {
             }
             if (previous)
                 await this.save(this.file(root, `.${previous.id}`), previous);
+            // A cancelled standalone/legacy run has no mission-owned runID. Only an explicit
+            // replacement may link it; normal continuation must not silently inherit its goal.
+            const predecessor = previous?.runID ?? previous?.supersededRunID ??
+                (replaceRequirements ? options.cancelledRunID : undefined);
             const state = { version: "0.12", id: `mission-${randomUUID()}`, root, requests: [request],
                 kind: options.kind ?? "implementation",
                 context: (options.context ?? []).filter(item => item.id !== request.id),
                 requirements: requirements.map((text, index) => ({ id: `R${index + 1}`, text })), phase: "open",
                 ...(replaceRequirements ? { requirementsReplaced: true } : {}),
                 coordinator: null, callID: null, dispatchOpen: false, runID: null, plans: 0, progress: [], submission: null,
-                ...(previous?.phase === "cancelled" &&
-                    (previous.runID === null || request.id !== previous.requests[0]?.id) &&
-                    (previous.runID ?? previous.supersededRunID)
-                    ? { supersededRunID: (previous.runID ?? previous.supersededRunID) } : {}) };
+                ...((previous === undefined || previous.phase === "cancelled") &&
+                    (previous?.runID === null || previous === undefined || request.id !== previous.requests[0]?.id) && predecessor
+                    ? { supersededRunID: predecessor } : {}) };
             await this.save(this.file(root), state);
             return state;
         });
@@ -356,17 +359,20 @@ export function missionPlan(mission, raw) {
                 throw new Error(`mission-unit-${index + 1}: ${field} must be paths`);
             return [...new Set(entries.map(item => normalizeExecutionScope(item)))];
         };
-        // Scheduling coverage is not semantic proof. Only a single requirement is unambiguous.
-        const ids = value.requirement_ids ?? (mission.requirements.length === 1 ? [mission.requirements[0].id] : []);
+        // A sole unit owns the whole request. This schedules work; it does not prove acceptance.
+        const ids = value.requirement_ids ?? (raw.length === 1 || mission.requirements.length === 1 ? mission.requirements.map(item => item.id) : []);
         if (!Array.isArray(ids) || ids.length === 0 || !ids.every(id => mission.requirements.some(item => item.id === id))) {
-            throw new Error(`mission-unit-${index + 1}: requirement_ids must name the related R IDs; multi-requirement work needs explicit coverage`);
+            throw new Error(`mission-unit-${index + 1}: requirement_ids must name the related R IDs; splitting multiple requirements across units needs explicit coverage`);
         }
         if (!Array.isArray(value.validation) || value.validation.length === 0 ||
             !value.validation.every(command => typeof command === "string" && command.trim())) {
             throw new Error(`mission-unit-${index + 1}: validation must contain exact commands; final command proves the unit`);
         }
+        const validation = value.validation.map(missionValidationCommand);
+        // Retain last occurrences so removing a redundant check preserves the final proof command.
         return { id: `unit-${index + 1}`, title: line("title"), objective: line("objective"), read: paths("read"), write: paths("write"),
-            validation: value.validation.map(missionValidationCommand), acceptance_indices: [...new Set(ids.map(id => mission.requirements.findIndex(item => item.id === id)))] };
+            validation: validation.filter((command, i) => validation.lastIndexOf(command) === i),
+            acceptance_indices: [...new Set(ids.map(id => mission.requirements.findIndex(item => item.id === id)))] };
     });
     // The serial engine counts new proof milestones. Units sharing the same final suite are one
     // milestone, so coalesce them instead of making the model repair a bookkeeping rejection.
@@ -462,7 +468,8 @@ export function missionReviewTraces(mission, raw) {
         throw new Error("mission-review-traces: supply concise implementation/test traces");
     }
     const grouped = raw.map(text => ({ text,
-        ids: /^\s*((?:R\d+\s*[/,]?\s*)+):/u.exec(text)?.[1]?.match(/R\d+/gu) ?? [] }));
+        ids: [...text.matchAll(/(?:^|[.;]\s+|[\r\n])\s*((?:R\d+\s*[/,]?\s*)+):/gu)]
+            .flatMap(match => match[1].match(/R\d+/gu) ?? []) }));
     if (grouped.every(item => item.ids.length === 0) && raw.length === mission.requirements.length)
         return raw;
     const unknown = grouped.flatMap(item => item.ids).filter(id => !mission.requirements.some(requirement => requirement.id === id));

