Getting Started
Test Agent
Test Agent is responsible for test case generation and management, supports declarative workflow engine driven, multi-project independent process configuration.
v3.0.7 (Latest)
Release Date: 2026-08-21 | Base Version: v3.0.6
1. Overview
| Item | Content |
|---|---|
| Release Date | 2026-08-21 |
| Version | v3.0.7 |
| Base Version | v3.0.6 |
2. Release Notes
2.1 New Features
| Module | Details |
|---|---|
| Test Case Generation Skill | Added automatic filtering of document structural sections: in single-document / dual-document modes, non-requirement content such as change history, cover title blocks, abbreviations and glossaries, and format template descriptions is automatically excluded, so invalid test cases are no longer generated for such sections; each exclusion records a reason for audit, and when over-filtering occurs the system automatically falls back to full coverage to avoid false positives Added pre-run cost estimation prompt: before dispatch, users are informed of the processing scope, total number of requirement items, concurrency, and estimated duration, allowing them to assess task scale in advance and decide whether to adjust the scope Added configurable knowledge base recall: recall service address, retrieval corpus, number of recalls and other parameters are all externalized to configuration files, supporting two-level override by project (Geely / DNKT) and workspace; switching knowledge base environments no longer requires modifying the flow Added distinction between "Not Run" and "Broken Link" states in traceability reports: the traceability matrix and summary report clearly mark which requirements have not yet been processed and which genuinely lack links after processing, and provide run/not-run counts along with an overall run status conclusion, so users can tell at a glance whether the task did not finish or finished with issues |
2.2 Changes
| Module | Details |
|---|---|
| Test Case Generation Skill | Processing scope strategy adjusted to "full by default": when the user does not explicitly specify a scope, all requirement items are processed, instead of being automatically trimmed to the first few items; when the user specifies a requirement range, section, or functional module in the input, execution follows the specified scope, and the basis for determining the processing scope is written into the results for traceability Knowledge base recall changed from in-model inline execution to scripted deterministic execution, improving stability and reproducibility of recall results; the applicable scope of recall is also clarified (item-by-item recall is performed only for plain-text requirement scenarios, and document-type inputs no longer trigger repeated recall), reducing invalid calls and time consumption Requirement analysis entry logic streamlined: "problem ticket" and "structured change points" are no longer treated as independent branches and are unified into the plain-text requirement processing path; capabilities remain unchanged but with fewer flow branches and more consistent behavior Added test point missing interception: when a requirement item produces no test points, test cases are no longer forcibly generated; instead an empty result is produced with a reason marked, preventing non-traceable test cases from polluting the traceability matrix Built-in fallback test case format completed with "Requirement ID", "Requirement Name", "Test Point" and other columns, delivering a more complete deliverable column structure in default scenarios |
2.3 Removed
| Module | Details |
|---|---|
| None | None |
2.4 Bug Fixes
| Module | Details |
|---|---|
| Test Case Generation Skill | Fixed issue where large single-document files only processed some requirement items: previously large specification documents were automatically trimmed, causing test cases not to be generated for many requirements; now processed in full with the scale clearly communicated Fixed issue where "Requirement ID" and "Requirement Name" appeared blank in the deliverable Excel: these two columns are now forcibly backfilled from requirement items with non-null validation, so test cases can be fully traced back to their source requirements; sections in the document that genuinely have no requirement ID are correctly distinguished and no longer falsely reported as failures Fixed issue where names were empty when requirement item titles were missing: when the title is missing, a fallback is applied in the order of section number → file name → item number, so test case files and Excel names no longer appear empty |
v3.0.6
Release Date: 2026-07-24 | Base Version: v3.0.5
1. Overview
