vibe-coding-master 0.7.33 → 0.7.35

package/dist/backend/services/harness-feedback-service.js

@@ -250,13 +250,14 @@ export function createHarnessFeedbackService(deps) {
                     "Auto Memory Review:",
                     `Role drafts: ${memoryReview.roleDraftsPath}`,
                     `Current memory snapshot: ${memoryReview.currentMemoryPath}`,
+                    "Active memory files:",
+                    ...memoryReview.activeMemoryPaths.map((memoryPath) => `- ${memoryPath}`),
                     ...(memoryReview.planningCandidatePath
                         ? [`Architect planning-session candidate: ${memoryReview.planningCandidatePath}`]
                         : []),
-                    `Write the complete reviewed memory set to: ${memoryReview.reviewedMemoryPath}`,
                     "",
                     "Review every memory candidate against final task evidence while performing this retrospective.",
-                    "Each snapshot file contains only the matching <VCM-memory> block content. Edit every existing reviewed-memory file in place.",
+                    "The snapshot files contain only the matching pre-review <VCM-memory> block content.",
                     "Before evaluating proposals, review every substantive entry in every current memory snapshot against current code, documentation, and final task evidence.",
                     "For each existing entry, decide retain, update, remove, or move-to-durable-doc. Record the decision reason, the impact of removing it, and whether memory or a durable document is the correct source.",
                     "Complete this full existing-memory review even when every proposal says no-change.",
@@ -265,12 +266,14 @@ export function createHarnessFeedbackService(deps) {
                     "Do not copy the proposer rationale as the review. Verify it against current code, durable documentation, and final task evidence.",
                     "Keep only verified, durable, reusable project knowledge. Merge duplicates and keep role-specific knowledge in the matching role file.",
                     "Do not record task narrative, temporary state, unverified conclusions, or Harness rules in memory.",
-                    "Final content must be one exact line written to the selected reviewed-memory file. Use none when the decision does not keep memory.",
+                    "Final content must be one exact line written to the selected active <VCM-memory> block. Use none when the decision does not keep memory.",
                     "For keep-in-memory use Durable doc disposition: memory. For keep-memory-reference use memory-reference. For move-to-durable-doc use durable-doc.",
                     "Existing-memory Target must be shared or the exact role name. Decision must be retain, update, remove, or move-to-durable-doc.",
                     "Durable doc disposition must be memory, durable-doc, or memory-reference. Use Durable doc path: none with memory and an actual path with the other dispositions.",
                     "Do not keep full content in memory when durable-doc is correct. Use memory-reference only when an ongoing role needs the document pointer.",
                     "Use none as the complete Existing Memory Decisions body only when no substantive existing memory entry exists.",
+                    "Apply the reviewed result directly to the <VCM-memory> blocks in the listed active memory files. Do not change any content outside those blocks.",
+                    "If memory changes, commit only the changed active memory files before ending the turn. Use commit message: chore: update VCM memory. If memory is unchanged, do not create a commit.",
                     "Use this exact block in the retrospective report and replace each option or placeholder with one allowed value or a concise summary:",
                     "",
                     "## Memory Review",