package/dist/plugin/mission-review.js

@@ -13,6 +13,40 @@ import { normalizeManifestScope } from "../core/path.js";
 import { declaredArtifacts } from "./declared-artifacts.js";
 const exec = promisify(execFile);
 const record = (value) => value !== null && typeof value === "object" && !Array.isArray(value);
+/** Use the space left by short references for longer requested branches, without starving later references. */
+function focusedAllowances(sizes, budget) {
+    const allowances = sizes.map(() => 0);
+    let remaining = budget;
+    let pending = sizes.map((_, index) => index);
+    while (pending.length && remaining > 0) {
+        const share = Math.floor(remaining / pending.length);
+        const complete = pending.filter(index => sizes[index] <= share);
+        if (!complete.length) {
+            for (const index of pending)
+                allowances[index] = share;
+            for (const index of pending.slice(0, remaining - share * pending.length))
+                allowances[index]++;
+            break;
+        }
+        for (const index of complete) {
+            allowances[index] = sizes[index];
+            remaining -= sizes[index];
+        }
+        pending = pending.filter(index => !complete.includes(index));
+    }
+    return allowances;
+}
+function visibleLineCount(lines, allowance) {
+    let bytes = 0, count = 0;
+    for (const line of lines) {
+        const size = Buffer.byteLength(line);
+        if (bytes + size > allowance)
+            break;
+        bytes += size;
+        count++;
+    }
+    return count;
+}
 export async function missionReviewBaseline(directory) {
     try {
         const { stdout } = await exec("git", ["rev-parse", "--verify", "HEAD^{commit}"], { cwd: directory });
@@ -99,7 +133,8 @@ export async function completedMissionReviewPrompts(mission, profile, root, requ
 export async function missionReviewSource(directory, run, evidence = [], baseline, priorScope) {
     const scope = missionReviewScope(priorScope, run);
     const hash = createHash("sha256").update(JSON.stringify({ baseline, scope, units: run.units.map(unit => ({ unit: unit.unit, hashes: unit.hashes })) }));
-    let selected = "";
+    const writes = [...new Set(scope.write.map(path => path === "." ? path : normalizeManifestScope(path).path))];
+    const focused = [];
     for (const entry of evidence) {
         const absolute = resolve(directory, entry.path);
         const local = relative(resolve(directory), absolute);
@@ -125,21 +160,15 @@ export async function missionReviewSource(directory, run, evidence = [], baselin
             stream.on("data", chunk => hash.update(chunk));
             const lines = createInterface({ input: stream, crlfDelay: Infinity });
             stream.on("error", error => lines.emit("error", error));
-            selected += `\n--- evidence: ${entry.path}:${entry.offset} ---\n`;
-            // Share the focused budget rather than silently starving later requested error branches.
-            const allowance = Math.floor(11_000 / evidence.length);
-            let number = 0, bytes = 0, truncated = false;
+            const excerpt = [];
+            let number = 0, bytes = 0;
             try {
                 for await (const line of lines) {
                     number++;
                     if (number >= entry.offset && number < entry.offset + entry.limit) {
-                        const text = `${number}: ${line.slice(0, 2_000)}\n`, size = Buffer.byteLength(text);
-                        if (!truncated && bytes + size <= allowance) {
-                            selected += text;
-                            bytes += size;
-                        }
-                        else
-                            truncated = true;
+                        const text = `${number}: ${line.slice(0, 2_000)}\n`;
+                        excerpt.push(text);
+                        bytes += Buffer.byteLength(text);
                     }
                 }
             }
@@ -149,14 +178,42 @@ export async function missionReviewSource(directory, run, evidence = [], baselin
             }
             if (entry.offset > number)
                 throw new Error(`offset ${entry.offset} exceeds ${number} lines; select existing lines`);
-            if (truncated)
-                selected += `[FOCUSED EXCERPT TRUNCATED: ${entry.path}:${entry.offset}; request a smaller range]\n`;
+            focused.push({ entry, lines: excerpt, bytes });
         }
         catch (error) {
             throw fail(error instanceof Error ? error.message : String(error));
         }
     }
-    const writes = [...new Set(scope.write.map(path => path === "." ? path : normalizeManifestScope(path).path))];
+    // Keep room for the baseline diff. Reserve notices only after the allocated whole lines show
+    // truncation; pre-reserving every possible notice can hide all six otherwise fitting references.
+    const heading = (entry) => `\n--- evidence: ${entry.path}:${entry.offset} ---\n`;
+    const truncated = (entry) => `[FOCUSED EXCERPT TRUNCATED: ${entry.path}:${entry.offset}; request a smaller range]\n`;
+    // Read-only reviews retain their smaller existing envelope; code reviews may use the space
+    // previously reserved for automatic diff prefixes, while leaving room for changed-file context.
+    const limit = writes.length ? 18_000 : 11_000;
+    const headings = focused.reduce((size, { entry }) => size + Buffer.byteLength(heading(entry)), 0);
+    const needsNotice = focused.map(() => false);
+    let allowances;
+    while (true) {
+        const notices = focused.reduce((size, { entry }, index) => size + (needsNotice[index] ? Buffer.byteLength(truncated(entry)) : 0), 0);
+        allowances = focusedAllowances(focused.map(item => item.bytes), Math.max(0, limit - headings - notices));
+        let changed = false;
+        for (const [index, item] of focused.entries()) {
+            if (!needsNotice[index] && visibleLineCount(item.lines, allowances[index]) < item.lines.length) {
+                needsNotice[index] = true;
+                changed = true;
+            }
+        }
+        if (!changed)
+            break;
+    }
+    let selected = "";
+    for (const [index, item] of focused.entries()) {
+        selected += heading(item.entry);
+        selected += item.lines.slice(0, visibleLineCount(item.lines, allowances[index])).join("");
+        if (needsNotice[index])
+            selected += truncated(item.entry);
+    }
     if (writes.length === 0)
         return { fingerprint: `sha256:${hash.digest("hex")}`,
             excerpt: selected + "[Read-only units: no declared output files. Review the supplied observations, traces and validation evidence.]" };

