Getting Started

REQ Agent

REQ Agent is responsible for requirements analysis tasks, transforming requirements into executable specifications.

---

v3.0.4 (Latest)

Release Date: 2026-08-21 | Base Version: v3.0.3

1. Overview

Item	Content
Release Date	2026-08-21
Version	v3.0.4
Base Version	v3.0.3

2. Report Items

2.1 New Features

Module	Details
Requirements analysis skill	Added ReqAgent requirements engineering capability (request-analysis). Takes controlled old/new requirement specifications as input, and automatically produces the full set of deliverables: diff report, requirements analysis, analysis review, SRS (including BDD/SDD-A/verification analysis), final review report, and editable Excel workbook. Trigger words: "Execute ReqAgent", "Requirements Diff", "SWE1 Report"
Requirements analysis skill	Supports paired old/new input in Word/Excel/PDF/Markdown, single specifications with additions only or deletions only, and verification intake for an existing change-point list
Requirements analysis skill	Checkpoint resume and stage jump: can continue from any stage using natural language; before starting a stage, automatically verifies that the prerequisite deliverables for that stage are complete, and explicitly reports missing items instead of generating them out of thin air
Requirements analysis skill	Generation and review isolation: each generation stage is paired with an independent review; the reviewer independently re-verifies the inputs and deliverables without modifying the generation result; issues found trigger bounded rework, and the re-review judges each issue individually as "Resolved/Still Present/Reverted"
Requirements analysis skill	Tightened no-change determination: a no-change result is only returned when a deterministic diff can prove there are no changes, followed by minimal wrap-up; tool execution failures are never written as "no change", and no fabricated SRS, review report, or Excel is produced
Requirements analysis skill	Explicit registration of missing evidence: when evidence is insufficient, it is registered as Q&A, gaps, suspensions, or tentative items, without inventing project facts; when the code root directory is not configured, the item is suspended rather than interpreting "knowledge base miss" as "no impact"
Requirements analysis skill	Automatic parallelization for large change sets: when the number of change points exceeds the concurrency level, they are automatically split into multiple shards that generate the SRS in parallel, then merged by a deterministic script (evidence-chain stitching, requirement deduplication, reference remapping, issue-list merging); default is 4-way concurrency and is adjustable per project; when there are few change points it automatically falls back to serial, producing the same deliverables as serial execution
Requirements analysis skill	Shard-level failure retry: a failed shard only reruns that shard without affecting completed results of other shards; its retry budget is independent of the semantic rework budget
Requirements analysis skill	Excel deliverables are a deterministic projection of the reviewed Markdown/JSON, not a second-round AI-generated conclusion, and can be edited and delivered directly
Requirements analysis skill	Reusable common capabilities consolidated: three modules—document diff (Markdown/PDF/Excel/Word physical comparison), controlled artifact tools (hash, Schema validation, path restriction), and software design knowledge base (init/check/index/query/timeliness)—can be copied wholesale for use by other skills
Requirements analysis skill	Changed to self-contained deployment: the project knowledge bases (requirements knowledge base and HMI software design knowledge base), configuration, scripts, and contracts are distributed with the package; Excel export no longer depends on the node runtime or native binaries and is cross-platform
Requirements document generation skill	Added the ability to reconstruct tables by text coordinates, covering signal definition tables, signal name tables, signal send/receive info tables, DID definition tables, state/action description tables, and two bit-encoding tables, with the reconstruction source annotated in the result for review
Requirements document generation skill	Added incremental debugging: supports rerunning only a specified requirement ID, and appending extraction while preserving existing Markdown and images, so revisions do not require rerunning the entire PDF
Requirements document generation skill	Added structured output of requirement metadata (version, page range, status, table and image counts, continued-table reconstruction status); metadata is separated from the main text, and old-format files remain compatible
Project initiation research document	Added "Customer Pain Points and Business Value Research Report" and "Technical Path and Risk Assessment Report" for three directions: SSR generation, requirements analysis, and requirements generation, covering business positioning, current-state pain points, business value, as well as implementation process breakdown and landing risk assessment