package/dist/backend/services/session-service.js

@@ -31,11 +31,11 @@ const SESSION_READY_MAX_POLLS = 60;
 export function createSessionService(deps) {
     const now = deps.now ?? (() => new Date().toISOString());
     const isProcessAlive = deps.isProcessAlive ?? defaultIsProcessAlive;
-    async function readCurrentHarnessRevision(repoRoot) {
-        return (await readHarnessRevisionState(deps.fs, repoRoot)).revision;
+    async function readCurrentHarnessRevision(harnessRepoRoot) {
+        return (await readHarnessRevisionState(deps.fs, harnessRepoRoot)).revision;
     }
-    async function withHarnessRevisionView(repoRoot, record) {
-        const currentRevision = await readCurrentHarnessRevision(repoRoot);
+    async function withHarnessRevisionView(harnessRepoRoot, record) {
+        const currentRevision = await readCurrentHarnessRevision(harnessRepoRoot);
         const sessionRevision = normalizeHarnessRevision(record.harnessRevision);
         return {
             ...record,
@@ -45,13 +45,13 @@ export function createSessionService(deps) {
         };
     }
     async function launchRoleSession(repoRoot, taskSlug, role, input, launchMode) {
-        const live = toRoleSessionRecordView(getRegisteredRoleSession(deps.registry, deps.runtime, taskSlug, role), deps.runtime);
-        if (live && live.status === "running") {
-            return withHarnessRevisionView(repoRoot, live);
-        }
         const config = await deps.projectService.loadConfig(repoRoot);
         const task = await deps.taskService.loadTask(repoRoot, taskSlug);
         const taskRepoRoot = getTaskRuntimeRepoRoot(task);
+        const live = toRoleSessionRecordView(getRegisteredRoleSession(deps.registry, deps.runtime, taskSlug, role), deps.runtime);
+        if (live && live.status === "running") {
+            return withHarnessRevisionView(taskRepoRoot, live);
+        }
         const paths = deps.artifactService.getHandoffPaths(taskRepoRoot, task.handoffDir);
         const persisted = await loadPersistedRoleRecordForRole(deps.fs, repoRoot, taskRepoRoot, config.stateRoot, taskSlug, role);
         const permissionMode = normalizeClaudePermissionMode(input.permissionMode ?? persisted?.permissionMode);
@@ -106,7 +106,7 @@ export function createSessionService(deps) {
             rows: input.rows
         });
         const timestamp = now();
-        const harnessRevision = await readCurrentHarnessRevision(repoRoot);
+        const harnessRevision = await readCurrentHarnessRevision(taskRepoRoot);
         const record = {
             id: runtimeSession.id,
             runtimeSessionToken,
@@ -139,7 +139,7 @@ export function createSessionService(deps) {
         if (role === "project-manager") {
             await restoreProjectManagerWorkflowContext(record, taskRepoRoot, config.stateRoot);
         }
-        return withHarnessRevisionView(repoRoot, record);
+        return withHarnessRevisionView(taskRepoRoot, record);
     }
     async function restoreProjectManagerWorkflowContext(record, taskRepoRoot, stateRoot) {
         if (!deps.taskWorkflowService || record.status !== "running") {
@@ -165,7 +165,7 @@ export function createSessionService(deps) {
         const taskContext = await resolveProjectToolTaskContext(repoRoot, input, "Translator");
         const live = toRoleSessionRecordView(getRegisteredProjectTranslatorSession(deps.registry, deps.runtime), deps.runtime);
         if (live && live.status === "running") {
-            return withHarnessRevisionView(repoRoot, await migrateRunningProjectToolSessionCwd(repoRoot, live, taskContext.taskRepoRoot));
+            return withHarnessRevisionView(taskContext.taskRepoRoot, await migrateRunningProjectToolSessionCwd(repoRoot, live, taskContext.taskRepoRoot));
         }
         const config = await deps.projectService.loadConfig(repoRoot);
         const persisted = await loadPersistedTranslatorSession(deps.fs, repoRoot);
@@ -236,7 +236,7 @@ export function createSessionService(deps) {
             rows: input.rows
         });
         const timestamp = now();
-        const harnessRevision = await readCurrentHarnessRevision(repoRoot);
+        const harnessRevision = await readCurrentHarnessRevision(taskContext.taskRepoRoot);
         const record = {
             id: runtimeSession.id,
             runtimeSessionToken,
@@ -277,15 +277,15 @@ export function createSessionService(deps) {
                 await clearPersistedTranslatorSession(deps.fs, repoRoot);
                 return launchProjectTranslatorSession(repoRoot, input, "fresh");
             }
-            return withHarnessRevisionView(repoRoot, migrated);
+            return withHarnessRevisionView(taskContext.taskRepoRoot, migrated);
         }
-        return withHarnessRevisionView(repoRoot, await migrateRunningProjectToolSessionCwd(repoRoot, record, taskContext.taskRepoRoot));
+        return withHarnessRevisionView(taskContext.taskRepoRoot, await migrateRunningProjectToolSessionCwd(repoRoot, record, taskContext.taskRepoRoot));
     }
     async function launchProjectHarnessEngineerSession(repoRoot, input, launchMode) {