package/dist/plugin/profiled.js

@@ -1,4 +1,5 @@
 import { V010_RUNTIME_ASSET_VERSION } from "../asset-version.js";
+import { MISSION_BEHAVIOR_REVIEW } from "../runtime-mission-assets.js";
 import { cancelledMissionRetainsAcceptance, OperatorContractError, OperatorRuntime } from "../core/operator-runtime.js";
 import { DEFAULT_OPERATOR_PROPOSAL_BUDGET, OPERATOR_APPROVAL_CONTRACT, OPERATOR_PROPOSAL_BUDGET_CAPS, OPERATOR_PROPOSAL_REVISION_CONTRACT, OperatorProposalBudgetError, OperatorProposalRuntime } from "../core/operator-proposal.js";
 import { CANONICAL_AGENT_ROLES, canonicalAgent, profileAgent, profileTool, V010_RUNTIME_PROFILE } from "../core/runtime-profile.js";
@@ -561,7 +562,9 @@ export function createProfiledPlugin(profile, assetVersion) {
             // Only the same source can retain a cancelled run's acceptance as a prefix.
             if (previous.sourceRefs[0] !== `user:${mission.requests[0]?.id}` ||
                 !previous.sourceRefs.every(ref => mission.requests.some(request => ref === `user:${request.id}`))) {
-                throw new Error("mission-cancelled-source-unproven");
+                throw new Error("mission-cancelled-source-unproven: the cancelled run's source_refs do not prove this is the same request. " +
+                    "If the user changed or narrowed the goal, start_mission with intent=replace and the complete current requirements; " +
+                    "otherwise preserve the earlier accepted criteria and establish the original source before continuing.");
             }
             return previous.acceptance.every((text, index) => mission.requirements[index]?.text === text)
                 ? mission : missions.carryForward(root, mission.id, previous.acceptance);
@@ -1355,9 +1358,11 @@ export function createProfiledPlugin(profile, assetVersion) {
                         if (relocated)
                             return JSON.stringify({ ...relocated, budget: await control.currentBudget(context.sessionID) });
                     }
+                    const prior = await operators.read(context.sessionID);
                     let mission = await retainCancelledMissionAcceptance(context.sessionID, await missions.start(context.sessionID, args.requirements, args.intent === "replace" || args.intent === "new", {
                         kind: args.kind === "operation" ? "operation" : "implementation",
                         context: missionConversationContext(await messages(context.sessionID)),
+                        ...(args.intent === "replace" && prior?.phase === "cancelled" ? { cancelledRunID: prior.runID } : {}),
                     }));
                     if (!mission.reviewBaseline) {
                         const baseline = await missionReviewBaseline(input.directory);
@@ -1454,13 +1459,21 @@ export function createProfiledPlugin(profile, assetVersion) {
             args: { units: { type: "array", minItems: 1, maxItems: 32, items: { type: "object", additionalProperties: false,
                         properties: { title: { type: "string" }, objective: { type: "string" },
                             read: { ...stringList, description: "Inputs that affect validation, not an allowlist for observation. Do not include whole live session/database/log trees just to inspect them." }, write: stringList,
-                            validation: stringList, requirement_ids: stringList }, required: ["title", "objective", "write", "validation"] } },
+                            validation: stringList, requirement_ids: { ...stringList, description: "Related requirement IDs. A single unit inherits all requirements when omitted; specify coverage when splitting work across units." } }, required: ["title", "objective", "write", "validation"] } },
                 execution: { type: "object", properties: { commands: stringList, directory: { type: "string" } },
                     required: ["commands", "directory"], additionalProperties: false,
                     description: "For an operation: the actual run/grade commands, not a preflight or NO_START check. Host observes their native shell completion; no handwritten proof file is needed.", "x-sortie-optional": true },
                 reason: { type: "string", "x-sortie-optional": true } }, execute: async (args, context) => {
                 const { root, mission } = await missionAuthority(context.sessionID);
-                return serializeDispatchTransition(root, () => declareMissionUnits(root, context.sessionID, mission, args.units, args.reason, args.execution));
+                try {
+                    return await serializeDispatchTransition(root, () => declareMissionUnits(root, context.sessionID, mission, args.units, args.reason, args.execution));
+                }
+                catch (error) {
+                    if (!(error instanceof OperatorContractError))
+                        throw error;
+                    return JSON.stringify({ status: "invalid-plan", diagnostics: error.diagnostics, diagnostics_truncated: error.diagnostics_truncated,
+                        next_action: "Correct the reported field or control-storage problem and retry plan_units directly. Keep the original requirements and existing run; do not cancel or repeat passed work to repair the plan." });
+                }
             } };
         tools[expandUnit] = { description: "Coordinator: extend the stopped unit's write scope immediately within the original request. Keeps original requirements and cumulative budget; generates a replacement Worker contract. No Operator approval or handwritten manifest repair is needed. Use only after the Worker has returned.",
             args: { unit_id: stringSchema, paths: stringList, reason: stringSchema }, execute: async (args, context) => {
@@ -1517,9 +1530,10 @@ export function createProfiledPlugin(profile, assetVersion) {
                     description: `🔎 ${run.units[0].unit.title}`, prompt: [
                         `candidate_id: ${mission.id}`, `review_phase: ${phase}`, "canonical_validation_exit: 0", `risk_tags: [${risk.join(", ")}]`,
                         "Review this candidate independently. Use the language of the requirements/traces. Invoke no tools. First line: exactly PASS, FINDINGS or EVIDENCE_GAPS.",
-                        "Use EVIDENCE_GAPS only when no concrete source/test defect is established and a specific acceptance-relevant behavior or required validation cannot be settled by the supplied artifact. Name the affected path and why the missing evidence matters; do not request a generic route inventory or raw history for incidental workflow constraints already covered by a concise Worker trace. Any concrete defect uses FINDINGS.",
+                        "Use EVIDENCE_GAPS only when no concrete source/test defect is established and a specific acceptance-relevant behavior or required validation cannot be settled by the supplied artifact. Name the affected path and why the missing evidence matters; do not request a generic route inventory. Any concrete defect uses FINDINGS.",
                         "This Reviewer's native outcome and final acceptance can only be observed after this review. List those as deferred Operator checks, not as a reason to request another review. Still assess all available source, validation and historical evidence independently.",
                         "For changed failure paths, assess the public return value, error and post-failure state together against existing API behavior; matching error text alone does not establish compatibility.",
+                        MISSION_BEHAVIOR_REVIEW,
                         `acceptance: ${JSON.stringify(run.acceptance)}`, `changedLogicSummary: ${JSON.stringify(traces)}`,
                         ...run.acceptance.map((_, i) => `acceptance[${i}] -> changedLogicSummary[${i}]`),
                         `manifest: ${JSON.stringify(run.units.map(unit => unit.unit))}`, `sourceFingerprint: ${source.fingerprint}`,

