humanish 0.123.0 → 0.124.0

package/dist/run/run.js

@@ -5,6 +5,7 @@
 import { randomBytes } from "node:crypto";
 import { redactText } from "../evidence/redaction.js";
 import { buildObserverData } from "../observer/data.js";
+import { splitObserverData } from "../observer/participant-items.js";
 import { renderObserver } from "../observer/render.js";
 import { RUN_BUNDLE_FILE, PUBLIC_TARGET_CWD, buildRunSource, } from "./bundle.js";
 import { judgeExecution, } from "./judge.js";
@@ -187,7 +188,7 @@ function runPublisher(args) {
         await writeContainedOutputFile(paths, "review.md", options.renderReview(publicBundle, status), "utf8");
         const events = publicBundle.events.map((event) => JSON.stringify(event)).join("\n");
         await writeContainedOutputFile(paths, "events.ndjson", `${events}\n`, "utf8");
-        await writeContainedOutputFile(paths, "observer/observer-data.json", json(buildObserverData(publicBundle, undefined, status)), "utf8");
+        await writeContainedOutputFile(paths, "observer/observer-data.json", json(splitObserverData(buildObserverData(publicBundle, undefined, status)).shell), "utf8");
     };
     const snapshot = async (bundle) => {
         await writeBundleFiles(bundle, undefined, async () => { });

package/dist/study/plan.js

@@ -15,6 +15,7 @@ import { routeOf } from "./routing.js";
 import { automaticAnalysisRouteReason, taskProtocolValidationReason } from "./validation.js";
 import { declaredParticipantCount } from "./study-fields.js";
 import { recordedStudyWarnings } from "./warnings.js";
+import { plural } from "../run/text.js";
 export { routeOf } from "./routing.js";
 /** Resolve dry-run: explicit override wins, else the scenario mode, else the given fallback. */
 export function resolveStudyDryRun(config, override, fallback) {
@@ -85,10 +86,12 @@ export function planStudy(config, options, deps = {}) {
     const bundle = plannedBundle(result.plan);
     // The manifest the CLI resolved enters the plan here and nowhere else; the routes read
     // plan.study for the run's status record and bundle, and plan.warnings for its bundle events.
+    const closing = closingReportWarning(result.plan);
     const warnings = [
         ...recordedStudyWarnings(study),
         ...(result.plan.warnings ?? []),
         ...(bundle.warning === undefined ? [] : [bundle.warning]),
+        ...(closing === undefined ? [] : [closing]),
     ];
     const plan = {
         ...result.plan,
@@ -111,6 +114,23 @@ function plannedBundle(plan) {
     const warning = plan.dryRun ? undefined : bundleLimitReason(estimate);
     return warning === undefined ? { estimate } : { estimate, warning };
 }
+/**
+ * The warning for participants with a scheduled end on Claude as a local-agent participant model.
+ * Its session has no read-only closing request, so each such participant ends with the scheduled
+ * end as its reason and no account of its session.
+ */
+function closingReportWarning(plan) {
+    if (plan.route !== "computer-use")
+        return undefined;
+    const { brain } = plan.runner;
+    if (brain.kind !== "local-agent" || brain.agent !== "claude")
+        return undefined;
+    const scheduled = plan.runner.participants.filter((participant) => participant.endAfterMs !== undefined).length;
+    if (scheduled === 0)
+        return undefined;
+    const subject = scheduled === 1 ? "it ends" : "each ends";
+    return `${plural(scheduled, "participant")} ${scheduled === 1 ? "has" : "have"} a scheduled end (participants[].endAfterMs), and Claude as a local-agent participant model cannot be asked for a closing report, so ${subject} with the scheduled end as its reason and no closing account. A computer-use model (actor.type: openai-computer-use) or Codex (actor.localAgent: codex) writes one.`;
+}
 function planRoute(route, study, options, input, deps, e2b) {
     const { sandboxCeiling } = e2b;
     switch (route) {

