@pome-sh/cli 0.2.0 → 0.4.0

package/dist/src/runner/runScenarioHosted.js

@@ -11,7 +11,7 @@ import { parseScenarioFile } from "../scenario/parseScenario.js";
 import { createHostedClient, perTwinReturnedByCloud, } from "../hosted/client.js";
 import { ensureMcpSuffix } from "../cli/session.js";
 import { redactJsonl, scoreFromFinalizeResponse, uploadRunBlobs, } from "../hosted/uploadAndFinalize.js";
-import { normalizeConfigAgentId, normalizeConfigAgentSdk, readProjectConfig, } from "../cli/project-config.js";
+import { resolveRunAgentIdentity } from "../cli/agent-identity.js";
 import { HostedOrchError, HostedTrialError } from "../hosted/errors.js";
 /** Build the agent subprocess env for a hosted run.
  *
@@ -96,6 +96,12 @@ export function buildAgentEnv(params) {
             // would reject.
             env.POME_SLACK_TOKEN = session.agent_token;
         }
+        else if (twin === "gmail") {
+            // Gmail uses the Pome session JWT directly. This is an alias for agent
+            // SDKs that conventionally read a provider-specific bearer variable;
+            // there is deliberately no provider_credentials.gmail shape.
+            env.POME_GMAIL_TOKEN = session.agent_token;
+        }
     }
     return env;
 }
@@ -107,18 +113,19 @@ export async function runScenarioHosted(options) {
     const startedAt = new Date().toISOString();
     const client = options.client ??
         createHostedClient({ baseUrl: options.hosted.baseUrl, apiKey: options.hosted.apiKey });
-    // ADR-013 — resolve agentId + agent.sdk from pome.config.json. Hosted runs
-    // are grouped under the agent the user registered with `pome register
-    // agent`. `agent.sdk` is a free-form string ("claude-agent-sdk",
-    // "openai-agents", "openclaw", "hermes", "custom") that the dashboard
-    // surfaces as a badge on every run. Both are optional during rollout.
-    let agentId;
-    let agentSdk = null;
-    const configRead = await readProjectConfig(dirname(options.scenarioPath));
-    if (configRead) {
-        agentId = normalizeConfigAgentId(configRead.config);
-        agentSdk = normalizeConfigAgentSdk(configRead.config);
-    }
+    // F-819 — resolve the agent id + version from the manifest (`agent.slug` →
+    // `.pome/link.json` cache, team-gated, re-resolved by slug on miss) and the
+    // `agent.framework` (the dashboard "SDK badge", formerly `agent.sdk`). The
+    // `agent_version` stamps the session; `--agent-version` overrides it. All are
+    // optional during rollout — a manifest-less repo runs unattributed.
+    const identity = await resolveRunAgentIdentity({
+        startDir: dirname(options.scenarioPath),
+        apiBaseUrl: options.hosted.baseUrl,
+        agentVersionOverride: options.agentVersion,
+    });
+    const agentId = identity.agentId;
+    const agentVersion = identity.agentVersion;
+    const agentSdk = identity.framework ?? null;
     // Per-run signals file. POME_ADAPTER_SIGNALS_PATH is injected into the agent
     // env so adapter-rich correlator gets step/tool_call signals when the agent
     // emits them; otherwise the heuristic correlator path runs over events alone.
@@ -138,6 +145,7 @@ export async function runScenarioHosted(options) {
         scenarioSource,
         twins: scenario.config.twins,
         agentId,
+        agentVersion,
         seed: scenario.seedState,
     });
     // Multi-twin (M3) locals. `twins[0]` is the primary twin: the source of the
@@ -204,6 +212,14 @@ export async function runScenarioHosted(options) {
         });
         let agentResult = preflight;
         if (preflight.exitCode === 0) {
+            // F-771 — the preflight probe appended its own signals (turns/steps/
+            // tool_calls) to the shared POME_ADAPTER_SIGNALS_PATH file. Those must
+            // NOT reach the uploaded signals.jsonl / usage ledger. Truncate before
+            // the real run so the blob counts real-run telemetry only. runAgentCommand
+            // awaits child exit, so the preflight's writes are already flushed here.
