@arjunkhera/atlas 0.3.36 → 0.3.38

package/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "atlas",
-  "version": "0.3.36",
+  "version": "0.3.38",
   "description": "Atlas: the delivery lifecycle, its crews and the Atlas tools, for any repository.",
   "author": {
     "name": "Arjun Khera"

package/door/kit-releases.json

@@ -32,6 +32,14 @@
     {
       "kit_sha256": "c4a506ec261d2d2fd7884943c9d317fb7b522704f4f63669b5bd37d32326bde5",
       "first_package": "0.3.32"
+    },
+    {
+      "kit_sha256": "b5c36f685ba476e09a54e3207e2444b098cc6a26dded8dca65b01e751fe2cbe4",
+      "first_package": "0.3.34"
+    },
+    {
+      "kit_sha256": "c00e87170dd5f1b477d144eca976db76580d94ede6104d76609aac040851d488",
+      "first_package": "0.3.37"
     }
   ]
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@arjunkhera/atlas",
-  "version": "0.3.36",
+  "version": "0.3.38",
   "description": "Atlas: the delivery lifecycle, its crews and the Atlas tools, as a Claude Code plugin for any repository.",
   "type": "module",
   "license": "UNLICENSED",

package/tests/contract.mjs

@@ -117,6 +117,12 @@ function load(path, options) {
 }
 
 const listOf = (x) => (Array.isArray(x) ? x.map(String) : x === undefined ? [] : [String(x)]);
+const oneLine = (t) => String(t ?? '').replace(/\s+/g, ' ').trim();
+// The reason of each run of a repeat that did not pass, for the log: "; run 2: ..., run 3: ...".
+const whyNot = (list) => {
+  const lines = list.filter((r) => r.result !== 'pass' && r.result !== 'not here').map((r) => `run ${r.run}: ${oneLine(r.error ?? r.why ?? r.stopped ?? r.result).slice(0, 240)}`);
+  return lines.length ? `; ${lines.join('; ')}` : '';
+};
 
 export function scenario(path, body, options = {}) {
   const ctx = load(path, options);
@@ -238,7 +244,7 @@ export function scenario(path, body, options = {}) {
             a.result = blocked ? 'blocked' : ok ? 'pass' : unsure && passed + unsure >= repeat.pass ? 'unsure' : 'fail';
             a.proof = {
               repeat: { runs: repeat.runs, pass: repeat.pass, passed }, runs: list,
-              ...(blocked ? { why: `the repeat stopped in run ${list.length}: ${state.redactor.text(error.message)}` } : ok ? {} : { error: `passed in ${passed} of ${list.length} runs; the rule is ${repeat.pass} of ${repeat.runs}` }),
+              ...(blocked ? { why: `the repeat stopped in run ${list.length}: ${state.redactor.text(error.message)}` } : ok ? {} : { error: `passed in ${passed} of ${list.length} runs; the rule is ${repeat.pass} of ${repeat.runs}${whyNot(list)}` }),
             };
           }
           ev.write();
@@ -454,7 +460,9 @@ function makeRun({ ctx, state, way, env, ev, adapter, round = null }) {
       if (!key) throw new Blocked(`the judge "${names[0]}" needs the key ${keyName}: list it in secrets of tests.yaml, and give its value`);
       const out = await askJudge({ name: names[0], judge, words: a.words, evidence, key, meter: state.meter, redactor: state.redactor });
       open('run.judge');
-      ev.setResult(a.fullId, target, out.result, { judge: names[0], answers: out.answers, cost: out.cost, calls, settled_by: way, ...(out.result === 'pass' ? {} : { error: out.why }) });
+      // A judge that did not pass says why in its own words: the first answer that is not YES.
+      const doubt = out.answers.find((x) => x.answer !== 'YES')?.reply;
+      ev.setResult(a.fullId, target, out.result, { judge: names[0], answers: out.answers, cost: out.cost, calls, settled_by: way, ...(out.result === 'pass' ? {} : { error: doubt ? `${out.why}: ${state.redactor.text(oneLine(doubt)).slice(0, 200)}` : out.why }) });
     },
     // Money that a product's own model call spent (the router, say), read from its answer.
     // It counts against budget.money with the judge and the agent calls.

package/tests/model.mjs

@@ -289,11 +289,14 @@ export async function runClaude({ prompt, args = [], key, cwd, meter, what, reda
     // A call with no cost in its answer counts its whole hold: the meter never guesses low. The hold is at
     // most a quarter of the cap, so one bad call does not end the night.
     meter.add(Number.isFinite(cost) ? cost : left, what, id);
-    if (!envelope) throw new Error(`${what}: the model command gave no answer (exit ${out.exit}): ${(redactor ? redactor.text(out.stderr || out.stdout) : out.stderr || out.stdout).slice(0, 400)}`);
     const clean = (t) => (redactor ? redactor.text(t) : t);
+    // No answer at all: the command could not run (a missing binary, say). That is no product result
+    // either, so it reads blocked, and the log shows what the command printed.
+    if (!envelope) throw new Blocked(`${what}: the model command gave no answer (exit ${out.exit}): ${clean(String(out.stderr || out.stdout).trim()).slice(0, 400)}`);
     // An error answer with no cost never reached the model: a refused sign-in, no credit, a bad model
-    // name. The scenario cannot pass or fail on it, so it reads blocked, with the reason.
-    if (envelope.is_error && !Number.isFinite(cost)) throw new Blocked(`${what}: the model call failed before the model answered: ${clean(String(envelope.result ?? envelope.error ?? (out.stderr || out.stdout))).slice(0, 400)}`);
+    // name. Newer claude commands give a cost of 0 there. The scenario cannot pass or fail on it, so it
+    // reads blocked, with the reason.
+    if (envelope.is_error && !(cost > 0)) throw new Blocked(`${what}: the model call failed before the model answered: ${clean(String(envelope.result ?? envelope.error ?? (out.stderr || out.stdout))).slice(0, 400)}`);
     if (out.signal === 'SIGKILL') throw new Error(`${what}: the model call ran past ${Math.round(timeoutMs / 1000)} seconds and was stopped`);
     return { exit: out.exit, stdout: clean(out.stdout), stderr: clean(out.stderr), envelope, events, cost: Number.isFinite(cost) ? cost : null };
   } finally {