package/docs/architecture/observer.md

@@ -24,9 +24,26 @@ Observer commands may add derived artifacts:
     observer-data.json
 ```
 
-`run.json` remains the source bundle. `observer/observer-data.json` is the
-normalized view model consumed by the Observer. `events.ndjson` is the appendable
-event stream contract that live adapters update while a run is active.
+`run.json` remains the source bundle. `observer/observer-data.json`
+(`humanish.observer-data.v2`) is the normalized view model consumed by the Observer,
+with every participant except their trace items and the screenshot entries of their
+artifact list. A participant working for hours records thousands of items and a
+screenshot each turn, so these are most of a long run's bytes; apart, the saved file
+stays small enough for `verify` to read. The items stay in the run's traces
+(`actors/<stream>.json`). `events.ndjson` is the appendable event stream
+contract that live adapters update while a run is active.
+
+`observer/index.html` carries both parts: the data slot holds observer-data.json's
+content, and `<script id="observer-items" type="application/octet-stream"
+data-encoding="gzip">` holds the items and full artifact lists (`humanish.observer-items.v1`, by stream id)
+gzip-compressed in base64, which a `file://` page decodes without a fetch. Real
+trace items compress about eight times. The page joins the two before it renders
+(`joinObserverData` in `src/observer/participant-items.ts`), so every view reads the
+same data as before. A stream whose screenshot entries moved records how many in
+`screenshotArtifacts`, and the page renders nothing from items that are another
+run's or miss a participant's items or screenshot entries. The server answers a
+saved v2 observer-data.json that leaves items out as unavailable, since it has no
+items to join.
 
 ## Stream Model
 
@@ -55,7 +72,9 @@ public-safe metadata.
 4. opens the served Observer URL;
 5. keeps the shell attached until Ctrl-C.
 
-The browser polls `observer-data.json` with `no-store` caching. Static
+The browser polls `observer-feed.json` with `no-store` caching: the server's
+projection of the run as `{ data, items }`, observer-data.json's content and the
+items, from one read of the run per poll. Static
 `file://` opening still works for immutable review, but follow mode is the
 operator path. Agents and CI should use `humanish watch --json --no-open` for
 the same fresh evidence without browser open or a long-running process.
@@ -253,7 +272,7 @@ These remain adapter work:
 ## The rendering layer
 
 The renderer is the `observer/` workspace: a Vite single-file build on the
-`@humanish` registry tokens, frozen against `humanish.observer-data.v1`
+`@humanish` registry tokens, frozen against `humanish.observer-data.v2`
 (`tests/observer/data-contract.test.ts`). The root build copies the workspace
 artifact to `dist/observer-app.html`; in a repo checkout a missing or stale
 artifact auto-builds, and an unconditional preflight at CLI startup makes a

package/docs/contracts/run-bundle.md

@@ -77,7 +77,9 @@ used as a bundle answers 404 with the reason, never the saved projection. In saf
 (`observe --all --safe`) every read is checked against the files admitted with the run's
 share_ready verification, so a run.json changed after admission is refused and the admitted
 projection is served.
-observer-data.json is built from the loaded bundle and has the same shape as before.
+observer-data.json is built from the loaded bundle. From 0.124.0 it is
+`humanish.observer-data.v2`: the same data without each participant's trace items, which stay in
+the traces (docs/architecture/observer.md).
 
 Compatibility: humanish 0.122.0 and earlier read only v1 and refuse a v2 run.json by its schema
 (`run schema` fails in verify, analysis refuses the source). This release reads v1 bundles as

package/docs/contracts/study-analysis.md

@@ -367,7 +367,7 @@ and an impression alone supports only a finding about what the participant said.
 
 ## Durable records
 
-The frozen `humanish.observer-data.v1` schema is unchanged. A companion
+The `humanish.observer-data.v2` schema is unchanged by analysis. A companion
 `humanish.study-analysis.v1` artifact records source/config/input hashes,
 participant context, evidence manifest, coverage, provider/model/prompt version,
 status, usage and validated findings. The Observer's projection of it,

package/docs/decisions/0002-observer-is-one-self-contained-file.md

@@ -13,7 +13,8 @@ exactly those cases.
 The `observer/` workspace builds with Vite into a single HTML file with fonts inlined and no
 network references. The root build copies it to `dist/observer-app.html`. `renderObserverHtml`
 is the one path every surface (observe, observe --all, watch, studies) uses; it injects the run's
-`humanish.observer-data.v1` snapshot into that file. There is no second renderer and no feature
+`humanish.observer-data.v2` snapshot into that file, with the participants' trace items beside it
+gzip-compressed (docs/architecture/observer.md). There is no second renderer and no feature
 flag; rolling back means pinning an older package version.
 
 ## Consequences

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "humanish",
-  "version": "0.123.0",
+  "version": "0.124.0",
   "description": "Synthetic user research for apps, CLIs, and agent-facing product flows.",
   "keywords": [
     "cli",
@@ -94,7 +94,7 @@
     "comms:agentmail:contract": "node --import tsx scripts/agentmail-contract-check.mjs",
     "format": "oxfmt",
     "format:check": "oxfmt --check",
-    "lint": "node scripts/check-lint-cap.mjs --max-warnings=426",
+    "lint": "node scripts/check-lint-cap.mjs --max-warnings=424",
     "knip": "knip",
     "prose:check": "node scripts/check-code-prose.mjs",
     "vocabulary:check": "tsx scripts/check-retired-words.ts",