package/dist/runtime-assets-v010.js

@@ -4,7 +4,7 @@ import { V010_RUNTIME_PROFILE as profile, profileAgent, renderProfileInstruction
 import { GOAL_DECLARATION_FORMAT } from "./core/goal-declaration-format.js";
 import { STRATEGY_TRIGGERS, SOURCE_REVIEW_PHASES, SOURCE_REVIEW_RISK_TAGS } from "./core/consultation.js";
 import { SCOUT_EVIDENCE_CODES } from "./core/scout-contract.js";
-import { missionOperatorContent, missionCoordinatorContent, missionWorkerContent } from "./runtime-mission-assets.js";
+import { MISSION_BEHAVIOR_REVIEW, missionOperatorContent, missionCoordinatorContent, missionWorkerContent } from "./runtime-mission-assets.js";
 const coordinator = profileAgent(profile, "dog-coordinator");
 const operator = profileAgent(profile, "dog-operator");
 const worker = profileAgent(profile, "dog-worker");
@@ -420,10 +420,9 @@ For a multi-target change, verify each affected target when the source shows ind
 
 Do not demand a generic creation/binding/mutation inventory, every possible value representation, or a
 cross-product of independent dimensions just because the construct already exists. Absence of that
-inventory alone is not an evidence gap. Do not demand raw logs or historical process traces for
-incidental workflow constraints when a concise Worker trace reports them and the supplied record does
-not contradict it. The Operator owns final comparison to the original request; still flag a missing
-required behavioral result or canonical validation evidence.
+inventory alone is not an evidence gap.
+
+${MISSION_BEHAVIOR_REVIEW}
 
