npm - @code-partner/codepipe-hub - Versions diffs - 0.14.1-dev.399.g96d6330f → 0.14.1-dev.404.g2e5e29f1 - Mend

@code-partner/codepipe-hub 0.14.1-dev.399.g96d6330f → 0.14.1-dev.404.g2e5e29f1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (7) hide show

package/admin/dist/assets/{index-C1be47HC.js → index-B2EvJfMD.js} +8 -8
package/admin/dist/index.html +1 -1
package/dashboard/dist/assets/{ModelCanvas-B2M4zQnV.js → ModelCanvas-3KSH8-83.js} +1 -1
package/dashboard/dist/assets/{index-Cpy8aKZb.js → index-T-kK12wa.js} +121 -121
package/dashboard/dist/index.html +1 -1
package/hub/dist/index.js +239 -49
package/package.json +1 -1

package/dashboard/dist/index.html CHANGED Viewed

@@ -27,7 +27,7 @@
         localStorage.getItem('theme') ?? 'dark'
       );
     </script>
-    <script type="module" crossorigin src="/assets/index-Cpy8aKZb.js"></script>
+    <script type="module" crossorigin src="/assets/index-T-kK12wa.js"></script>
     <link rel="stylesheet" crossorigin href="/assets/index-CD8JrtmW.css">
   </head>
   <body>

package/hub/dist/index.js CHANGED Viewed

@@ -13,7 +13,7 @@ var PLATFORM_VERSION;
 var init_version_generated = __esm({
   "../packages/shared/dist/version.generated.js"() {
     "use strict";
-    PLATFORM_VERSION = "0.14.1-dev.399.g96d6330f";
+    PLATFORM_VERSION = "0.14.1-dev.404.g2e5e29f1";
   }
 });
@@ -394,22 +394,25 @@ var init_job = __esm({
 });
 // ../packages/shared/dist/llm-models.js
+function modelForHarness(model, harness) {
+  if (!harness)
+    return model;
+  return HARNESS_MODEL_LADDER[harness]?.[model] ?? model;
+}
 function parseWorkflowPreset(v) {
   return typeof v === "string" && WORKFLOW_PRESETS.includes(v.trim()) ? v.trim() : null;
 }
 function emptyModelConfig() {
   return { default: null, roles: {} };
 }
-function resolveModelForRole(cfg, role, preset = DEFAULT_WORKFLOW_PRESET) {
+function resolveModelForRole(cfg, role, preset = DEFAULT_WORKFLOW_PRESET, harness) {
   const explicit = cfg?.roles?.[role];
   if (typeof explicit === "string" && explicit.trim().length > 0)
     return explicit.trim();
   if (cfg?.default && cfg.default.trim().length > 0)
     return cfg.default.trim();
   const presetModel = PRESET_ROLE_MODELS[preset]?.[role];
-  if (presetModel)
-    return presetModel;
-  return RECOMMENDED_ROLE_MODELS[role] ?? FALLBACK_MODEL;
+  return modelForHarness(presetModel ?? RECOMMENDED_ROLE_MODELS[role] ?? FALLBACK_MODEL, harness);
 }
 function resolveProvider(cfg) {
   const p = cfg?.provider;
@@ -532,7 +535,7 @@ function deserializeModelConfig(json) {
     return emptyModelConfig();
   }
 }
-var DEFAULT_LLM_PROVIDER, LLM_PROVIDER_SUGGESTIONS, MODEL_OPUS, MODEL_SONNET, MODEL_HAIKU, MODEL_ALIAS_SUGGESTIONS, BUILD_DISCOVERY, CODE_REVIEW_VERIFY, MODEL_SUB_ROLES, ALL_MODEL_ROLE_KEYS, RECOMMENDED_ROLE_MODELS, FALLBACK_MODEL, WORKFLOW_PRESETS, DEFAULT_WORKFLOW_PRESET, PRESET_ROLE_MODELS, DEFAULT_REGIMEN_ALIASES, EFFORT_LEVELS, EFFORT_LEVEL_SUGGESTIONS, RECOMMENDED_ROLE_EFFORT;