         const taskContext = await resolveProjectToolTaskContext(repoRoot, input, "Harness Engineer");
         const live = toRoleSessionRecordView(getRegisteredProjectHarnessEngineerSession(deps.registry, deps.runtime), deps.runtime);
         if (live && live.status === "running") {
-            return withHarnessRevisionView(repoRoot, await migrateRunningProjectToolSessionCwd(repoRoot, live, taskContext.taskRepoRoot));
+            return withHarnessRevisionView(taskContext.taskRepoRoot, await migrateRunningProjectToolSessionCwd(repoRoot, live, taskContext.taskRepoRoot));
         }
         const config = await deps.projectService.loadConfig(repoRoot);
         const persisted = await loadPersistedHarnessEngineerSession(deps.fs, repoRoot);
@@ -352,7 +352,7 @@ export function createSessionService(deps) {
             rows: input.rows
         });
         const timestamp = now();
-        const harnessRevision = await readCurrentHarnessRevision(repoRoot);
+        const harnessRevision = await readCurrentHarnessRevision(taskContext.taskRepoRoot);
         const record = {
             id: runtimeSession.id,
             runtimeSessionToken,
@@ -392,9 +392,9 @@ export function createSessionService(deps) {
                 await clearPersistedHarnessEngineerSession(deps.fs, repoRoot);
                 return launchProjectHarnessEngineerSession(repoRoot, input, "fresh");
             }
-            return withHarnessRevisionView(repoRoot, migrated);
+            return withHarnessRevisionView(taskContext.taskRepoRoot, migrated);
         }
-        return withHarnessRevisionView(repoRoot, await migrateRunningProjectToolSessionCwd(repoRoot, record, taskContext.taskRepoRoot));
+        return withHarnessRevisionView(taskContext.taskRepoRoot, await migrateRunningProjectToolSessionCwd(repoRoot, record, taskContext.taskRepoRoot));
     }
     async function resolveProjectToolTaskContext(repoRoot, input, roleLabel) {
         const taskSlug = input.taskSlug?.trim();
@@ -620,7 +620,7 @@ export function createSessionService(deps) {
             await persistHarnessEngineerSession(deps.fs, repoRoot, session);
         }
     }
-    async function notifyHarnessUpdatedForSession(repoRoot, session) {
+    async function buildHarnessUpdatedSession(harnessRepoRoot, session) {
         const runtimeSession = deps.runtime.getSession(session.id);
         if (!runtimeSession || runtimeSession.status !== "running") {
             throw new VcmError({
@@ -630,10 +630,10 @@ export function createSessionService(deps) {
                 hint: "Resume or restart the role to load the latest harness settings."
             });
         }
-        const currentRevision = await readCurrentHarnessRevision(repoRoot);
+        const currentRevision = await readCurrentHarnessRevision(harnessRepoRoot);
         const timestamp = now();
         await submitTerminalInput(deps.runtime, session.id, buildHarnessRefreshPrompt(session.role));
-        const updated = {
+        return {
             ...session,
             harnessRevision: currentRevision,
             harnessCurrentRevision: currentRevision,
@@ -641,28 +641,6 @@ export function createSessionService(deps) {
             lastHarnessNotifyAt: timestamp,
             updatedAt: timestamp
         };
-        deps.registry.upsert(normalizeProjectScopedRecordForPersistence(updated));
-        await persistNotifiedHarnessSession(repoRoot, updated);
-        return withHarnessRevisionView(repoRoot, updated);
-    }
-    async function persistNotifiedHarnessSession(repoRoot, session) {
-        if (session.role === TRANSLATOR_ROLE) {
-            await persistTranslatorSession(deps.fs, repoRoot, {
-                ...session,
-                taskSlug: PROJECT_TRANSLATOR_SCOPE
-            });
-            return;
-        }
-        if (session.role === HARNESS_ENGINEER_ROLE) {
-            await persistHarnessEngineerSession(deps.fs, repoRoot, {
-                ...session,
-                taskSlug: PROJECT_HARNESS_ENGINEER_SCOPE
-            });
-            return;
-        }
-        const config = await deps.projectService.loadConfig(repoRoot);
-        const task = await deps.taskService.loadTask(repoRoot, session.taskSlug);
-        await persistRoleSessionRecord(deps.fs, repoRoot, getTaskRuntimeRepoRoot(task), config.stateRoot, session);
     }
     async function getProjectToolSessionView(repoRoot, role) {
         const record = role === TRANSLATOR_ROLE
@@ -671,7 +649,7 @@ export function createSessionService(deps) {
             : getRegisteredProjectHarnessEngineerSession(deps.registry, deps.runtime)
                 ?? await loadPersistedHarnessEngineerSession(deps.fs, repoRoot);
         const view = toRoleSessionRecordView(record, deps.runtime);
-        return view ? withHarnessRevisionView(repoRoot, view) : undefined;
+        return view ? withHarnessRevisionView(view.cwd, view) : undefined;
     }
     async function markProjectToolActivityIdle(repoRoot, current, persist) {
         const timestamp = now();
@@ -692,7 +670,7 @@ export function createSessionService(deps) {