 ## Mission review verdict
 

package/dist/runtime-assets.d.ts

@@ -1514,7 +1514,7 @@ END_INTERNAL_TERMINAL_PROOF_FIXTURE
     readonly name: "dog-reviewer";
     readonly version: "0.3.89-completion-proof-v1";
     readonly installPath: "agent/dog-reviewer.md";
-    readonly content: "---\ndescription: Independent source reviewer for dog-coordinator\nmode: subagent\npermission:\n  bash: deny\n  webfetch: deny\n  task: deny\n  question: deny\n  glob: deny\n  grep: deny\n  edit: deny\n  list: deny\n  write: deny\n  patch: deny\n  read: deny\ntools:\n  bash: false\n  webfetch: false\n  task: false\n  question: false\n  glob: false\n  grep: false\n  edit: false\n  list: false\n  write: false\n  patch: false\n  read: false\n---\n# dog-reviewer\n\nAccept only one bounded SourceReview request from dog-coordinator after canonical\nvalidation for one high-risk candidate. Review only the supplied acceptance criteria, exact\nmanifest, changedLogicSummary, supplied changed-code excerpts, and validation evidence. Confirm every acceptance item explicitly\nmaps to at least one changedLogicSummary entry and assess that changed logic against the mapped\nacceptance item. Missing or incomplete coverage is a concrete finding, never PASS.\nRequire one indexed acceptance[i] -> changedLogicSummary[j] mapping line per acceptance item and\nreject a missing index or unequal mapping count before assessing the changed logic.\nAlso require each acceptance item to map to a concrete exercising test/input/branch and result. A broad\nsuite PASS without criterion-level exercise evidence is insufficient. When one item contains materially\ndifferent syntax forms, value shapes, scopes, or error paths, reject PASS unless representative traces cover\neach path or the artifact proves they share one implementation path.\nCheck contract-derived expectations only for accepted behavior. For an applicable failure or composite-value\ncriterion, error-only or helper-only assertions can leave public result/state behavior unproved. Accept justified\nN/A dimensions; do not demand new behavior, new review rounds, or redundant checks outside accepted scope.\nWhen changed files include generator inputs or checked-in generated outputs, require the canonical generator\ncommand, a stable post-generation diff, and validation executed after generation. Reject PASS if helper logic\nexists only in a generated output, regeneration removes behavior, or validation predates the generated candidate.\nDo not request raw logs or full source files, review low-risk candidates, expand scope, or dispatch\nanother agent.\nTreat those supplied fields as the complete bounded SourceReview artifact; use only that artifact and invoke no tools.\nDo not infer that a branch or exemption is absent from source because a prose summary omits it.\nIf the supplied excerpts do not establish a claim, report an evidence gap and request the exact\nbranch/helper excerpt in the next artifact; do not prescribe a source fix for an unproven defect.\n\nReturn one concise PASS or concrete-finding response only to dog-coordinator before the\ncoordinator commit. Write every finding, evidence, and required-fix sentence in the language the\nsupplied artifact uses for its own prose, one statement per line, and keep verdict values,\nidentifiers, paths, and commands verbatim. Do not implement, remediate, resolve blockers, edit,\nstage, commit, or become user-facing. Remain host-routed: do not require or identify a provider, vendor, model, variant,\nor transport.\n";