+var DEFAULT_LLM_PROVIDER, LLM_PROVIDER_SUGGESTIONS, MODEL_OPUS, MODEL_SONNET, MODEL_HAIKU, MODEL_ALIAS_SUGGESTIONS, BUILD_DISCOVERY, CODE_REVIEW_VERIFY, MODEL_SUB_ROLES, ALL_MODEL_ROLE_KEYS, RECOMMENDED_ROLE_MODELS, FALLBACK_MODEL, HARNESS_MODEL_LADDER, WORKFLOW_PRESETS, DEFAULT_WORKFLOW_PRESET, PRESET_ROLE_MODELS, DEFAULT_REGIMEN_ALIASES, EFFORT_LEVELS, EFFORT_LEVEL_SUGGESTIONS, RECOMMENDED_ROLE_EFFORT;
 var init_llm_models = __esm({
   "../packages/shared/dist/llm-models.js"() {
     "use strict";
@@ -604,6 +607,13 @@ var init_llm_models = __esm({
       testing: MODEL_OPUS
     };
     FALLBACK_MODEL = MODEL_OPUS;
+    HARNESS_MODEL_LADDER = {
+      codex: {
+        [MODEL_OPUS]: "gpt-5.6-sol",
+        [MODEL_SONNET]: "gpt-5.6-terra",
+        [MODEL_HAIKU]: "gpt-5.6-terra"
+      }
+    };
     WORKFLOW_PRESETS = ["express", "balanced", "full"];
     DEFAULT_WORKFLOW_PRESET = "full";
     PRESET_ROLE_MODELS = {
@@ -711,6 +721,78 @@ var init_actions = __esm({
   }
 });
+// ../packages/shared/dist/capability-registry.js
+function harnessesForBackend(llm) {
+  switch (llm) {
+    case "claude-headless":
+      return ["claude-code"];
+    case "codex":
+      return ["codex"];
+    case "mock":
+      return ["mock", "claude-code", "codex"];
+    default:
+      return [];
+  }
+}
+var CAPABILITY_REGISTRY, SELECTABLE_HARNESSES;
+var init_capability_registry = __esm({
+  "../packages/shared/dist/capability-registry.js"() {
+    "use strict";
+    CAPABILITY_REGISTRY = [
+      {
+        harness: "claude-code",
+        endpoint: "anthropic-managed",
+        modelPattern: "*",
+        credentialKinds: ["subscription_oauth", "api_key"],
+        status: "supported",
+        capabilities: {
+          headless: true,
+          // Claude Code can resume, but the product deliberately does not glue
+          // sessions across jobs (DEV-682) — so no resume is advertised.
+          resume: false,
+          streamEvents: true,
+          usage: "tokens",
+          toolPolicy: "vendor"
+        },
+        testedWith: { runtimeVersion: "claude-code 2.x (stream-json)", protocolVersion: 7 }
+      },
+      {
+        harness: "codex",
+        endpoint: "openai-managed",
+        modelPattern: "*",
+        credentialKinds: ["subscription_oauth", "api_key"],
+        // EXPERIMENTAL until a cross-harness eval run compares it with the Claude
+        // path on the same corpus (§12) — claiming "supported" before that would
+        // be a promise nothing backs.
+        status: "experimental",
+        capabilities: {
+          headless: true,
+          // The CLI can resume a session, but the adapter exposes no resume() and
+          // the product does not glue sessions across JOBS — the registry must
+          // say what the ADAPTER can do, or a job requiring resume would be
+          // matched to a worker that cannot. (DEV-897 uses `exec resume` WITHIN a
+          // chat job; that is one session, not glue between jobs.)
