@mechasm/cli 0.1.3 → 0.1.4

package/AGENT_GUIDE.md

@@ -22,8 +22,21 @@ mechasm runs status RUN_ID --project PROJECT_ID --json
 
 Log streaming alone does not report pass/fail through its exit code. Do not infer success from a closed stream or an individual log line. For scripts that only need the final result, `--wait` on create/edit/run or `runs wait` waits silently and returns the final result without displaying live logs. With `--json`, it returns one final JSON result.
 
+Let Mechasm's healer work while the run is active. Failed actions, unexpected pages and recovery attempts can be corrected automatically. Wait for the overall run to reach `passed` or `failed`; do not cancel because a few minutes passed or one step failed. Stop when the user requests it, or when you deliberately need to change the prompt before continuing. `runs status` includes `terminal`, the current `progress` phase/start time when available, and safe failure details with a suggested next step. Missing progress is not proof the run is stuck.
+
+Before reporting the outcome, review the logs and available evidence against the user's intended result:
+
+```sh
+mechasm runs trace RUN_ID --project PROJECT_ID --out run-trace.zip
+mechasm runs video RUN_ID --project PROJECT_ID --out run.webm
+```
+
+Check artifact availability in `runs status`; uploads can finish after the run result. Inspect trace screenshots/page snapshots, actions and errors, using video as supporting evidence. Verify that the expected behavior really happened; do not treat an existing file as proof of success. If evidence is unavailable or cannot be opened, report that limit.
+
+If evidence shows a genuine failure of the user's requirement, accept and report the failure. If your prompt introduced an unsupported assumption or was ambiguous, clarify it without weakening the user's requirement, then use `tests edit --prompt-file FILE` and inspect the new run. Ask the user if their intended result is unclear. Do not repeatedly rerun or remove assertions just to get a pass. A manually stopped run proves cancellation, not what the natural result would have been.
+
 Use explicit project, test, and run IDs. Do not assume a default project. Use `--json` for scripts; stdout contains JSON and errors go to stderr. Log following emits JSON lines. Save run IDs and log cursors in workflow state. Resume logs with `--after CURSOR` after reconnecting.
 
 Exit code `5` from a status/get/wait command or a test command using `--wait` means the run failed or was stopped. Code `2` means invalid arguments, `3` authentication failed, and `4` permission was denied. The optional npm update notice runs only in interactive non-JSON sessions; CI skips it.
 
-See [CLI documentation](https://mechasm.ai/docs/mechasm-cli/getting-started) for full command details.
+See [CLI documentation](https://mechasm.ai/docs/mechasm-cli/getting-started) and [agent workflows](https://mechasm.ai/docs/mechasm-cli/agent-workflows) for full command details.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mechasm/cli",
-  "version": "0.1.3",
+  "version": "0.1.4",
   "description": "Mechasm command line client",
   "homepage": "https://mechasm.ai/docs/mechasm-cli/getting-started",
   "license": "SEE LICENSE IN LICENSE",

package/src/cli.js

@@ -181,7 +181,10 @@ function output(value, json) {
     console.log(`Signed in${identity ? ` as ${identity}` : ''}${value.deviceName ? ` (${value.deviceName})` : ''}.`);
   }
   else if (Array.isArray(value?.projects)) {
-    if (!value.projects.length) console.log('No projects available.');
+    if (!value.projects.length) {
+      console.log(value.message || 'No projects available to this device.');
+      if (value.suggestedNextStep) console.log(value.suggestedNextStep);
+    }
     else for (const project of value.projects) console.log(`${project.id}  ${project.name}  ${project.url}`);
   } else if (Array.isArray(value?.tests)) {
     if (!value.tests.length) console.log('No tests found.');