@tasksai/install 0.1.40 → 0.1.41

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tasksai/install",
-  "version": "0.1.40",
+  "version": "0.1.41",
   "description": "Shared TasksAI MCP installer CLI",
   "type": "module",
   "bin": {

package/runtime/workflows/catalog_generation.py

@@ -104,6 +104,9 @@ def _generate(workspace,job_id,sequence,revision,request_id,user_request,documen
                 'If a gate requires refusal or handoff, return that status with the exact required response, empty content_markdown, tables and source_refs, and no work product. '
                 'If the released skill requires information collection, a source-refresh stop, or another prerequisite before any file, return needs_input with its required response and only the missing questions in response and missing; leave content_markdown, tables and source_refs empty. Never turn an intake-only or source-blocked response into a document. If a required tool or source guide is unavailable in this environment, state that limitation accurately; do not pretend it was called or ask the user to operate an internal tool. '
                 'For allowed work, follow the entire output structure. Use draft status when information is missing; list concise missing questions. '
+                'This application, not the drafting model, creates and saves the Word document after accepting a ready result. A missing model-side save tool is not a missing document source or a drafting blocker. Do not write that no file was saved, that saving failed, or that the customer must call a save tool; leave file-save confirmation to the application response. Keep signature, delivery, publication and professional approval pending unless evidenced. '
+                'Describe missing authority as an evidence limitation (for example, reviewed jurisdiction coverage has not been established), not unavailable tools or registry mechanics. Never imply an unperformed lookup occurred. When the released skill permits provisional professional sources, preserve that status and all source limitations. '
+                'Consolidate each unresolved action once in the designated next-actions list, using missing-question markers there. Other required sections may record the underlying unknown fact or refer to that action briefly; do not repeat its question, explanation or action text across flags, tables and checklists. Do not duplicate an item label in its own description. '
                 'Honor explicit user length and formatting constraints. Unless the user excludes them, a document word maximum includes headings, tables, source-register entries and appended missing questions. Budget space for every required section; shorten repeated explanations and group source entries without losing provenance. Do not claim compliance with a word limit merely because the narrative alone fits. '
                 'For a short page target, budget the complete packet, including table wrapping and blank evidence fields. Use compact labels, concise cells and one occurrence of each explanation; put required source details in one concise register rather than repeating them throughout. Preserve every required section, unresolved issue and evidence distinction. Do not claim a page count before the document has been rendered. '
                 'For a standard-size packet with several tables and a one- or two-page target, use roughly 250–300 words per requested page as a conservative drafting budget across all rendered text, including headings, table cells, inserted missing questions and the source register. This is a planning estimate, not a guaranteed rendered page count or permission to omit required content. Reserve part of that budget for sources before drafting the main sections. Source entries should identify their source and the facts supported with short locators or concise field names, not repeat every fact already in the packet. Do not apply this estimate to large print or an explicit user word-count requirement. '