+          resume: false,
+          streamEvents: true,
+          usage: "quota",
+          toolPolicy: "vendor"
+        },
+        testedWith: { runtimeVersion: "codex-cli 0.152 (exec --json, exec resume)", protocolVersion: 7 }
+      },
+      {
+        harness: "mock",
+        endpoint: "none",
+        modelPattern: "*",
+        credentialKinds: [],
+        status: "supported",
+        capabilities: { headless: true, resume: false, streamEvents: false, usage: "none", toolPolicy: "vendor" },
+        testedWith: { runtimeVersion: "built-in stub", protocolVersion: 7 }
+      }
+    ];
+    SELECTABLE_HARNESSES = CAPABILITY_REGISTRY.filter((e) => e.harness !== "mock").map((e) => e.harness);
+  }
+});
 // ../packages/shared/dist/project.js
 function slugify(input) {
   const lowered = input.toLowerCase();
@@ -1162,7 +1244,9 @@ var init_project_settings = __esm({
   "../packages/shared/dist/project-settings.js"() {
     "use strict";
     init_actions();
+    init_capability_registry();
     init_git_flow();
+    init_job();
     init_llm_models();
     init_project_map();
     init_project_language();
@@ -1313,10 +1397,58 @@ var init_project_settings = __esm({
             // DEV-896: the old text ("currently only claude is available") outlived
             // its truth and read as "there is nothing to choose" — the one thing
             // help must never say about a live control.
-            help: "The agent harness every role runs on by default. Per-role overrides (and the fallback agent) live in the shared config's routing block."
+            // DEV-934: per-role agents are a control now, not a YAML instruction —
+            // point at it, and say which of the two wins, because they live in
+            // different files and one is personal.
+            help: "The agent every role runs on by default \u2014 your own choice, not the team's. \xABAgent by role\xBB below is shared and outranks it."
           }
         ],
         groups: [
+          {
+            // DEV-934: the model was pickable per role from the start (below) and
+            // the AGENT was not — a project mixing vendors (analysis on one,
+            // implementation on the other) could only be expressed by hand-editing
+            // project.yaml, which put the whole DEV-759 routing feature out of
+            // reach of anyone who does not edit YAML.
+            id: "routing",
+            label: "Agent by role",
+            help: "Which agent runs which stage. Shared with the team (project.yaml). Empty \u21D2 the role follows the default below it, then your own provider setting above.",
+            fields: [
+              {
+                id: "routing.default.harness",
+                file: "project.yaml",
+                path: ["routing", "default", "harness"],
+                type: "enum",
+                label: "default (whole project)",
+                nullable: true,
+                suggestions: [...SELECTABLE_HARNESSES],
+                help: "The team-wide default. Set, it outranks each member's own provider setting."
+              },
+              ...ALL_JOB_ROLES.map((role) => ({
+                id: `routing.${role}.harness`,
+                file: "project.yaml",
+                path: ["routing", role, "harness"],
+                type: "enum",
+                label: role,
+                nullable: true,
+                suggestions: [...SELECTABLE_HARNESSES],
+                help: "Empty \u21D2 this role follows the default."
+              })),
+              {
+                id: "routing.fallback.harness",
+                file: "project.yaml",
+                path: ["routing", "fallback", "harness"],
+                type: "enum",
+                label: "fallback (retry after a vendor failure)",
+                nullable: true,
+                suggestions: [...SELECTABLE_HARNESSES],
+                // DEV-759 §9: the ONE sanctioned way to change vendor mid-task, and
+                // only for infrastructure trouble — a wrong result is a rework, not
+                // a reason to try the other vendor.
+                help: "Where a retry goes when the vendor itself was unusable (rate limit, 5xx). Empty \u21D2 retries stay on the same agent."
