@dtlabs/ai-toolkit 0.15.1 → 0.16.1

package/CHANGELOG.md

@@ -1,4 +1,25 @@
 ## 0.1.4 (2026-07-21)
+## [0.16.1](https://github.com/dtlab-tech/ai-toolkit/compare/v0.16.0...v0.16.1) (2026-10-02)
+
+
+### Bug Fixes
+
+* **wb-validate:** require groupingRationale only when outputCount > 1 ([e3cb71a](https://github.com/dtlab-tech/ai-toolkit/commit/e3cb71a7f38fbd974c70f3dbdd99223a6aa05048))
+* **wb-validate:** require groupingRationale only when outputCount > 1 ([7406fbe](https://github.com/dtlab-tech/ai-toolkit/commit/7406fbe336304c2ebcd012fc6a6b1b4980121a64))
+
+## [0.16.0](https://github.com/dtlab-tech/ai-toolkit/compare/v0.15.1...v0.16.0) (2026-10-02)
+
+
+### Features
+
+* Add FTR-017 and FTR-018 to the Feature Registry with detailed summaries and keywords ([9be5d95](https://github.com/dtlab-tech/ai-toolkit/commit/9be5d95abf934cca91ebde11ade60bda837ea719))
+
+
+### Bug Fixes
+
+* **workflows:** make structured-output extraction delimiter-agnostic ([0ad511b](https://github.com/dtlab-tech/ai-toolkit/commit/0ad511bfa3892a3c02ff91c9c8256cec5fbdd841))
+* **workflows:** make structured-output extraction delimiter-agnostic ([3498c00](https://github.com/dtlab-tech/ai-toolkit/commit/3498c001725d1b5c52039506f79cacda31b1bc4e))
+
 ## [0.15.1](https://github.com/dtlab-tech/ai-toolkit/compare/v0.15.0...v0.15.1) (2026-10-01)
 
 

package/lib/workflow-control.js

@@ -296,30 +296,32 @@ function createControl(options) {
   return api;
 }
 
-// Tries each known convention a model has been observed using to delimit its final
-// structured answer inside otherwise-free text, in order of explicitness, and returns the
-// first candidate that is both valid JSON and schema-conformant. Returns null (never throws)
-// if no candidate qualifies, so the caller's original "no structured output" error still fires
-// unchanged when the text holds nothing usable.
+// The model is not consistent about which delimiter it uses to mark its final structured
+// answer inside otherwise-free text — observed empirically, same agent and schema, run to
+// run: an <structured_output> tag, a ```json fence, and an unprompted <json> tag. Growing
+// one hardcoded pattern per newly observed convention doesn't converge, so matching is
+// delimiter-agnostic: any <tag>...</tag> pair and any fenced code block are tried as
+// candidates, most-recent-first (the real answer conventionally comes after any earlier
+// in-text example), each independently re-validated as schema-conformant JSON before being
+// accepted — an unmatched or malformed text never silently satisfies the contract. A
+// non-greedy delimiter match is itself unsafe when the schema has a free-text field (e.g. a
+// "report") that legitimately embeds its own nested, identically-delimited example before
+// the real closing delimiter, so brace-balance recovery anchored after the last ```json or
+// <json> marker is tried next — immune to that nesting because it tracks braces, not
+// delimiters — followed by an unanchored whole-text brace span as the last resort.
 function extractDelimitedStructuredOutput(text, schema) {
   const candidates = [];
-  const tagMatch = text.match(/<structured_output>([\s\S]*?)<\/structured_output>/);
-  if (tagMatch) candidates.push(tagMatch[1]);
-  const fenceMatches = [...text.matchAll(/```json\s*([\s\S]*?)```/gi)];
-  if (fenceMatches.length) candidates.push(fenceMatches[fenceMatches.length - 1][1]);
-  // The non-greedy fence match above stops at the FIRST ``` following ```json, which is
-  // wrong when a schema string field (e.g. a "report") legitimately embeds its own
-  // untagged ``` ... ``` example (a markdown table, say) before the real closing fence —
-  // observed empirically with gaia-validate-feature-docs. Recover with a brace-balance
-  // span instead of a fence-delimiter match, anchored after the last ```json marker so it
-  // is immune to backtick nesting inside string values.
-  const lastJsonFence = text.lastIndexOf('```json');
-  if (lastJsonFence !== -1) {
-    const tail = text.slice(lastJsonFence + '```json'.length);
+  for (const m of [...text.matchAll(/<(\w+)>([\s\S]*?)<\/\1>/g)].reverse()) candidates.push(m[2]);
+  for (const m of [...text.matchAll(/```[a-zA-Z]*\s*\n?([\s\S]*?)```/g)].reverse()) candidates.push(m[1]);
+  for (const start of [...text.matchAll(/```json|<json>/gi)].map(m => m.index + m[0].length).reverse()) {
+    const tail = text.slice(start);
     const first = tail.indexOf('{');
     const last = tail.lastIndexOf('}');
     if (first !== -1 && last > first) candidates.push(tail.slice(first, last + 1));
   }
+  const first = text.indexOf('{');
+  const last = text.lastIndexOf('}');
+  if (first !== -1 && last > first) candidates.push(text.slice(first, last + 1));
   for (const raw of candidates) {
     try {
       const parsed = JSON.parse(raw.trim());

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@dtlabs/ai-toolkit",
-  "version": "0.15.1",
+  "version": "0.16.1",
   "description": "AI Toolkit — reusable agents and procedures for developers to build AI-powered applications.",
   "files": [
     "docs/",

package/src/claude/scripts/wb-validate.js

@@ -359,7 +359,7 @@ for (const phase of phases) {
 
     const topLevelFields = [
       'id', 'title', 'outcome', 'domain', 'agentType', 'dependsOn',
-      'acceptanceCriteria', 'outputCount', 'groupingRationale', 'commit',
+      'acceptanceCriteria', 'outputCount', 'commit',
     ]
     for (const field of topLevelFields) {
       if (task[field] == null) {
@@ -374,6 +374,20 @@ for (const phase of phases) {
       }
     }
 
+    // groupingRationale is required only when outputCount > 1 (the generator agent's own
+    // contract: a single-output task has nothing to justify grouping, so it emits null) —
+    // null is the documented, correct value there, not a missing field.
+    if (task.outputCount > 1 && task.groupingRationale == null) {
+      report.errors.push({
+        category: ERRORS.MISSING_FIELD,
+        severity: 'error',
+        taskId,
+        field: 'groupingRationale',
+        message: `Task "${taskId}" has outputCount > 1 but is missing required field "groupingRationale"`,
+        details: { taskId, field: 'groupingRationale' },
+      })
+    }
+
     // Nested: verification.commands — existence and array type only; emptiness checked later
     if (task.verification == null || task.verification.commands == null) {
       report.errors.push({