         const record = getRegisteredRoleSession(deps.registry, deps.runtime, taskSlug, role)
             ?? await loadPersistedRoleRecordForRole(deps.fs, repoRoot, taskRepoRoot, config.stateRoot, taskSlug, role);
         const view = toRoleSessionRecordView(record, deps.runtime);
-        return view ? withHarnessRevisionView(repoRoot, view) : undefined;
+        return view ? withHarnessRevisionView(taskRepoRoot, view) : undefined;
     }
     async function markTaskRoleActivityIdle(repoRoot, taskSlug, role, expectedSessionId) {
         const current = await getTaskRoleSessionView(repoRoot, taskSlug, role);
@@ -816,7 +794,7 @@ export function createSessionService(deps) {
             const record = getRegisteredProjectTranslatorSession(deps.registry, deps.runtime)
                 ?? await loadPersistedTranslatorSession(deps.fs, repoRoot);
             const view = toRoleSessionRecordView(record, deps.runtime);
-            return view ? withHarnessRevisionView(repoRoot, view) : undefined;
+            return view ? withHarnessRevisionView(view.cwd, view) : undefined;
         },
         async ensureProjectTranslatorSession(repoRoot, input = {}) {
             const existing = await this.getProjectTranslatorSession(repoRoot);
@@ -878,7 +856,10 @@ export function createSessionService(deps) {
                     statusCode: 404
                 });
             }
-            return notifyHarnessUpdatedForSession(repoRoot, current);
+            const updated = await buildHarnessUpdatedSession(current.cwd, current);
+            deps.registry.upsert(normalizeProjectScopedRecordForPersistence(updated));
+            await persistTranslatorSession(deps.fs, repoRoot, updated);
+            return withHarnessRevisionView(current.cwd, updated);
         },
         startProjectHarnessEngineerSession(repoRoot, input = {}) {
             return launchProjectHarnessEngineerSession(repoRoot, input, "fresh");
@@ -942,7 +923,7 @@ export function createSessionService(deps) {
             const record = getRegisteredProjectHarnessEngineerSession(deps.registry, deps.runtime)
                 ?? await loadPersistedHarnessEngineerSession(deps.fs, repoRoot);
             const view = toRoleSessionRecordView(record, deps.runtime);
-            return view ? withHarnessRevisionView(repoRoot, view) : undefined;
+            return view ? withHarnessRevisionView(view.cwd, view) : undefined;
         },
         async ensureProjectHarnessEngineerSession(repoRoot, input = {}) {
             const existing = await this.getProjectHarnessEngineerSession(repoRoot);
@@ -1004,7 +985,10 @@ export function createSessionService(deps) {
                     statusCode: 404
                 });
             }
-            return notifyHarnessUpdatedForSession(repoRoot, current);
+            const updated = await buildHarnessUpdatedSession(current.cwd, current);
+            deps.registry.upsert(normalizeProjectScopedRecordForPersistence(updated));
+            await persistHarnessEngineerSession(deps.fs, repoRoot, updated);
+            return withHarnessRevisionView(current.cwd, updated);
         },
         startRoleSession(repoRoot, taskSlug, role, input = {}) {
             return launchRoleSession(repoRoot, taskSlug, role, input, "fresh");
@@ -1061,7 +1045,7 @@ export function createSessionService(deps) {
                 return undefined;
             }
             const view = toRoleSessionRecordView(record, deps.runtime);
-            return view ? withHarnessRevisionView(repoRoot, view) : undefined;
+            return view ? withHarnessRevisionView(taskRepoRoot, view) : undefined;
         },
         async listRoleSessions(repoRoot, taskSlug) {
             const sessions = [];
@@ -1077,7 +1061,7 @@ export function createSessionService(deps) {
                     sessions.push(session);
                 }
             }
-            return Promise.all(sessions.map((session) => withHarnessRevisionView(repoRoot, session)));
+            return Promise.all(sessions.map((session) => withHarnessRevisionView(taskRepoRoot, session)));
         },
         async notifyRoleHarnessUpdated(repoRoot, taskSlug, role) {
             const current = await this.getRoleSession(repoRoot, taskSlug, role);
@@ -1088,7 +1072,13 @@ export function createSessionService(deps) {
                     statusCode: 404
                 });
             }
-            return notifyHarnessUpdatedForSession(repoRoot, current);
+            const config = await deps.projectService.loadConfig(repoRoot);
+            const task = await deps.taskService.loadTask(repoRoot, taskSlug);
+            const taskRepoRoot = getTaskRuntimeRepoRoot(task);
+            const updated = await buildHarnessUpdatedSession(taskRepoRoot, current);
+            deps.registry.upsert(updated);
+            await persistRoleSessionRecord(deps.fs, repoRoot, taskRepoRoot, config.stateRoot, updated);
+            return withHarnessRevisionView(taskRepoRoot, updated);
         },
         async recordRoleHookEvent(repoRoot, input) {
             const current = await this.getRoleSession(repoRoot, input.taskSlug, input.role);