+              }
+            ]
+          },
           {
             id: "llm.models",
             label: "Models by role",
@@ -2988,25 +3120,6 @@ var init_credential_broker = __esm({
   }
 });
-// ../packages/shared/dist/capability-registry.js
-function harnessesForBackend(llm) {
-  switch (llm) {
-    case "claude-headless":
-      return ["claude-code"];
-    case "codex":
-      return ["codex"];
-    case "mock":
-      return ["mock", "claude-code", "codex"];
-    default:
-      return [];
-  }
-}
-var init_capability_registry = __esm({
-  "../packages/shared/dist/capability-registry.js"() {
-    "use strict";
-  }
-});
 // ../packages/shared/dist/structured-llm.js
 var init_structured_llm = __esm({
   "../packages/shared/dist/structured-llm.js"() {
@@ -6457,6 +6570,17 @@ function mergeSandboxCapabilities(stored, live) {
   }
   return merged;
 }
+async function claimCapabilities(db, sandboxId, stored, live) {
+  const merged = mergeSandboxCapabilities(stored, live);
+  if (JSON.stringify(merged) !== JSON.stringify(stored)) {
+    await db.run(
+      `UPDATE sandboxes SET capabilities_json = ? WHERE id = ?`,
+      JSON.stringify(merged),
+      sandboxId
+    );
+  }
+  return merged;
+}
 async function refreshSandboxCapabilities(db, sandboxId, live) {
   const row = await db.get(
     `SELECT capabilities_json as capabilitiesJson FROM sandboxes WHERE id = ?`,
@@ -10691,6 +10815,63 @@ async function heldByTrancheDeferral(db) {
   );
   return rows.map((r) => r.id);
 }
+async function resolveJobAxes(db, projectId, role, taskWorkflowPreset) {
+  const mc = await loadModelConfig(db, projectId);
+  const provider = resolveProvider(mc);
+  const { preset: projectPreset } = await loadWorkflowSettings(db, projectId);
+  const workflowPreset = parseWorkflowPreset(taskWorkflowPreset) ?? projectPreset;
+  const routing = resolveRoutingForRole(await loadRoutingConfig(db, projectId), role, {
+    defaultHarness: harnessForProvider(provider)
+  });
+  return {
+    routing,
+    provider: HARNESS_PROVIDER[routing.harness] ?? provider,
+    model: routing.model || resolveModelForRole(mc, role, workflowPreset, routing.harness),
+    modelDiscovery: role === "build" ? resolveModelForRole(mc, "build-discovery", workflowPreset, routing.harness) : void 0
+  };
+}
+function axesOnSpec(axes) {
+  return {
+    harness: axes.routing.harness,
+    model: axes.model,
+    endpoint: axes.routing.endpoint,
+    account: axes.routing.account,
+    // The harness a run needs is a capability of the worker (DEV-758 matches
+    // on these).
+    requiredCapabilities: [`harness:${axes.routing.harness}`, `provider:${axes.provider}`]
+  };
+}
+async function restampQueuedJobs(db, projectId) {
+  const rows = await db.all(
+    `SELECT id, role, task_id AS taskId, spec_json AS specJson
+       FROM jobs WHERE project_id = ? AND state = 'queued'`,
+    projectId
+  );
+  let restamped = 0;
+  for (const row of rows) {
+    let spec;
+    try {
+      spec = JSON.parse(row.specJson);
+    } catch {
+      continue;
+    }
+    if (!spec.executionSpec) continue;
+    const task = await getTask(db, row.taskId);
+    const axes = await resolveJobAxes(db, projectId, row.role, task?.workflowPreset);
+    const next = {
+      ...spec,
+      model: axes.model,
+      provider: axes.provider,
+      ...axes.modelDiscovery ? { modelDiscovery: axes.modelDiscovery } : {},
+      executionSpec: { ...spec.executionSpec, ...axesOnSpec(axes) }
+    };
+    const nextJson = JSON.stringify(next);
+    if (nextJson === row.specJson) continue;
+    await engineStore(db).jobs.patchSpec(row.id, nextJson);
+    restamped += 1;
+  }
+  return restamped;
+}
 async function createJob(db, input) {
   const id = `job_${nanoid12(16)}`;