| Item | Content |
|---|---|
| Release Date | 2026-07-24 |
| Version | v3.0.6 |
| Base Version | v3.0.5 |
2. Release Notes
2.1 New Features
| Module | Details |
|---|---|
| Test Case Generation Skill | Added "problem ticket" input format: when user input is a three-segment problem ticket of "phenomenon + Root cause + Solution", directly derives change points from the problem ticket using the four-step method, without relying on SRD Added Excel single document input format (execute_excel_single), improved DNKT Excel requirement parsing and normalization (openpyxl parsing, Excel requirement normalization) per-item pipeline added test case Excel export sub-step (M3-Excel), test cases for single change point can be directly exported to Excel |
2.2 Changes
| Module | Details |
|---|---|
| Test Case Generation Skill | Change point description filtering enhanced: in single_doc mode, semantic pre-filtering of split items based on user's "change point description" (raw_input), only passing hit requirement items to pipeline, avoiding generating large amounts of irrelevant test cases based on entire specification book, test case focus and accuracy improved Generalized skill usage issue fixes and polishing: M0 requirement analysis, M7 test case execution, M8 test case filtering and other step descriptions and configuration optimized; added DNKT / Geely project execution configuration templates (execution_config.json), execution platform access configuration clearer Built-in fallback generation rule (builtin_fallback) improved, test case generation more robust in default scenarios |
2.3 Removed
| Module | Details |
|---|---|
| None | None |
2.4 Bug Fixes
| Module | Details |
|---|---|
| Test Case Generation Skill (Geely) | Fixed Geely project feedback issues: optimized M0 change point extraction and document diff comparison logic, Excel requirement parsing stability improved Fixed issue where all items in single_doc mode generate in full, with large amounts of test cases unrelated to change points |
v3.0.5
Release Date: 2026-07-17 | Base Version: v3.0.4
1. Overview
| Item | Content |
|---|---|
| Release Date | 2026-07-17 |
| Version | v3.0.5 |
| Base Version | v3.0.4 |
2. Release Notes
2.1 New Features
| Module | Details |
|---|---|
| Test Case Generation Skill | Added parallel pipeline capability: workflow engine added parallel_for_each / aggregate step types and concurrency control (semaphore max_in_flight), supports parallel generation and result aggregation for multiple change points / requirement modules, significantly improving generation throughput and overall time performanceDNKT project added Excel input format support, can directly use Excel requirements as input to generate test cases, and added test case Excel export capability |
2.2 Changes
| Module | Details |
|---|---|
| Test Case Generation Skill | Process overall generalization (de-projectization): originally separated multiple independent processes by project (jili / dnkt) merged into unified single workflow.json and common step directory, adding new projects or input formats without copying entire process, maintenance cost significantly reduced Four business input formats (plain text requirements / single requirement document / new-old dual-version document comparison / new requirements) unified as common execution template in each step, single input change point processing logic generalized Process acceleration and step streamlining: M0 automatically skips M1 when test perspective is detailed enough; M1, M3 steps merged to reduce steps and directory count; review feedback loop (loop) optimized, rework path more efficient |
2.3 Removed
| Module | Details |
|---|---|
| None | None |
2.4 Bug Fixes
| Module | Details |
|---|---|
| Test Case Generation Skill | Fixed DNKT / Geely project Excel generation related defects, Excel export result stability improved |
v3.0.4
Release Date: 2026-07-07 | Base Version: v3.0.3
1. Overview
| Item | Content |
|---|---|
| Release Date | 2026-07-07 |
| Version | v3.0.4 |
| Base Version | v3.0.3 |
2. Release Notes
2.1 New Features
| Module | Details |
|---|---|
| Test Case Generation Skill | Added support for three business input formats (plain text requirements / single requirement document / new-old dual-version document comparison), users can flexibly choose input method according to actual work scenarios Added document diff analysis capability, supports rule-based comparison of various content formats including text, tables, images, and sequence diagrams, providing underlying capability support for requirement change scenarios DNKT project added remote automatic test case execution capability, users can directly trigger execution tasks after generating test cases, reducing manual handling Geely project added automatic requirement change diff extraction and requirement module structured splitting capability, providing support for test case generation in dual-document comparison scenarios |