package/dist/backend/templates/handoff.js

@@ -1,7 +1,10 @@
 import { ARCHITECTURE_BRIEF_STATUSES, ARCHITECTURE_PLAN_RESULTS, DOCS_SYNC_DECISIONS, FINAL_ACCEPTANCE_DECISIONS, L3_ACTIONS, L3_REQUIRED_VALUES, STRICT_NONE_VALUE, TEST_INFRASTRUCTURE_STATUSES, TEST_RESULTS, renderArtifactOptions } from "../../shared/validation/artifact-contract.js";
+const CURRENT_HANDOFF_NOTICE = "<!-- VCM current handoff: replace this file with one complete, self-contained snapshot of the current result. Restate all still-relevant evidence; do not refer to a prior revision, route message, Session, or transcript as evidence. -->";
 export function renderArchitectureBriefTemplate(taskSlug) {
     return `# Architecture Brief: ${taskSlug}
 
+${CURRENT_HANDOFF_NOTICE}
+
 Architecture Brief Status: ${renderArtifactOptions(ARCHITECTURE_BRIEF_STATUSES)}
 
 ## Accepted Outcome
@@ -28,6 +31,8 @@ TBD
 export function renderArchitecturePlanTemplate(taskSlug) {
     return `# Architecture Plan: ${taskSlug}
 
+${CURRENT_HANDOFF_NOTICE}
+
 Planning Result: ${renderArtifactOptions(ARCHITECTURE_PLAN_RESULTS)}
 
 ## Accepted Scope
@@ -135,6 +140,8 @@ TBD
 export function renderKnownIssuesTemplate(taskSlug) {
     return `# Known Issues: ${taskSlug}
 
+${CURRENT_HANDOFF_NOTICE}
+
 ## Task Issues
 
 No unresolved task issues recorded yet.
@@ -147,6 +154,8 @@ At task close, promote still-relevant confirmed issues to \`docs/known-issues.md
 export function renderTestReportTemplate(taskSlug) {
     return `# Test Report: ${taskSlug}
 
+${CURRENT_HANDOFF_NOTICE}
+
 Test Result: ${renderArtifactOptions(TEST_RESULTS)}
 
 ## Evidence Reviewed
@@ -249,6 +258,8 @@ ${STRICT_NONE_VALUE}
 export function renderCoderCompletionTemplate(taskSlug) {
     return `# Coder Completion: ${taskSlug}
 
+${CURRENT_HANDOFF_NOTICE}
+
 Decision: ready_for_review|incomplete|failed
 
 ## Scaffold Completion
@@ -293,6 +304,8 @@ TBD
 export function renderArchitectDebugTemplate(taskSlug) {
     return `# Architect Debug: ${taskSlug}
 
+${CURRENT_HANDOFF_NOTICE}
+
 Status: pending|completed
 
 ## PM-Routed Failure
@@ -342,6 +355,8 @@ TBD
 export function renderDocsSyncReportTemplate(taskSlug) {
     return `# Docs Sync Report: ${taskSlug}
 
+${CURRENT_HANDOFF_NOTICE}
+
 ## Summary
 
 TBD
@@ -378,6 +393,8 @@ ${renderArtifactOptions(DOCS_SYNC_DECISIONS)}
 export function renderFinalAcceptanceTemplate(taskSlug) {
     return `# Final Acceptance: ${taskSlug}
 
+${CURRENT_HANDOFF_NOTICE}
+
 ## Decision
 
 ${renderArtifactOptions(FINAL_ACCEPTANCE_DECISIONS)}

package/dist/backend/templates/harness/architect-agent.js

@@ -74,7 +74,7 @@ ${renderRoleMemoryRules("architect")}
 
 - \`architecture-plan.md\` must start with \`Planning Result: complete|incomplete|user clarification required\` and use these sections: Accepted Scope, Current Code Reality, Existing Assumptions And Class Coverage, Architecture Decision, Module/File Plan, Public Surface Impact, Scaffold Manifest, Scaffold Build Evidence, Tester Coverage Hints, Docs Impact, Known Risks, and Coder Handoff Notes.
 - Use \`Planning Result: complete\` only when: the plan document is complete; the Scaffold Manifest ledger reconciles one to one against the committed markers; and \`Scaffold Build Evidence\` records a green compile/typecheck run at the current scaffold commit hash. Include the same Planning Result in the route message to project-manager; do not select the next route.
-- \`architecture-plan.md\` is the current executable plan, not a changelog. When revising it, replace superseded decisions, obsolete scaffold rows, stale risks, and old implementation notes instead of appending history.
+- \`architecture-plan.md\` is the complete, self-contained current executable plan, not a changelog. Each revision must restate every still-current decision, constraint, evidence reference, scaffold row, risk, and implementation instruction needed to execute and review the plan without a prior revision. Replace superseded decisions, obsolete scaffold rows, stale risks, and old implementation notes instead of appending history.
 - \`Accepted Scope\`: state the PM-routed task scope and the confirmed brief's required user-visible outcome and decisions, plus any explicit non-scope that prevents accidental expansion.
 - \`Current Code Reality\`: cite \`architecture-evidence.md\` and summarize only the verified facts that constrain the architecture decision. Do not duplicate the full evidence inventory. For any module whose build configuration the plan changes, the evidence artifact must quote its complete direct dependency list from the package manifest, never a summary or selection.
 - Any enumeration the plan presents as complete over the codebase — call-site inventories, module or file lists, symbol sets — must either record the deterministic, repository-local command that generates it (run at the scaffold commit, the set transcribed from its output) or be explicitly marked as judgment-derived with the evidence basis for its completeness. A complete-claimed enumeration with neither is not evidence.
@@ -149,7 +149,7 @@ ${renderRoleMemoryRules("architect")}
 - Before handing off an architect-completed Debug Mode fix, run the smallest relevant L0 fast checks for the touched files or changed modules: format, lint, typecheck, boundary, dependency, or project-defined equivalents. If a check cannot run, report the exact reason.
 - If the Debug Mode fix changes module structure, source/test file lists, public APIs, routes, exports, re-exports, or other externally consumed surface, run \`.ai/tools/generate-module-index\` / \`.ai/tools/generate-public-surface\` or their \`--check\` mode as applicable.
 - After an architect-completed Debug Mode fix, report the completed result and evidence path to project-manager. Do not select the next route.
-- Before reporting a completed Debug Mode code fix, replace \`.ai/vcm/handoffs/architect-debug.md\` with current evidence. Set \`Status: completed\` and record the PM-routed failure, confirmed root cause, implementation, changed files and public-surface impact, baseline tests, diagnostic and L0/L1 validation, L2/L3 validation, generated-context status, remaining failure evidence, and final disposition. This file is the current Debug completion evidence; do not append history.
+- Before reporting a completed Debug Mode code fix, replace \`.ai/vcm/handoffs/architect-debug.md\` with complete, self-contained current evidence. Set \`Status: completed\` and restate the PM-routed failure, confirmed root cause, implementation, changed files and public-surface impact, baseline tests, diagnostic and L0/L1 validation, L2/L3 validation, generated-context status, remaining failure evidence, and final disposition needed to review the result without a prior revision. Remove superseded evidence instead of appending history.
 - Final disposition must be one of: local fix completed, normal architecture plan required, or user clarification required.
 - Report root cause, changed files, scope and public-surface impact, L0/L1 results, applicable L2/L3 results, baseline tests added or skipped with reason, generated-context regeneration or freshness check when applicable, final disposition, and the Debug completion evidence path when code was changed.
 
