baro-ai 0.133.1 → 0.133.2

package/dist/cli.mjs

@@ -91240,6 +91240,30 @@ function parseContractConflicts(value) {
   }
   return conflicts;
 }
+function parseUnavailableChecks(value) {
+  if (!Array.isArray(value)) return [];
+  const seen = /* @__PURE__ */ new Set();
+  const checks = [];
+  for (const entry of value) {
+    if (typeof entry !== "object" || entry === null) continue;
+    const { criterion, reason, kind } = entry;
+    if (kind !== "check_unavailable") continue;
+    if (typeof criterion !== "string" || !criterion.trim() || seen.has(criterion)) continue;
+    if (typeof reason !== "string" || !reason.trim()) continue;
+    seen.add(criterion);
+    checks.push({ criterion, reason: reason.trim().slice(0, MAX_CONFLICT_REASON_CHARS) });
+    if (checks.length >= MAX_CONTRACT_CONFLICTS) break;
+  }
+  return checks;
+}
+function parsedConflictFields(value) {
+  const conflicts = parseContractConflicts(value);
+  const unavailable = parseUnavailableChecks(value);
+  return {
+    ...conflicts.length > 0 ? { contractConflicts: conflicts } : {},
+    ...unavailable.length > 0 ? { unavailableChecks: unavailable } : {}
+  };
+}
 function applyContractConflicts(evaluation, criteria, enabled = CONTRACT_CONFLICTS_ENABLED) {
   const conflicts = (enabled ? evaluation.contractConflicts ?? [] : []).filter(
     (conflict) => criteria.includes(conflict.criterion)
@@ -91269,9 +91293,10 @@ or
   const remediationRule = options.remediationGroups ? `
 - For "fail", "remediation_groups" is required and must partition "violated_criteria" exactly once. Every group needs a concise "root_cause" and one or more exact failed criterion strings. For "pass" or "inconclusive", omit "remediation_groups".` : "";
   const conflictsRule = options.contractConflicts ? `
-- "contract_conflicts" (optional, default []) lists criteria this story cannot satisfy as written, each as {"criterion":"exact criterion string","kind":"outside_write_surface" or "contradiction","reason":"\u2026"}. Use "outside_write_surface" only when meeting the criterion requires changing files outside the "## Write surface" Baro lists for this story, and name those files. Use "contradiction" only when no change could meet it together with another of this story's criteria, and quote that other criterion.
+- "contract_conflicts" (optional, default []) lists criteria this story cannot satisfy as written, each as {"criterion":"exact criterion string","kind":"outside_write_surface", "contradiction" or "check_unavailable","reason":"\u2026"}. Use "outside_write_surface" only when meeting the criterion requires changing files outside the "## Write surface" Baro lists for this story, and name those files. Use "contradiction" only when no change could meet it together with another of this story's criteria, and quote that other criterion.
 - A conflicted criterion is not this story's to meet: leave it out of "violated_criteria" and decide "verdict" on every other criterion. Baro moves an "outside_write_surface" criterion to a follow-up story that may change those files, and records a "contradiction" for the run-level reviews.
-- Never list a criterion the story could meet within its write surface but did not, or one that merely lacks evidence: those are violations. Without a "## Write surface" section, only the contradiction case applies.` : "";
+- A command whose runtime status reads "could not start" did not run: its tool is not installed in this worktree. That is a fact about the environment, not about the work. When the only thing a criterion lacks is a passing run of such a command, use kind "check_unavailable" and name the command; leave the criterion out of "violated_criteria". Never use it for a command that ran and failed, or when no command is marked "could not start".
+- Never list a criterion the story could meet within its write surface but did not, or one that merely lacks evidence: those are violations. Without a "## Write surface" section, only the contradiction and check_unavailable cases apply.` : "";
   return `You are a strict acceptance-criteria evaluator. You will receive:
 1. A list of acceptance criteria that must ALL be satisfied.
 2. Optionally, the accepted architecture decisions the agent was bound to.
@@ -91648,6 +91673,8 @@ var MAX_FINGERPRINT_PATHS = 5e3;
 var MAX_FINGERPRINT_FILE_BYTES = 64 * 1024 * 1024;
 var MAX_FINGERPRINT_TOTAL_BYTES = 256 * 1024 * 1024;
 var CRITIC_MAX_PROMPT_CHARS = 14e4;
+var COULD_NOT_START_STATUS = "could not start; the command's tool is not available in this worktree";
+var TOOL_UNAVAILABLE_ISSUE = "a verification command could not start: its tool is not available in the worktree";
 var PublishedNoteCollector = class extends BusMember {
   constructor(maxNotesPerStory = 6) {
     super();
@@ -91768,7 +91795,8 @@ var CriticCommandEvidenceCollector = class extends BusMember {
         entry.fingerprint,
         currentFingerprint
       );
-      const status = output === null ? "pending/no output observed" : commandOutputLooksFailed(output) ? "completed; output indicates failure" : "completed; consult captured output for the result";
+      const toolUnavailable = output !== null && outputLooksToolUnavailable(output);
+      const status = output === null ? "pending/no output observed" : toolUnavailable ? COULD_NOT_START_STATUS : commandOutputLooksFailed(output) ? "completed; output indicates failure" : "completed; consult captured output for the result";
       return {
         chronologicalIndex: index,
         command: entry.command,
@@ -91776,7 +91804,8 @@ var CriticCommandEvidenceCollector = class extends BusMember {
           command: entry.command,
           terminal: output !== null,
           freshness: freshness.state,
-          sandboxBlocked: output !== null && outputLooksSandboxBlocked(output)
+          sandboxBlocked: output !== null && outputLooksSandboxBlocked(output),
+          toolUnavailable
         },
         rendered: [
           `### Command ${index + 1}`,
@@ -92278,7 +92307,8 @@ async function prepareCriticEvaluation(criteria, resultText, agentId, source, es
     issues,
     staleCommands: staleCommandsToRerun(commandEvidence, issues),
     repositoryFingerprint: issues.length === 0 ? afterFingerprint.value : null,
-    established
+    established,
+    unavailableCommands: (commandEvidence.commands ?? []).filter((command) => command.toolUnavailable && command.freshness === "fresh").map((command) => command.command)
   };
 }
 function mergeSegmentVerdicts(verdicts) {
@@ -92301,6 +92331,11 @@ function mergeSegmentVerdicts(verdicts) {
       verdicts.flatMap((verdict) => verdict.contractConflicts ?? []).map((conflict) => [conflict.criterion, conflict])
     ).values()
   ];
+  const unavailable = [
+    ...new Map(
+      verdicts.flatMap((verdict) => verdict.unavailableChecks ?? []).map((check) => [check.criterion, check])
+    ).values()
+  ];
   const reasoning = decisive.length > 0 ? decisive.map(({ verdict, pass }) => `[pass ${pass}] ${verdict.reasoning}`).join(" ") : first.reasoning;
   return {
     ...first,
@@ -92308,7 +92343,32 @@ function mergeSegmentVerdicts(verdicts) {
     verdict: decisive.length > 0 ? "fail" : "pass",
     reasoning,
     violatedCriteria: violated,
-    ...conflicts.length > 0 ? { contractConflicts: conflicts } : {}
+    ...conflicts.length > 0 ? { contractConflicts: conflicts } : {},
+    ...unavailable.length > 0 ? { unavailableChecks: unavailable } : {}
+  };
+}
+var EVIDENCE_UNAVAILABLE_MARKER = "[acceptance evidence unavailable]";
+function applyUnavailableChecks(evaluation, criteria, unavailableCommands) {
+  const { unavailableChecks, ...rest } = evaluation;
+  const settled = rest;
+  const claims = (unavailableChecks ?? []).filter(({ criterion }) => criteria.includes(criterion));
+  if (claims.length === 0 || (settled.status ?? "evaluated") !== "evaluated") return settled;
+  const claimed = new Set(claims.map(({ criterion }) => criterion));
+  if (unavailableCommands.length === 0) {
+    return {
+      ...settled,
+      verdict: "fail",
+      violatedCriteria: [.../* @__PURE__ */ new Set([...settled.violatedCriteria, ...claimed])]
+    };
+  }
+  const violated = settled.violatedCriteria.filter((criterion) => !claimed.has(criterion));
+  if (violated.length > 0) return { ...settled, verdict: "fail", violatedCriteria: violated };
+  return {
+    ...settled,
+    status: "inconclusive",
+    verdict: "fail",