2.2 Changes
| Module | Details |
|---|---|
| Test Case Generation Skill (DNKT) | Overall upgrade of test case generation process to new version, redundant steps merged and streamlined, fewer operation steps and clearer path from requirement input to test case output Requirement analysis capability enhanced, supports directly identifying change scope from requirement change blocks, reducing manual sorting cost Test case execution interface call stability improved, remote interaction unified, predictability in exception scenarios enhanced |
| Test Case Generation Skill (Geely) | Connected three business input scenarios (plain text requirements / single requirement document / new-old dual-version document comparison), covering actual work scenarios such as new requirement review, requirement change, and daily addition Requirement change scenarios can automatically identify new-old document diffs and locate affected modules, directly driving test case addition and modification Review feedback loop optimized: review opinions to test case modification formed as separate step, rework path clearer, test case quality and first-pass rate improved Knowledge base recall strategy upgraded to hierarchical recall, hit relevance and test case coverage improved synchronously Test case generation accuracy significantly improved in single text input scenario, M4 to M10 outputs completed throughout the process, delivery completeness improved |
2.3 Removed
| Module | Details |
|---|---|
| Test Case Generation Skill | Decommissioned all support for ndsd project, related processes and supporting resources no longer maintained DNKT project old version test case generation process fully decommissioned, replaced by the merged and optimized new process in this version Geely project decommissioned historical steps no longer used such as review analysis, modification confirmation, test case execution and test case filtering, process more focused on current actual business chain |
2.4 Bug Fixes
| Module | Details |
|---|---|
| None | None |
v3.0.3
Release Date: 2026-06-12 | Base Version: v3.0.2
1. Overview
| Item | Content |
|---|---|
| Release Date | 2026-06-12 |
| Version | v3.0.3 |
| Base Version | v3.0.2 |
2. Release Notes
2.1 New Features
| Module | Details |
|---|---|
| None | None |
2.2 Changes
| Module | Details |
|---|---|
| Test Case Generation Skill | Refactored workflow to declarative engine Added config/workflow.schema.json and project-independent assets/{project}/workflow.json, main agent no longer hardcodes business processes, changed to reading workflow.json to drive executionSynchronously rewrote SKILL.md and references/STRUCTURE.md, new projects do not need to modify main documentation Enforced fork sub-agent execution: engine only responsible for scheduling, all steps must execute via forked sub-agents, inline business logic prohibited Unified I/O and artifact format: re-aligned input/output formats for all steps, removed *_history.json saving logic, settings.json updated synchronouslyProcess optimization under each project, generation accuracy improved |
2.3 Removed
| Module | Details |
|---|---|
| None | None |
2.4 Bug Fixes
| Module | Details |
|---|---|
| None | None |
v3.0.2
Release Date: 2026-06-05 | Base Version: v3.0.1
1. Overview
| Item | Content |
|---|---|
| Release Date | 2026-06-05 |
| Version | v3.0.2 |
| Base Version | v3.0.1 |
2. Release Notes
2.1 New Features
| Module | Details |
|---|---|
| Test Case Generation Skill | Added test case generation logic according to different project rules |
2.2 Changes
| Module | Details |
|---|---|
| Test Case Generation Skill | Optimized execution step processing operations, added review rules |
2.3 Removed
| Module | Details |
|---|---|
| None | None |
2.4 Bug Fixes
| Module | Details |
|---|---|
| None | None |
v3.0.1
Release Date: 2026-05-22 | Base Version: v3.0.0
1. Overview
| Item | Content |
|---|---|
| Release Date | 2026-05-22 |
| Version | v3.0.1 |
| Base Version | v3.0.0 |
2. Release Notes
2.1 New Features
| Module | Details |
|---|---|
| None | None |
2.2 Changes
| Module | Details |
|---|---|
| Test Case Generation Skill | Added test perspective generation step, optimized modifications, added steps, added test case execution steps |
2.3 Removed
| Module | Details |
|---|---|
| None | None |
2.4 Bug Fixes
| Module | Details |
|---|---|
| None | None |