@@ -219,6 +219,8 @@ Small diff, minimum change, localized fix, or preserving the current implementat
 9. \`Implementation And Validation\`
 10. \`Final Disposition\`
 
+Each rewritten \`architecture-diagnosis.md\` must be a complete, self-contained current diagnosis and implementation result. Restate all still-relevant code-reading closure, evidence, architecture findings, changes, validation, and remaining failure evidence; do not refer to a prior round or superseded diagnosis as evidence.
+
 \`Implementation And Validation\` must use these subsections: \`Changed Files And Public Surface\`, \`Baseline Tests\`, \`Diagnostic And L0/L1 Validation\`, \`L2/L3 Validation\`, \`Generated Context\`, and \`Commit\`.
 
 \`L2/L3 Validation\` must use this table:
@@ -307,6 +309,7 @@ Small diff, minimum change, localized fix, or preserving the current implementat
 - Write \`.ai/vcm/handoffs/docs-sync-report.md\` for post-validation docs sync in Code-Change Flow, Architect Debug Flow, or a code-producing Architecture Diagnosis Flow. Do not write it for Docs-Only Flow or a Debug/Diagnosis Branch.
 - In Docs-Only Flow, the Architect role result must record the decision, changed documents, evidence reviewed, checks performed, and commit.
 - The report records decision, evidence reviewed, current-truth reconciliation, generated-context freshness, cross-document consistency, architecture docs, active plans, testing-doc consistency, known-issues disposition, durable-doc audit command and result, docs updated, docs left unchanged, remaining documentation risks, and handoff notes.
+- Each rewritten \`docs-sync-report.md\` must be a complete, self-contained snapshot of the current docs-sync result and must not rely on a prior report revision.
 - \`Decision\` must be \`synced\`, \`unchanged\`, or \`blocked\`.
 
 ### Background Jobs

package/dist/backend/templates/harness/claude-root.js

@@ -66,6 +66,14 @@ If a reusable harness problem is suspected, it is enough to record a concise fee
 - Runtime task records and handoffs under \`.ai/vcm/\` are temporary. Durable facts must move into code, tests, PR text, commit history, or long-term docs.
 - Only architect writes \`.ai/vcm/handoffs/known-issues.md\`; other roles report unresolved findings back through their own handoff artifacts.
 
+## VCM Current Handoff Contract
+
+- A role-owned handoff under \`.ai/vcm/handoffs/\` is the complete current result for that artifact, not an append-only log or a pointer to an earlier revision.
+- Whenever a handoff is rewritten, make the new revision self-contained: restate every still-relevant decision, evidence item, command, result, coverage mapping, finding, approval, and remaining action needed to interpret the current result without another round's artifact.
+- Remove or replace superseded content. Do not use a prior round, prior report revision, consumed route message, role Session, or transcript as a substitute for evidence in the current handoff.
+- A handoff may cite current code, durable docs, commits, preserved job output, or request-scoped Gate Review evidence that still exists at the cited path.
+- Before routing an artifact reference, confirm the referenced handoff satisfies this contract.
+
 ## User Communication
 
 - A message without a VCM marker is user communication.