2.2 Changes

Module	Details
Requirements analysis skill	Positioning upgraded from "requirements change analysis" to "requirements engineering workflow": changed from the original 5-stage general change analysis to a fixed 6-stage pipeline; the evidence basis changed from external databases and ticket systems to controlled inputs and on-disk evidence; trigger words and deliverable structure adjusted accordingly
Requirements document generation skill	Redivided chapter content ownership: the public attributes chapter is narrowed to output only static definition tables, while conditional value logic and NVM save strategy are uniformly moved into the requirements chapter, avoiding duplication or contradiction between the two places
Requirements document generation skill	Introduced a two-path extraction method: for the vehicle usage mode and public attributes chapters, first determine whether the source text contains a dedicated requirement; if so, directly crop the source table; if not, induct from the behavioral requirement semantics and explicitly label it as inferred; when the source table is garbled, it can be cross-recovered from behavioral requirements
Requirements document generation skill	No longer reverse-infers missing content: when the source text has no dedicated fault handling requirement, the fault handling and fault recovery sections are directly filled with N/A; columns not mentioned in the usage mode are changed from the default "No" to "N/A"
Requirements document generation skill	Refined chapter recognition rules: configuration items add recognition of the "number + all-uppercase config name" notation; the ports chapter adds naming correspondence and bus field filling rules; the requirements chapter unifies the item starting format and requires cross-references between requirements to be copied verbatim from the source
Requirements document generation skill	Word wide-table layout optimization: auto-adapts font size by column count, calculates column widths by content and fixes the table layout, supports in-cell line wrapping for enumeration values and measurement values, and supports strikethrough rendering
Requirements document generation skill	The vehicle applicability matrix allows at most 8 requirement ID columns per table; excess columns are automatically split vertically into multiple tables
Requirements document generation skill	Execution-time optimization: removed the M1 content health-check step; no longer dispatches health-check sub-tasks per feature in parallel and then aggregates, simplifying to two steps: script execution and checklist validation
Requirements document generation skill	Skill name stripped of the customer-name prefix, unified to srd-generation; trigger words remain unchanged

2.3 Removed

Module	Details
Requirements generation skill	Decommissioned the requirement-generation skill; its stages—intent reception, context retrieval, analysis aggregation, SRS report, and per-requirement pipeline—are removed together, and the skill is no longer available
Requirements analysis skill	Removed dependencies on Neo4j/PostgreSQL database queries and Jira/Redmine ticket tracking; physical diff is now handled by the packaged common capabilities, and the runtime no longer needs to connect to external systems
Requirements document generation skill	Removed the M1 content health-check step and its aggregated output; the output contract removed the health-check report accordingly

2.4 Bug Fixes

Module	Details
Requirements analysis skill	Fixed an output anomaly caused by inconsistency between the stage envelope contract and deliverable assembly
Requirements document generation skill	Fixed an issue where PDF table recognition failure produced misaligned or truncated tables; changed to reconstruct tables by text coordinates
Requirements document generation skill	Fixed an issue where wide tables exceeded the page boundary in the generated Word document
Requirements document generation skill	Fixed an issue where table content was mistakenly swallowed when vector images overlapped with table regions
Requirements document generation skill	Fixed an issue where voided content with strikethrough was extracted as valid content

---

v3.0.3

Release Date: 2026-07-31 | Base Version: v3.0.2

1. Overview

Item	Content
Release Date	2026-07-31
Version	v3.0.3
Base Version	v3.0.2

2. Report Items

This version mainly delivers two workflow features: Functional Safety SSR Specification Generation (ssr-generation) and SRD Generation (SRD-generation).

2.1 New Features

2.1.1 Functional Safety SSR Specification Generation Workflow (ssr-generation)

Takes SDS xlsx + one or more TSC xlsx as input, outputs Functional Safety SSR Specification Excel. Trigger words: functional safety, SSR specification generation.