+            // Gated on preflight success: the preflight-failed → single-run path keeps
+            // its current behavior (it uploads the preflight's signals).
+            await writeFile(signalsPath, "");
             agentResult = await runAgentCommand({
                 command: options.agentCommand,
                 env,
@@ -410,15 +426,16 @@ export async function runScenarioHosted(options) {
         //    storage, calls the managed judge via AI Gateway, persists the run
         //    row, and returns the authoritative score the dashboard records.
         //    The CLI prints this score on the `score:` line.
-        // `CriterionDef.kind` on the finalize request is still the legacy `D`/`P`
-        // wire vocabulary; map the canonical `code`/`model` kind back for the wire.
+        // The finalize wire carries the canonical `code`/`model` kinds (F-778;
+        // needs shared-types >= 0.10.0 cloud-side, whose tolerant reader also
+        // still accepts the legacy `D`/`P` spellings from older released CLIs).
         // Multi-twin (M3): carry the per-criterion twin attribution so the cloud
-        // judge checks each [D:<twin>] against that twin's state. Absent = the
+        // judge checks each criterion against its twin's state. Absent = the
         // primary twin (single-twin scenarios omit it entirely).
         const criteriaDefs = scenario.criteria.map((c, idx) => ({
             id: `crit_${idx}`,
             text: c.text,
-            kind: c.type === "code" ? "D" : "P",
+            kind: c.type,
             ...(c.twin ? { twin: c.twin } : {}),
         }));
         const stopReason = agentResult.timedOut

package/dist/src/runner/runTrialGroup.d.ts

@@ -12,7 +12,7 @@ export interface RunTrialGroupOptions {
     scenarioPath: string;
     agentCommand: string;
     /** Where the agent command came from, for the header copy
-     *  ("pome.config.json" | "--agent" | "built-in default"). */
+     *  ("pome.json" | "--agent" | "built-in default"). */
     agentCommandSource?: string;
     /** Effective k (2..20). k=1 must take the single-run path instead. */
     trials: number;
@@ -26,6 +26,8 @@ export interface RunTrialGroupOptions {
     dashboardBaseUrl: string;
     /** Informational; forwarded to every trial's `runs.agent_model`. */
     agentModel?: string;
+    /** F-819 — `--agent-version` override, forwarded to every trial's session. */
+    agentVersion?: string;
     /** FDRS-644 — the literal command the fix handoff tells the user to
      *  re-run after their coding agent applies a fix. The caller knows the
      *  invocation shape (bare `pome run` vs an explicit path + -n); default

package/dist/src/runner/runTrialGroup.js

@@ -42,7 +42,7 @@ import { newGroupId } from "../demo/ids.js";
 import { criterionPhrase } from "../demo/render.js";
 import { parseScenarioFile } from "../scenario/parseScenario.js";
 import { outcomeOf } from "../hosted/evalResultView.js";
-import { normalizeConfigAgentId, normalizeConfigAgentSlug, readProjectConfig, } from "../cli/project-config.js";
+import { resolveRunAgentIdentity } from "../cli/agent-identity.js";
 import { runScenarioHosted } from "./runScenarioHosted.js";
 import { fixHandoffLines, flagHintLine, groupExitCode, groupSummaryLines, mostCommonFailedCriterion, provisioningLine, shortReason, spawningAgentLine, trialRowLine, } from "./groupRender.js";
 /** Per-request transport timeout. Durable evaluation has a separate budget. */
@@ -65,18 +65,19 @@ export async function runTrialGroup(options) {
             timeoutMs: GROUP_FINALIZE_TIMEOUT_MS,
         });
     const groupId = options.groupId ?? newGroupId();
-    const agentCommandSource = options.agentCommandSource ?? "pome.config.json";
+    const agentCommandSource = options.agentCommandSource ?? "pome.json";
     const scenario = await parseScenarioFile(options.scenarioPath);
     const scenarioSource = await readFile(options.scenarioPath, "utf8");
-    // Same agent resolution as the single-run path (ADR-013): the group's
-    // trials are recorded under the agent pinned in pome.config.json.