+    reasoning: `Critic could not evaluate the candidate: ${claims.length} ${claims.length === 1 ? "criterion needs" : "criteria need"} a command that could not start in this worktree (${unavailableCommands.map((command) => boundText(command, 120)).join("; ")}). ` + settled.reasoning,
+    violatedCriteria: [EVIDENCE_UNAVAILABLE_MARKER]
   };
 }
 function inconclusiveEvidenceVerdict(issues) {
@@ -92317,7 +92377,7 @@ function inconclusiveEvidenceVerdict(issues) {
     status: "inconclusive",
     verdict: "fail",
     reasoning: `Critic could not evaluate the candidate: ${detail}`,
-    violatedCriteria: ["[acceptance evidence unavailable]"]
+    violatedCriteria: [EVIDENCE_UNAVAILABLE_MARKER]
   };
 }
 function buildEvalPrompt(criteria, resultText, commandEvidence = null, repositoryEvidence = null, decisionDocument = null, segment, publishedNotes = [], hostRunsWholeTreeVerification = false, altitudeSection = null, writeSurface = null, established = []) {
@@ -92510,7 +92570,7 @@ function evidenceReadinessIssues(commandEvidence, repositoryEvidence) {
     issues.push("no terminal shell command evidence was observed");
   } else if (commandEvidence.commands && commandEvidence.commands.length > 0) {
     const hasUsableFreshTerminalCommand = commandEvidence.commands.some(
-      (command) => command.freshness === "fresh" && command.terminal && !command.sandboxBlocked
+      (command) => command.freshness === "fresh" && command.terminal && !command.sandboxBlocked && !command.toolUnavailable
     );
     if (!hasUsableFreshTerminalCommand) {
       if (commandEvidence.commands.some(
@@ -92526,8 +92586,13 @@ function evidenceReadinessIssues(commandEvidence, repositoryEvidence) {
       )) {
         issues.push("verification command was blocked by the sandbox");
       }
+      if (commandEvidence.commands.some(
+        (command) => command.toolUnavailable
+      )) {
+        issues.push(TOOL_UNAVAILABLE_ISSUE);
+      }
       if (!issues.some(
-        (issue2) => issue2 === STALE_COMMAND_EVIDENCE || issue2 === "a verification command has no terminal output" || issue2 === "verification command was blocked by the sandbox"
+        (issue2) => issue2 === STALE_COMMAND_EVIDENCE || issue2 === "a verification command has no terminal output" || issue2 === "verification command was blocked by the sandbox" || issue2 === TOOL_UNAVAILABLE_ISSUE
       )) {
         issues.push("command evidence has no usable fresh terminal command");
       }
@@ -92563,11 +92628,13 @@ function parseRenderedCommandEvidence(evidence) {
     const command = /^command:\s*(.+)$/m.exec(metadata)?.[1]?.trim() ?? "";
     const freshnessText = /^freshness:\s*(.+)$/m.exec(metadata)?.[1]?.trim();
     const freshness = freshnessText?.startsWith("fresh") === true ? "fresh" : freshnessText?.startsWith("STALE/UNVERIFIABLE") === true ? "unverifiable" : "stale";
+    const toolUnavailable = runtime === COULD_NOT_START_STATUS || outputLooksToolUnavailable(output);
     return {
       command,
-      terminal: runtime?.startsWith("completed;") === true,
+      terminal: runtime?.startsWith("completed;") === true || toolUnavailable,
       freshness,
-      sandboxBlocked: outputLooksSandboxBlocked(output)
+      sandboxBlocked: outputLooksSandboxBlocked(output),
+      toolUnavailable
     };
   });
 }
@@ -93040,6 +93107,11 @@ function shellCommandSegments(command) {
     (segment) => segment.trim().replace(/^\(+\s*/u, "").replace(/^(?:[A-Za-z_][A-Za-z0-9_]*=[^\s]+\s+)+/u, "")
   );
 }
+function outputLooksToolUnavailable(output) {
+  return /\b(?:exit(?:ed)?(?: with)? (?:code|status)|status=)\s*127\b|(?:^|\n)\s*(?:(?:ba|z|da)?sh|npm error|npx)(?::\s*(?:line\s*)?\d+)?:\s.*(?:command not found|: not found)\b|\bcommand not found: \S+|\bis not recognized as an internal or external command\b|\bspawn \S+ ENOENT\b/i.test(
+    output
+  );
+}
 function outputLooksSandboxBlocked(output) {
   return /sandbox-exec|operation not permitted|permission denied|seatbelt|denied by sandbox/i.test(
     output
@@ -93143,13 +93215,12 @@ var OneShotCritic = class extends BusMember {
   /** Turn raw evaluator text into a verdict; throw to mark inconclusive. */
   parseVerdict(text3) {
     const parsed = JSON.parse(extractVerdictJson(text3.trim()));
-    const conflicts = parseContractConflicts(parsed.contract_conflicts);
     return {