package/dist/backend/templates/harness/coder-agent.js

@@ -70,7 +70,7 @@ ${renderRoleMemoryRules("coder")}
 
 ### Handoff
 
-- Write \`.ai/vcm/handoffs/coder-completion.md\` before routing back to project-manager. This file is the current implementation completion evidence, not a log; replace stale content instead of appending history.
+- Write \`.ai/vcm/handoffs/coder-completion.md\` before routing back to project-manager. This file is the complete, self-contained current implementation completion evidence, not a log. Each revision must restate every Scaffold Manifest disposition, changed file, helper, deviation, generated-context result, baseline-test change, L0/L1 command and result, worker result, commit, and objective failure still needed to review the current implementation without a prior revision. Replace stale content instead of appending history.
 - \`coder-completion.md\` must include \`Decision: ready_for_review | incomplete | failed\`.
 - \`coder-completion.md\` must report every Scaffold Manifest item disposition in the fixed Scaffold Completion table, plus changed files, private helpers added, manifest deviations as report-only facts, generated context status, baseline tests added or updated, L0/L1 commands and results, worker commits and integration status when workers were used, and compile/typecheck or L0/L1 failures.
 - Use this structure:

package/dist/backend/templates/harness/gate-review.js

@@ -8,6 +8,8 @@ ${renderRoleMemoryRules("reviewer")}
 
 Review only the gate in the VCM prompt. Use the task and worktree paths named there. Project memory may orient you, but only current worktree evidence can decide the gate.
 
+When the VCM prompt maps a task-evidence path to an immutable request snapshot, review that snapshot as the handoff or prior-Gate input for this request. Do not substitute a later rewritten live handoff. Continue to inspect current code, tests, durable docs, generated context, and the named commit range wherever the gate requires current-worktree evidence.
+
 Use only these decisions:
 
 - \`approve\`: required gate evidence is present, current, internally consistent, sufficient for that gate, and has no gate-blocking finding.

package/dist/backend/templates/harness/harness-engineer-agent.js

@@ -39,21 +39,22 @@ You are not part of the task workflow round state.
   yourself.
 - Retrospective Mode: analyze a completed task for reusable harness problems.
   When the assigned prompt includes Auto Memory Review, also review the memory
-  proposals and write only the reviewed-memory output files assigned by VCM.
+  proposals, update the active \`<VCM-memory>\` blocks, and commit those memory
+  changes yourself.
 - VCM Feedback Mode: draft VCM product, installer, UI, or fixed-template issue
   feedback. Do not submit without explicit in-session user authorization.
 
 ## Change Policy
 
-- Apply edits only in Bootstrap Apply Mode, to assigned reviewed-memory output
-  files during Retrospective Mode, or when VCM explicitly asks you to apply an
-  approved harness change.
+- Apply edits only in Bootstrap Apply Mode, to active \`<VCM-memory>\` blocks
+  during an assigned Auto Memory Retrospective, or when VCM explicitly asks you
+  to apply an approved harness change.
 - When applying edits, work only in the active task worktree named by VCM. Do not
   edit the base repository root unless VCM explicitly says so.
 - In Proposal Mode, do not edit files.
 - In Retrospective Mode, write the assigned retrospective report and, only when
-  Auto Memory Review is included in the prompt, the assigned reviewed-memory
-  output files. After every assigned pending feedback has a recorded
+  Auto Memory Review is included in the prompt, directly update the assigned
+  active memory blocks. After every assigned pending feedback has a recorded
   disposition, delete those processed feedback files.
 - Commit every applied harness change yourself before ending your turn.
 - Do not overwrite VCM fixed managed blocks.
@@ -92,9 +93,13 @@ You are not part of the task workflow round state.
   assigned by VCM.
 - Do not record task narrative, temporary state, unverified conclusions, or
   Harness rules in memory.
-- Edit only the review output paths assigned by VCM. Do not edit active
-  \`<VCM-memory>\` blocks directly. VCM applies the reviewed output, records the
-  diff, and commits the changed host files.
+- Edit only the \`<VCM-memory>\` blocks in the active memory files assigned by
+  VCM. Do not change surrounding role definitions, project context, or managed
+  Harness blocks during Auto Memory Review.
+- If the reviewed memory changes, commit only the changed active memory files
+  with commit message \`chore: update VCM memory\` before ending the turn. If
+  memory is unchanged, do not create a commit. VCM records the committed result,
+  diff, and review history; it does not apply or commit the memory for you.
 