Module	Details
Workflow Engine	Declarative workflow engine: assets/{project}/workflow.json declares step sequence and artifact dependencies; scripts/orchestrator.py handles workflow parsing, input resolution ($entry.* / $<step>.<key>), fan-out scheduling and artifact validation; the main agent only forks sub-agents and outputs the checklist, without executing business logic
Orchestrator Script	Provides subcommands such as resolve-inputs; supports @<utf8-json-file> file parameter passing and ASCII-safe JSON output, avoiding command-line encoding issues with paths containing Chinese/Japanese characters on Windows
M1 File Parsing (ssr-m1-file-parse)	Parses SDS/TSC xlsx using only the conversation attachment paths injected by the engine (no longer recursively searches directories); produces requirement tree, source manifest, USER_META, TSR info, TSR hierarchy, diagram elements, element descriptions, SDS3.5, filtered manifest and other artifacts
M2-1 Task Split (ssr-m2-1-task-split)	DWG-first split strategy: for each (filter JSON, TSC object_TSR) combination, calls filter_tree.py separately; merging multiple TSCs is prohibited, and the same DWG appearing in multiple TSCs is kept as multiple independent sources; generates tasks grouped by DWG, integrates C source function body extraction (find_c_function / _extract_function_body / _extract_constant) with calib, ram_romsw, and subsystem knowledge base queries
M2-2 Per-task Processing (ssr-m2-2-process-task)	Scheduled in parallel with fan_out mode (max_concurrency=4); each sub-agent handles only 1 DWG task, and must read that DWG's SVG, *_r.json, funcbodies, and TSR2 references itself, then think and generate the SSR tree; batch-generating nodes via scripts is explicitly prohibited
M2-3 Aggregation (ssr-m2-3-aggregate)	Groups by filter_source, outputs one m2_viewpoint_*.json per element/source (containing merged_entries, node file references, funcbodies references and statistical meta); preserves each DWG root SSR and its branches[] subtree; same-name DWGs from different TSCs are not merged
M4 Summary Report (ssr-m4-summary-report)	Workflow endpoint: reads USER_META to get system_name/system_id/model_name, applies the fixed xlsx template to output ./outputs/FS-{system_id}-{model_name}-SSR-0000.xlsx; includes Chapter 4 software requirements generation, TSR-SSR traceability matrix, element insertion and system module replacement
Error Report	M4 additionally outputs a standalone error report FS-*-SSR-0000_error_report.xlsx, tracing node files, function misses and variable knowledge base query status
SSR Generation Rules & Template	Hierarchical generation rules: parent SSR contains only a brief description, and when there are logical paragraphs, the first child SSR must be a logical branch description; refines computational SSR generation rules and clarifies recursion limits; SSR template 4.1.1.XXXXX numbering format; alloc normalized to dwg (real sub-module names are preserved); note field supplements variable descriptions
Element Merge Rules	TSR2 with different elements from TSR1 is marked skip_ssr_generation=true by M1, only retained in the TSR filter tree, not entering the DWG table and M2-2, instead reflected as similar item notes
Knowledge Base Query	When the database hits, case is ignored; only variables obtained from C code have their prefixes stripped when querying the database
Fault Tolerance Fallback	M2-2 scheduling ledger to avoid tasks being marked as running but actually not running; fallback for multiple manifest scenarios; fallback strategy for file timestamp path errors caused by lost tasks
Module Tree Data Source (slwebview-html-extractor)	Companion new skill: reads RequrimrntList.svg, *_h_*.json/*_s.json, *_r.json from the unzipped Simulink WebView/HTML export directory, filters Eagle_Requirement links, only retains the hierarchy after Subsystem, merges and outputs modules_tree.json by source_folder for M2-1 use

2.1.2 SRD Generation Workflow (SRD-generation)

Converts Geely SWRS PDF into structured SRD documents (Markdown + Word .docx). Trigger words: SWRS, SRD, requirement document generation, PDF splitting, chapter restructuring. Parameters: --pdf <PDF path> --function-list <function list Excel path> [--workspace <project directory>].

Module	Details
Workflow Engine	Also a declarative workflow engine: assets/workflow.json declares the M0→M1→M2→M3 four-step sequence and artifact dependencies; the engine only schedules and does not execute business logic; each step is completed by forking an independent sub-agent
M0 PDF Split (srd-m0-pdf-split)	Splits the SWRS PDF into independent Markdown files by requirement ID, with structural processing (paragraph labels to headings, logical block formatting); run_m0.py integrates the split and structuring phases, supports --workers multi-processing; also provides inspect_pdf_pages.py for page troubleshooting
M1 Function-based Extraction (srd-m1-extract)	Reads the function list Excel and M0 output, groups by (requirement document, primary function, secondary function), outputs one JSON per function and a total manifest; optionally reads SWRT Excel for vehicle model applicability, supports recognizing vehicle configuration columns (e.g., W,X,G) from user input
M2 Chapter Restructuring (srd-m2-restructure)	Dispatches a sub-agent for each function, restructures requirement content using a 9-subsection template and assembles into a complete SRD Markdown; supports multi-agent parallel subsection processing and image recognition insertion
M2 Chapter Template	Subsection 1 Overview And Project, Subsection 2 VMM Requirement (CM/UM constraint table), Subsection 3 Configuration, Subsection 4 Ports (external bus signals and physical attributes), Subsection 5 Internal Variable and DIDs, Subsection 6 Public Attribute, Subsection 7 Requirement (preserve original structure as much as possible), Subsection 8/9 Fault Handling & Fault Recovery, plus quality requirements and requirement traceability, pre-write checklist
M2 External Data Query	preload_databases.py preloads CCDB and signal matrix Excel as JSON cache, avoiding repeated parsing by each sub-agent; sub-agents query signal physical attributes and configuration parameters via query_ccdb.py, query_signal_matrix.py
M3 Word Generation (srd-m3-docx)	Converts the SRD Markdown output by M2 into .docx, applying style_standard.json standard styles, table of contents fields and formatted tables

2.2 Changes

Module	Details
None	None

2.3 Removed

Module	Details
None	None

2.4 Bug Fixes

Module	Details
None	None

---

v3.0.2

Release Date: 2026-07-07 | Base Version: v3.0.1

2.1 New Features

Module	Details
Diff tool subagent	Added diff tool subagent

2.2 Changes

Module	Details
None	None

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

2.1 New Features

Module	Details
-	-

2.2 Changes

Module	Details
Requirements analysis skill	Modified requirements analysis skill to generic skill

2.3 Removed

Module	Details
None	None

2.4 Bug Fixes

Module	Details
None	None