       status: "evaluated",
       verdict: parsed.verdict === "pass" ? "pass" : "fail",
       reasoning: parsed.reasoning ?? "",
       violatedCriteria: Array.isArray(parsed.violated_criteria) ? parsed.violated_criteria : [],
-      ...conflicts.length > 0 ? { contractConflicts: conflicts } : {}
+      ...parsedConflictFields(parsed.contract_conflicts)
     };
   }
   /**
@@ -93196,15 +93267,19 @@ var OneShotCritic = class extends BusMember {
         staleCommands = preparation.staleCommands;
         preparationReady = preparation.status === "ready";
         evaluation = preparationReady ? honorEstablished(
-          applyContractConflicts(
-            mergeSegmentVerdicts(
-              await Promise.all(
-                preparation.prompts.map(
-                  (prompt) => this.evaluate(prompt, agentId, turn)
+          applyUnavailableChecks(
+            applyContractConflicts(
+              mergeSegmentVerdicts(
+                await Promise.all(
+                  preparation.prompts.map(
+                    (prompt) => this.evaluate(prompt, agentId, turn)
+                  )
                 )
-              )
+              ),
+              criteria
             ),
-            criteria
+            criteria,
+            preparation.unavailableCommands
           ),
           preparation.established
         ) : inconclusiveEvidenceVerdict(preparation.issues);
@@ -93420,13 +93495,12 @@ var Critic = class extends OneShotCritic {
       }
       const verdictJson = extractVerdictJson(verdictText);
       const parsed = JSON.parse(verdictJson);
-      const conflicts = parseContractConflicts(parsed.contract_conflicts);
       return {
         status: "evaluated",
         verdict: parsed.verdict === "pass" ? "pass" : "fail",
         reasoning: parsed.reasoning ?? "",
         violatedCriteria: Array.isArray(parsed.violated_criteria) ? parsed.violated_criteria : [],
-        ...conflicts.length > 0 ? { contractConflicts: conflicts } : {},
+        ...parsedConflictFields(parsed.contract_conflicts),
         telemetry
       };
     } catch (err) {
@@ -95763,13 +95837,12 @@ ${prompt}`,
     if (!isCodexVerdict(parsed)) {
       throw new Error("Codex critic returned an invalid verdict object");
     }
-    const conflicts = parseContractConflicts(parsed.contract_conflicts);
     return {
       status: "evaluated",
       verdict: parsed.verdict,
       reasoning: parsed.reasoning,
       violatedCriteria: [...parsed.violated_criteria],
-      ...conflicts.length > 0 ? { contractConflicts: conflicts } : {}
+      ...parsedConflictFields(parsed.contract_conflicts)
     };
   }
 };
@@ -96366,15 +96439,19 @@ var CriticOpenAI = class extends BusMember {
         (fingerprint) => this.ledger.established(agentId, fingerprint)
       );
       const evaluation = preparation.status === "ready" ? honorEstablished(
-        applyContractConflicts(
-          mergeSegmentVerdicts(
-            await Promise.all(
-              preparation.prompts.map(
-                (prompt) => this.evaluate(prompt, agentId, turn, correlation)
+        applyUnavailableChecks(
+          applyContractConflicts(
+            mergeSegmentVerdicts(
+              await Promise.all(
+                preparation.prompts.map(
+                  (prompt) => this.evaluate(prompt, agentId, turn, correlation)
+                )
               )
-            )
+            ),
+            criteria
           ),
-          criteria
+          criteria,
+          preparation.unavailableCommands
         ),
         preparation.established
       ) : inconclusiveEvidenceVerdict(preparation.issues);
@@ -96523,13 +96600,12 @@ var CriticOpenAI = class extends BusMember {
       }
       const verdictJson = extractVerdictJson(assistantText);
       const parsed = JSON.parse(verdictJson);
-      const conflicts = parseContractConflicts(parsed.contract_conflicts);
       return {
         status: "evaluated",
         verdict: parsed.verdict === "pass" ? "pass" : "fail",
         reasoning: parsed.reasoning ?? "",
         violatedCriteria: Array.isArray(parsed.violated_criteria) ? parsed.violated_criteria : [],
-        ...conflicts.length > 0 ? { contractConflicts: conflicts } : {},
+        ...parsedConflictFields(parsed.contract_conflicts),
         telemetry
       };
     } catch (err) {