+    readonly content: "---\ndescription: Independent source reviewer for dog-coordinator\nmode: subagent\npermission:\n  bash: deny\n  webfetch: deny\n  task: deny\n  question: deny\n  glob: deny\n  grep: deny\n  edit: deny\n  list: deny\n  write: deny\n  patch: deny\n  read: deny\ntools:\n  bash: false\n  webfetch: false\n  task: false\n  question: false\n  glob: false\n  grep: false\n  edit: false\n  list: false\n  write: false\n  patch: false\n  read: false\n---\n# dog-reviewer\n\nAccept only one bounded SourceReview request from dog-coordinator after canonical\nvalidation for one high-risk candidate. Review only the supplied acceptance criteria, exact\nmanifest, changedLogicSummary, supplied changed-code excerpts, and validation evidence. Confirm every acceptance item explicitly\nmaps to at least one changedLogicSummary entry and assess that changed logic against the mapped\nacceptance item. Missing behavioral evidence is an evidence gap; a demonstrated source/test defect is a finding.\nRequire one indexed acceptance[i] -> changedLogicSummary[j] mapping line per acceptance item and\nreject a missing index or unequal mapping count before assessing the changed logic.\nFor behavioral acceptance items, require a concrete exercising test/input/branch and result. A broad\nsuite PASS without criterion-level exercise evidence is insufficient. When one item contains materially\ndifferent syntax forms, value shapes, scopes, or error paths, reject PASS unless representative traces cover\neach path or the artifact proves they share one implementation path.\nCheck contract-derived expectations only for accepted behavior. For an applicable failure or composite-value\ncriterion, error-only or helper-only assertions can leave public result/state behavior unproved. Accept justified\nN/A dimensions; do not demand new behavior, new review rounds, or redundant checks outside accepted scope.\nWhen changed files include generator inputs or checked-in generated outputs, require the canonical generator\ncommand, a stable post-generation diff, and validation executed after generation. Reject PASS if helper logic\nexists only in a generated output, regeneration removes behavior, or validation predates the generated candidate.\nDo not request raw logs or full source files, review low-risk candidates, expand scope, or dispatch\nanother agent.\nTreat those supplied fields as the complete bounded SourceReview artifact; use only that artifact and invoke no tools.\nDo not infer that a branch or exemption is absent from source because a prose summary omits it.\nIf the supplied excerpts do not establish a claim, report an evidence gap and request the exact\nbranch/helper excerpt in the next artifact; do not prescribe a source fix for an unproven defect.\n\nReturn one concise PASS or concrete-finding response only to dog-coordinator before the\ncoordinator commit. Write every finding, evidence, and required-fix sentence in the language the\nsupplied artifact uses for its own prose, one statement per line, and keep verdict values,\nidentifiers, paths, and commands verbatim. Do not implement, remediate, resolve blockers, edit,\nstage, commit, or become user-facing. Remain host-routed: do not require or identify a provider, vendor, model, variant,\nor transport.\n";
 }, {
     readonly name: "dog-advisor";
     readonly version: "0.3.89-completion-proof-v1";

package/dist/runtime-assets.js

@@ -1824,10 +1824,10 @@ Accept only one bounded SourceReview request from dog-coordinator after canonica
 validation for one high-risk candidate. Review only the supplied acceptance criteria, exact
 manifest, changedLogicSummary, supplied changed-code excerpts, and validation evidence. Confirm every acceptance item explicitly
 maps to at least one changedLogicSummary entry and assess that changed logic against the mapped
-acceptance item. Missing or incomplete coverage is a concrete finding, never PASS.
+acceptance item. Missing behavioral evidence is an evidence gap; a demonstrated source/test defect is a finding.
 Require one indexed acceptance[i] -> changedLogicSummary[j] mapping line per acceptance item and
 reject a missing index or unequal mapping count before assessing the changed logic.
-Also require each acceptance item to map to a concrete exercising test/input/branch and result. A broad
+For behavioral acceptance items, require a concrete exercising test/input/branch and result. A broad
 suite PASS without criterion-level exercise evidence is insufficient. When one item contains materially
 different syntax forms, value shapes, scopes, or error paths, reject PASS unless representative traces cover
 each path or the artifact proves they share one implementation path.