 ## Task Harness Retrospective
 

package/dist/backend/templates/harness/tester-agent.js

@@ -179,7 +179,7 @@ L3 Required: yes|no
 - When \`L3 Required: yes\`, include at least one complete flow-to-case mapping. \`Action\` must be \`run-existing\`, \`updated\`, or \`added\`.
 - When \`L3 Required: no\`, use \`Not-Required Evidence\` to prove every condition in the L3 not-required rule.
 - In every flow, if tests, fixtures, test-only helpers, or \`docs/TESTING.md\` changed, commit those changes before reporting a terminal result and record the changed files and commit in \`test-report.md\`. If no tracked files changed, record that no commit was required.
-- \`test-report.md\` is the current validation evidence, not a log; when rewriting it, carry forward still-unresolved findings or explicitly mark them resolved instead of dropping them.
+- \`test-report.md\` is the complete, self-contained current validation evidence, not a log. Each revision must independently support its current \`Test Result\` by restating every still-relevant test or external check, coverage mapping, command and result, test-infrastructure fact, failed expectation, skipped check, gap, approval, blocking issue, and remaining validation needed to review that result without a prior revision. Remove superseded evidence, but never replace current evidence with “as recorded in the prior round”, a prior report reference, or a Session/transcript reference.
 - In \`Coverage Mapping\`, map each accepted changed behavior or relevant risk to its validation level, actual test file and case or external evidence, exercised entry path and key assertions, result, and any remaining gap.
 - In \`Validation Progress\`, record \`Completed Validation\` and \`Remaining Validation\`. A final \`pass\` report must set remaining validation to \`None\`.
 - Use \`pass\` only when required validation completed and no blocking test failure, missing required coverage, unacceptable test weakness, or unresolved validation risk remains.

package/dist/backend/templates/harness/vcm-final-acceptance-skill.js

@@ -123,6 +123,8 @@ accepted|accepted-with-known-risks|needs-coder-follow-up|needs-architect-follow-
 ## Final User Summary
 \`\`\`
 
+Each rewrite must be a complete, self-contained snapshot of the current acceptance decision and supporting evidence. Restate every still-relevant evidence result, file classification, validation result, Gate decision, docs-sync result, known-issues disposition, cleanup result, risk, and next action; do not rely on a prior acceptance revision, route message, Session, or transcript.
+
 The final user summary should be concise and include files changed, validation, docs updates, open risks, and next action.
 `;
 }

package/dist/backend/templates/harness/vcm-long-running-validation-skill.js

@@ -7,12 +7,14 @@ Never run the Bash tool with \`run_in_background: true\`, and never detach a pro
 
 The only sanctioned long-running mechanism is \`.ai/tools/run-long-check\` plus \`.ai/tools/watch-job\` through this skill.
 
+Each run-long-check or watch-job invocation must be the only top-level command in its Bash tool call. Do not pipe it, append or prepend another command, place it in a conditional chain or subshell, capture it with command substitution, or invoke it through a shell command string. Ordinary redirection is allowed. A directly passed script such as \`-- bash /tmp/check.sh\` is allowed; \`-- bash -c '...'\` is not.
+
 The hard ceiling is 60 minutes per job, enforced by the job worker itself. No approval can raise this ceiling; split larger operations into jobs that each fit within it.
 
 ## Protocol
 
 1. Start the command with an explicit ceiling: \`.ai/tools/run-long-check --timeout <duration> -- <command>\`. Pass the validation executable and its arguments directly. Do not use a shell command-string wrapper, pipeline its output, or append another command: run-long-check already captures stdout/stderr, and shell wrappers can hide the validation exit code. Pick the ceiling from \`docs/TESTING.md\` guidance or a realistic estimate, never above 60m. The tool prints the job id and creates job state under \`.ai/vcm/jobs/<job-id>/\`.
-2. In the same turn, run \`.ai/tools/watch-job <job-id>\`. The default watch window is 8 minutes.
+2. In the same turn, run \`.ai/tools/watch-job <job-id>\` as its own Bash tool call. The default watch window is 8 minutes.
 3. If watch-job exits 125, the job is still running: run \`.ai/tools/watch-job <job-id>\` again immediately. Do not end the turn between windows.
 4. Repeat until watch-job reports a terminal result.
 5. Read the final status and the relevant log tail.

package/dist/backend/templates/harness/vcm-route-message-skill.js

@@ -44,7 +44,7 @@ If the same route file already contains a not-yet-delivered message, update that
 
 ## Message Format
 
-Use the smallest body that is complete. Include artifact refs instead of copying long documents.
+Use the smallest body that is complete. Include artifact refs instead of copying long documents. Reference a role handoff only after confirming its current revision is complete and self-contained; never use an artifact ref to stand in for evidence that exists only in a prior revision, consumed route message, Session, or transcript.
 
 For simple user relay, use a lightweight body instead of the formal dispatch format.
 

package/dist/shared/types/session.js

@@ -61,6 +61,7 @@ export const CCR_GATEWAY_BASE_URLS = [
 ];
 export const CCR_GPT_MODEL_ID = "Codex API/gpt-5.6-sol";
 export const CCR_GPT_SESSION_MODEL = `ccr:${CCR_GPT_MODEL_ID}`;
+export const CCR_GPT_EFFECTIVE_CONTEXT_TOKENS = 258_400;
 export function isCcrSessionModel(model) {
     return model === CCR_GPT_SESSION_MODEL;
 }