   const now = Date.now();
@@ -10707,16 +10888,12 @@ async function createJob(db, input) {
   if (!isAutoLanguage(responseLanguage)) metadata["responseLanguage"] = responseLanguage;
   const attachDeliveries = input.role === "build" || input.role === "code-review" || input.attachDeliveries === true || input.role === "implement" && ((input.task.buildIteration ?? 0) > 0 || (input.task.reviewIteration ?? 0) > 0);
   const deliveries = attachDeliveries ? Object.entries(await loadDeliveries(db, input.task.id)).map(([name, d]) => ({ name, ...d })) : [];
-  const mc = await loadModelConfig(db, input.projectId);
-  const { preset: projectPreset } = await loadWorkflowSettings(db, input.projectId);
-  const workflowPreset = parseWorkflowPreset(input.task.workflowPreset) ?? projectPreset;
-  const model = resolveModelForRole(mc, input.role, workflowPreset);
+  const axes = await resolveJobAxes(db, input.projectId, input.role, input.task.workflowPreset);
+  const { modelDiscovery } = axes;
   const skillsRow = await db.get(
     `SELECT skills_json AS json FROM projects WHERE id = ?`,
     input.projectId
   );
-  const provider = resolveProvider(mc);
-  const modelDiscovery = input.role === "build" ? resolveModelForRole(mc, "build-discovery", workflowPreset) : void 0;
   const ec = await loadEffortConfig(db, input.projectId);
   const effort = resolveEffortForRole(ec, input.role);
   const effortDiscovery = input.role === "build" ? resolveEffortForRole(ec, "build-discovery") : void 0;
@@ -10725,23 +10902,13 @@ async function createJob(db, input) {
   const isReviewVerify = input.metadata?.["reviewVerify"] === true || input.metadata?.["externalReviewVerify"] === true;
   const councilPhase = input.role === "council" ? "questions" : input.role === "code-review" && !isReviewVerify ? "codeReview" : null;
   const council = councilPhase ? await loadCouncilSettings(db, input.projectId) : null;
-  const routing = resolveRoutingForRole(await loadRoutingConfig(db, input.projectId), input.role, {
-    defaultHarness: harnessForProvider(provider)
-  });
-  const effectiveModel = routing.model || model;
-  const effectiveProvider = HARNESS_PROVIDER[routing.harness] ?? provider;
+  const { model: effectiveModel, provider: effectiveProvider } = axes;
   const executionSpec = {
     schemaVersion: EXECUTION_SPEC_SCHEMA_VERSION,
     jobId: id,
     role: input.role,
-    harness: routing.harness,
-    model: effectiveModel,
-    endpoint: routing.endpoint,
-    account: routing.account,
-    budgets: {},
-    // The harness a run needs is a capability of the worker (DEV-758 matches
-    // on these).
-    requiredCapabilities: [`harness:${routing.harness}`, `provider:${effectiveProvider}`]
+    ...axesOnSpec(axes),
+    budgets: {}
   };
   const spec = {
     jobId: id,
@@ -16569,7 +16736,7 @@ async function enqueueExternalReviewVerifyJob(db, task, meta, firstVerdict, log)
     metadata: { externalReviewVerify: true, reviewPanel: false, externalReview: true }
   });
   const patched = await patchSpecForExternalReview(db, task.projectId, spec, meta, {});
-  patched.model = resolveModelForRole(await loadModelConfig(db, task.projectId), "code-review-verify", (await loadWorkflowSettings(db, task.projectId)).preset);
+  patched.model = resolveModelForRole(await loadModelConfig(db, task.projectId), "code-review-verify", (await loadWorkflowSettings(db, task.projectId)).preset, patched.executionSpec?.harness);
   const verifyEffort = resolveEffortForRole(
     await loadEffortConfig(db, task.projectId),
     "code-review-verify"
@@ -19563,7 +19730,9 @@ async function enqueueInternalReviewVerifyJob(db, task, firstVerdict, log) {
   const patched = {
     ...specRest,
     // DEV-660: the verifier runs on its own (cheaper) model key.