package/dist/runtime-mission-assets.d.ts

@@ -2,5 +2,7 @@ import { type RuntimeProfile } from "./core/runtime-profile.ts";
 /** Repository-local dependency environment shared by all units; excluded from reviewed and captured source. */
 export declare const TOOL_ENVIRONMENT = ".sortie-env";
 export declare function missionOperatorContent(profile: RuntimeProfile, version: string): string;
+/** Shared by the installed Reviewer and its host-generated mission prompt. */
+export declare const MISSION_BEHAVIOR_REVIEW = "For a changed failure handler, inspect the operation it calls and the public inputs reaching it,\nincluding failures not listed in the new handler. Use the supplied source/tests and established API behavior\nto identify a concrete input that could still violate the requested contract. A passing normal input does\nnot settle a different failure outcome of that same operation. A demonstrable defect is FINDINGS; ask for\nan excerpt or result only when a specific material outcome cannot be settled. Do not invent new behavior,\nrequire an exhaustive exception inventory, or recommend catching every exception.\n\nBehavioral requirements need concrete input/result evidence. For incidental workflow constraints such as\ncache settings or command-path spelling, use the existing host observations and concise compliance trace;\nabsence of a separate settings dump or historical log is not itself an evidence gap. Flag observed\ncontradictions. The Operator owns final comparison with the original request. If a missing check genuinely\naffects correctness or a requested deliverable, name that consequence and the smallest useful next check.";
 export declare function missionCoordinatorContent(profile: RuntimeProfile, version: string): string;
 export declare function missionWorkerContent(profile: RuntimeProfile): string;

package/dist/runtime-mission-assets.js

@@ -81,7 +81,8 @@ results to guide the following units. Do not invent preparation units or plan-ap
 For operations, plan_units.execution names the actual run/grade commands and working directory. Keep
 setup, execution and result collection in the same Worker. The host records native execution; NO_START
 or setup success cannot complete the operation. Reward/score zero is a result, not failure to execute.
-Use requirement_ids for each unit when there are multiple requirements; these are related requirements,
+Use requirement_ids when splitting multiple requirements across units; a single unit inherits all requirements
+when they are omitted. These are related requirements,
 not claims that a command proves every semantic obligation. Compare the final result yourself.
 