-    const configRead = await readProjectConfig(dirname(options.scenarioPath));
-    const agentId = configRead
-        ? normalizeConfigAgentId(configRead.config)
-        : undefined;
-    const agentSlug = configRead
-        ? normalizeConfigAgentSlug(configRead.config)
-        : undefined;
+    // Same agent resolution as the single-run path (F-819): the group's trials
+    // are recorded under the agent the manifest's `agent.slug` resolves to.
+    const identity = await resolveRunAgentIdentity({
+        startDir: dirname(options.scenarioPath),
+        apiBaseUrl: options.hosted.baseUrl,
+        agentVersionOverride: options.agentVersion,
+    });
+    const agentId = identity.agentId;
+    const agentVersion = identity.agentVersion;
+    const agentSlug = identity.agentSlug;
     out(flagHintLine(agentCommandSource));
     out("");
     const sleep = options.sleepFn ??
@@ -85,6 +86,7 @@ export async function runTrialGroup(options) {
         scenarioSource,
         twins: scenario.config.twins,
         agentId,
+        agentVersion,
         seed: scenario.seedState,
         groupId,
         // Fresh idempotency key per mint — the mint bodies are otherwise

package/dist/src/scenario/parseScenario.js

@@ -5,6 +5,8 @@ import { basename, extname } from "node:path";
 import { parse as parseYaml } from "yaml";
 import { z } from "zod";
 import { defaultSeedState, seedSchema } from "@pome-sh/twin-github";
+import { defaultSeedState as defaultGmailSeedState, gmailSeedSchema, } from "@pome-sh/twin-gmail";
+import { defaultSeedState as defaultLinearSeedState, linearSeedSchema as linearSeedStateSchema, } from "@pome-sh/twin-linear";
 import { parseGitHubSeedState } from "./githubSeedCompat.js";
 import { criterionSchema, scenarioConfigSchema, scenarioSchema, slackSeedStateSchema, stripeSeedStateSchema } from "./scenarioSchema.js";
 export async function parseScenarioFile(path) {
@@ -107,13 +109,13 @@ function buildSeedEnvelope(raw, twins) {
         return envelope;
     }
     if (raw === null || typeof raw !== "object" || Array.isArray(raw)) {
-        throw new Error(`Multi-twin scenarios need a per-twin seed envelope { <twin>: <seed> } for twins [${twins.join(", ")}], not a bare seed object.`);
+        throw new Error(`Multi-twin tasks need a per-twin seed envelope { <twin>: <seed> } for twins [${twins.join(", ")}], not a bare seed object.`);
     }
     const allowed = new Set(twins);
     const provided = raw;
     for (const key of Object.keys(provided)) {
         if (!allowed.has(key)) {
-            throw new Error(`Seed envelope key "${key}" is not one of the scenario's twins [${twins.join(", ")}].`);
+            throw new Error(`Seed envelope key "${key}" is not one of the task's twins [${twins.join(", ")}].`);
         }
     }
     for (const twin of twins) {
@@ -130,6 +132,10 @@ function parseSeedForTwin(twin, input) {
         return stripeSeedStateSchema.parse(input);
     if (twin === "slack")
         return slackSeedStateSchema.parse(input);
+    if (twin === "gmail")
+        return gmailSeedSchema.parse(input);
+    if (twin === "linear")
+        return linearSeedStateSchema.parse(input);
     return parseGitHubSeedState(input);
 }
 function defaultSeedForTwin(twin) {
@@ -141,19 +147,25 @@ function defaultSeedForTwin(twin) {
     if (twin === "slack") {
         return slackSeedStateSchema.parse({});
     }
+    if (twin === "gmail") {
+        return gmailSeedSchema.parse(defaultGmailSeedState());
+    }
+    if (twin === "linear") {
+        return linearSeedStateSchema.parse(defaultLinearSeedState());
+    }
     return seedSchema.parse(defaultSeedState());
 }
 function missingSidecarMessage(scenarioPath) {
-    const pathLabel = scenarioPath ?? "<scenario>.md";
+    const pathLabel = scenarioPath ?? "<task>.md";
     const sidecarLabel = scenarioPath
         ? scenarioPath.replace(/\.md$/i, ".seed.json")
-        : "<scenario>.seed.json";
+        : "<task>.seed.json";
     return [
-        `Scenario has a prose ## Seed State section but no compiled sidecar (${sidecarLabel}). Run:`,
+        `Task has a prose ## Seed State section but no compiled sidecar (${sidecarLabel}). Run:`,
         "",
         `    pome compile-seeds ${pathLabel}`,
         "",
-        "then re-run the scenario. (See `pome docs scenarios-github`.)",
+        "then re-run the task. (See `pome docs scenarios-github`.)",
     ].join("\n");
 }
 function stripSidecarMeta(seed) {
@@ -191,11 +203,17 @@ function splitSections(markdown) {
     }
     return sections;
 }
-// Criterion marker grammar (M3): `[D]` / `[P]` optionally carrying a twin tag,
-// `[D:<twin>]` / `[P:<twin>]`, where <twin> is `[a-z][a-z0-9_-]*`. The tag lands
-// on `criterion.twin`; a bare marker leaves it undefined (attributes to the
-// session's primary twin, `twins[0]`).