-    model: resolveModelForRole(await loadModelConfig(db, task.projectId), "code-review-verify", (await loadWorkflowSettings(db, task.projectId)).preset),
+    // DEV-930: resolved for the harness THIS job was routed to — the cheaper
+    // key is a tier, and the tier means a different model per vendor.
+    model: resolveModelForRole(await loadModelConfig(db, task.projectId), "code-review-verify", (await loadWorkflowSettings(db, task.projectId)).preset, spec.executionSpec?.harness),
     ...verifyEffort !== void 0 ? { effort: verifyEffort } : {}
   };
   await engineStore(db).jobs.patchSpec(spec.jobId, JSON.stringify(patched));
@@ -25173,6 +25342,8 @@ async function rowToSummary3(db, row) {
   const mc = await loadModelConfig(db, row.id);
   summary.llmProvider = resolveProvider(mc);
   if (mc.default !== null || Object.keys(mc.roles).length > 0) summary.modelConfig = mc;
+  const routing = await loadRoutingConfig(db, row.id);
+  if (routing) summary.routing = routing;
   const ec = await loadEffortConfig(db, row.id);
   if (ec.default !== null || Object.keys(ec.roles).length > 0) summary.effortConfig = ec;
   const intake = await loadObserverIntakeSettings(db, row.id);
@@ -27209,6 +27380,7 @@ init_lifecycle();
 init_cluster_recovery();
 // src/projects/context-observation.ts
+init_queue2();
 init_prompts();
 init_prompt_cache();
 init_ws();
@@ -27257,17 +27429,29 @@ function installContextObserver(db, log, opts = {}) {
     } catch (err) {
       log.warn({ err, projectId }, "config_observation: project.yaml apply failed");
     }
+    let regimenChanged = false;
     try {
-      await applyRegimenContent(
+      const res = await applyRegimenContent(
         db,
         projectId,
         typeof obs.regimenYaml === "string" ? obs.regimenYaml : null,
         log,
         sameDoc ? sharedParsed : void 0
       );
+      regimenChanged = res.changed;
     } catch (err) {
       log.warn({ err, projectId }, "config_observation: regimen.yaml apply failed");
     }
+    if (regimenChanged) {
+      try {
+        const restamped = await restampQueuedJobs(db, projectId);
+        if (restamped > 0) {
+          log.info({ projectId, jobs: restamped }, "queued jobs re-stamped for the project's routing");
+        }
+      } catch (err) {
+        log.warn({ err, projectId }, "config_observation: re-stamping queued jobs failed");
+      }
+    }
     try {
       applyActionsContent(projectId, obs.actions, obs.actionErrors);
     } catch (err) {
@@ -31049,6 +31233,7 @@ function latestQuestionsContent(artifacts) {
 // src/routes/jobs.ts
 init_context_outbox();
 init_deliveries();
+init_capabilities();
 init_sandbox_keys();
 init_sandbox2();
 init_ws();
@@ -31065,7 +31250,12 @@ async function jobsRoutes(app, deps) {
       app.log.warn({ sandboxId: sandbox.id }, "claim denied \u2014 SANDBOX protocol incompatible");
       return reply.status(204).send();
     }
-    const capabilities = JSON.parse(sandbox.capabilitiesJson);
+    const capabilities = await claimCapabilities(
+      db,
+      sandbox.id,
+      JSON.parse(sandbox.capabilitiesJson),
+      req.body?.capabilities
+    );
     const laneParam = req.query?.lane;
     const lane = laneParam === "planning" || laneParam === "implementation" ? laneParam : void 0;
     const deadline = Date.now() + config.jobClaimLongPollMs;

package/package.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "name": "@code-partner/codepipe-hub",
-  "version": "0.14.1-dev.399.g96d6330f",
+  "version": "0.14.1-dev.404.g2e5e29f1",
   "description": "The CodePipe hub, packaged for one machine: the CLI installs it on demand for the local profile.",
   "type": "module",
   "main": "./hub/dist/index.js",