 Copy returned task fields exactly (V2: subagent_type -> agent, task_id -> sessionID). Do not append to a
@@ -101,6 +102,19 @@ once. Do not invent scores, medals, costs, savings, models or successful checks.
 existing user authorization and project gates; npm publication remains manual.
 `;
 }
+/** Shared by the installed Reviewer and its host-generated mission prompt. */
+export const MISSION_BEHAVIOR_REVIEW = `For a changed failure handler, inspect the operation it calls and the public inputs reaching it,
+including failures not listed in the new handler. Use the supplied source/tests and established API behavior
+to identify a concrete input that could still violate the requested contract. A passing normal input does
+not settle a different failure outcome of that same operation. A demonstrable defect is FINDINGS; ask for
+an excerpt or result only when a specific material outcome cannot be settled. Do not invent new behavior,
+require an exhaustive exception inventory, or recommend catching every exception.
+
+Behavioral requirements need concrete input/result evidence. For incidental workflow constraints such as
+cache settings or command-path spelling, use the existing host observations and concise compliance trace;
+absence of a separate settings dump or historical log is not itself an evidence gap. Flag observed
+contradictions. The Operator owns final comparison with the original request. If a missing check genuinely
+affects correctness or a requested deliverable, name that consequence and the smallest useful next check.`;
 export function missionCoordinatorContent(profile, version) {
     return `---
 description: Sortie-dogs ${version} Coordinator — investigation, unit dispatch, scope extension and correction loop.
@@ -144,7 +158,8 @@ select the matching runner/profile before starting rather than changing the pinn
 Investigate only enough to start the first useful Worker. Prefer a targeted read/reproduction over a broad
 inventory or speculative full design. Call ${profile.toolPrefix}plan_units with concise units:
 title, objective, read/write file or directory scopes, validation commands, and related requirement_ids
-when there is more than one requirement. For operation missions, include execution with the actual
+when splitting multiple requirements across multiple units; a single unit inherits all requirements when
+requirement_ids is omitted. For operation missions, include execution with the actual
 run/grade commands and working directory. Setup, launch and result collection normally stay in one Worker;
 do not forbid execution while assigning that Worker the requirement to execute.
 When the issue includes a concrete public reproduction, pass its entrypoint, relevant input and observed
@@ -153,7 +168,9 @@ example depends on a working directory or package layout, preserve that context.
 existing analogous source/tests for the expected contract when available. Avoid separate investigation
 units just to restate the issue. A test of a neighboring name is not an adjacent check unless it runs
 the changed branch on a relevant different input; keep validation focused and do not require an extra
-test when the existing checks already exercise that boundary.
+test when the existing checks already exercise that boundary. For an exception fix, identify the failing
+operation and ask the Worker to consider its other source/API-backed failure inputs, including ones the
+new handler does not catch. A normal input alone does not check a different failure outcome.
 For read-only verification, use write: []; do not invent an output file or request write access to inputs.
 For ordinary diagnostics, use native read/search/shell directly, including while an old run is being
 reconciled. Do not create a dummy validation/console.log unit just to inspect status. The read list is
@@ -164,7 +181,7 @@ Use absolute native paths for requested global installations or other external o
 a directory including a not-yet-created tree. These are execution/evidence scopes, not an additional
 permission grant: the host's native permissions still apply. Include the actual external input/output
 paths in read/write so validation and review observe them; do not substitute a repository symlink.
-Keep all original requirements covered; only a single requirement can be inferred. Last validation command checks
+Keep all original requirements covered. Last validation command checks
 that unit. If your quick check shows repository-declared dependencies or the test runner are missing, keep
 setup inside the first unit: declare its checks through the repository-local tool environment ${TOOL_ENVIRONMENT}/
 (for Python, ${TOOL_ENVIRONMENT}/bin/python -m pytest ...) and let that Worker create it. Never plan a separate setup
@@ -198,7 +215,8 @@ and ask a bounded question in the user's language; do not send generic explorato
 If the user explicitly requested Advisor input before a decision, do not treat it as optional.
 
 After formal validation, call review_mission with risk_tags and one concise implementation/test trace per
-requirement. Include the observed result for required behavioral checks, not a speculative route inventory
+requirement. For changed failure behavior, connect the operation and concrete input to the contract-derived
+expected result and observed result in those existing traces. Include required behavioral checks, not a speculative route inventory
 or raw history to prove incidental process constraints. Recognized tags: ${SOURCE_REVIEW_RISK_TAGS.join(", ")}.
 High-risk changes require the generated independent ${profileAgent(profile, "dog-reviewer")} task.
 Low risk uses [] and the host records the skip. The host supplies source excerpts, manifest, requirement
@@ -253,11 +271,13 @@ formal acceptance evidence. Do not repeat a failed command without a concrete so
 repeat passed checks on unchanged source. Add meaningful tests only when needed by the change/request.
 For a reported bug, keep the public reproduction's entrypoint, input and layout intact during diagnosis;
 after a fix, run it again when the available environment permits. If it cannot run, report exactly
-what remains unverified instead of substituting a different passing check. When a changed condition
-or exception handler also covers an adjacent input demonstrated by the public source/tests, check that
-input on the same branch (e.g. a neighboring exception type), rather than checking only a nearby test
-name. Derive expected behavior from public code and tests, not hidden evaluator details; skip redundant
-checks already covered by formal validation. Return a concrete failed reproduction to Coordinator for
+what remains unverified instead of substituting a different passing check. For a changed condition or
+exception handler, inspect the underlying operation and inputs reaching it, including failures the new
+handler does not catch. Check a materially different failure input when public source/tests or established
+API behavior support the same requested contract; a normal input alone does not check that failure outcome.
+Choose the smallest complete fix, not a catch-all or an exhaustive exception matrix. Derive expected behavior
+from public code and tests, not hidden evaluator details; skip redundant checks already covered by formal validation.
+Return a concrete failed reproduction to Coordinator for
 the same-goal correction loop rather than declaring the whole task complete.
 When changing a failure path, check its public return value, error and post-failure state together against
 the existing API contract; do not stop assertions after matching error text.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "sortie-dogs",
-  "version": "0.12.18",
+  "version": "0.12.19",
   "description": "Bounded agent harness and validated orchestration loop plugin for OpenCode",
   "keywords": [
     "opencode",