-const CRITERION_LINE_RE = /^[-*]\s+\[([DP])(?::([a-z][a-z0-9_-]*))?\]\s+(.+)$/;
+// Criterion marker grammar (F-778): `[code]` / `[model]` optionally carrying a
+// twin tag, `[code:<twin>]` / `[model:<twin>]`, where <twin> is
+// `[a-z][a-z0-9_-]*`. The marker spells the canonical criterion kind directly.
+// The tag lands on `criterion.twin`; a bare marker leaves it undefined
+// (attributes to the session's primary twin, `twins[0]`).
+const CRITERION_LINE_RE = /^[-*]\s+\[(code|model)(?::([a-z][a-z0-9_-]*))?\]\s+(.+)$/;
+// The retired pre-F-778 marker spelling, matched ONLY to fail loudly. Without
+// this guard a legacy `[D]`/`[P]` line would fall through the silent
+// skip-non-criterion path below and the scenario would "pass" with fewer
+// criteria than its author wrote.
+const LEGACY_CRITERION_LINE_RE = /^[-*]\s+\[([DP])(?::([a-z][a-z0-9_-]*))?\]\s+(.+)$/;
 function parseCriteria(input, twins) {
     const multiTwin = twins.length > 1;
     const allowed = new Set(twins);
@@ -203,10 +221,15 @@ function parseCriteria(input, twins) {
     const criteria = [];
     for (const rawLine of input.split("\n")) {
         const line = rawLine.trim();
+        const legacy = line.match(LEGACY_CRITERION_LINE_RE);
+        if (legacy) {
+            const legacyMarker = `[${legacy[1]}${legacy[2] ? `:${legacy[2]}` : ""}]`;
+            throw new Error(`Criterion "${legacyMarker} ${legacy[3].trim()}" uses a retired marker — markers were renamed [D]→[code] and [P]→[model] (tags carry over: [D:github]→[code:github]).`);
+        }
         const match = line.match(CRITERION_LINE_RE);
         if (!match)
             continue;
-        const kind = match[1]; // "D" | "P"
+        const kind = match[1]; // "code" | "model"
         const tag = match[2]; // twin tag or undefined
         const text = match[3].trim();
         // Reconstruct the human-facing marker for error messages.
@@ -215,22 +238,23 @@ function parseCriteria(input, twins) {
             if (!multiTwin) {
                 // Single-twin: an explicit tag is allowed but must equal the sole twin.
                 if (tag !== primary) {
-                    throw new Error(`Criterion "${marker} ${text}" tags twin "${tag}", but this single-twin scenario runs "${primary}". Drop the tag or set config.twins to include "${tag}".`);
+                    throw new Error(`Criterion "${marker} ${text}" tags twin "${tag}", but this single-twin task runs "${primary}". Drop the tag or set config.twins to include "${tag}".`);
                 }
             }
             else if (!allowed.has(tag)) {
                 // Multi-twin: an explicit tag must name one of the scenario's twins.
-                throw new Error(`Criterion "${marker} ${text}" tags twin "${tag}", which is not in the scenario's twins [${twins.join(", ")}].`);
+                throw new Error(`Criterion "${marker} ${text}" tags twin "${tag}", which is not in the task's twins [${twins.join(", ")}].`);
             }
         }
-        else if (multiTwin && kind === "D") {
-            // Multi-twin: every deterministic [D] criterion MUST carry a tag so the
-            // cloud knows which twin's state to check it against. [P] may stay bare
+        else if (multiTwin && kind === "code") {
+            // Multi-twin: every [code] criterion MUST carry a tag so the cloud knows
+            // which twin's state to check it against. [model] may stay bare
             // (attributes to the primary twin).
-            throw new Error(`Criterion "${marker} ${text}" needs a twin tag ([D:<twin>]) in a multi-twin scenario (twins [${twins.join(", ")}]).`);
+            throw new Error(`Criterion "${marker} ${text}" needs a twin tag ([code:<twin>]) in a multi-twin task (twins [${twins.join(", ")}]).`);
         }
-        // Authors still write `[D]`/`[P]` in markdown; the published contract's
-        // tolerant reader normalizes those to the canonical `code`/`model` kinds.
+        // The marker spells the canonical kind; `criterionSchema` keeps accepting
+        // the legacy `D`/`P` enum values only for 0.3.0-era persisted artifacts
+        // (the published contract's tolerant reader), never from markdown.
         // The optional `twin` rides through untouched.
         criteria.push(criterionSchema.parse(tag !== undefined ? { type: kind, text, twin: tag } : { type: kind, text }));
     }
@@ -257,6 +281,10 @@ function parseSeedStateForScenario(input, config) {
         return stripeSeedStateSchema.parse(input);
     if (isSlackOnly(config.twins))
         return slackSeedStateSchema.parse(input);
+    if (isGmailOnly(config.twins))
+        return gmailSeedSchema.parse(input);
+    if (isLinearOnly(config.twins))
+        return linearSeedStateSchema.parse(input);
     return parseGitHubSeedState(input);
 }
 function defaultSeedStateForConfig(twins) {
@@ -271,13 +299,25 @@ function defaultSeedStateForConfig(twins) {
         // schema-valid floor.
         return slackSeedStateSchema.parse({});
     }
+    if (isGmailOnly(twins)) {
+        return gmailSeedSchema.parse(defaultGmailSeedState());
+    }
+    if (isLinearOnly(twins)) {
+        return linearSeedStateSchema.parse(defaultLinearSeedState());
+    }
     return seedSchema.parse(defaultSeedState());
 }
 function isStripeOnly(twins) {
-    return twins.includes("stripe") && !twins.includes("github");
+    return twins.length === 1 && twins[0] === "stripe";
 }
 function isSlackOnly(twins) {
-    return twins.includes("slack") && !twins.includes("github") && !twins.includes("stripe");
+    return twins.length === 1 && twins[0] === "slack";
+}
+function isGmailOnly(twins) {
+    return twins.length === 1 && twins[0] === "gmail";
+}
+function isLinearOnly(twins) {
+    return twins.length === 1 && twins[0] === "linear";
 }
 function stripFence(input) {
     const fence = input.match(/```(?:json|yaml)?\s*([\s\S]*?)```/i);
@@ -291,10 +331,10 @@ function slugFromPath(path) {
 }
 export function formatScenarioError(error, filePath) {
     if (error instanceof z.ZodError) {
-        return `Invalid scenario ${filePath}: ${error.issues.map((issue) => `${issue.path.join(".")}: ${issue.message}`).join("; ")}`;
+        return `Invalid task ${filePath}: ${error.issues.map((issue) => `${issue.path.join(".")}: ${issue.message}`).join("; ")}`;
     }
     if (error instanceof Error) {
-        return `Invalid scenario ${filePath}: ${error.message}`;
+        return `Invalid task ${filePath}: ${error.message}`;
     }
-    return `Invalid scenario ${filePath}`;
+    return `Invalid task ${filePath}`;
 }