跪拜 Guibai
← All articles
Frontend · Trae

PRD-to-Feature-List Automation Cuts a One-Hour Chore to 25 Minutes

By 优秀稳妥的JiaJi ·
Read original on juejin.cn ↗ Google Translate ↗ Alt translation

The bottleneck in turning a PRD into a buildable spec isn't intelligence — it's the repetitive mechanical steps of reading, screenshotting, and formatting. Automating that end-to-end eliminates omission errors and produces a list the whole team can use, not just cryptic personal notes. The Skill packaging means the workflow becomes a portable asset rather than a one-off script.

Summary

A frontend developer replaced the tedious pre-iteration ritual of manually copying PRD changes into an Excel feature list with a single TRAE Work command. The tool ingested an HTML PRD, an HTML interaction prototype, and an xlsx template, then extracted all six requirement changes — including rules, operation steps, and acceptance criteria that manual work routinely skipped. It spun up a local HTTP server to serve the prototype, navigated to each scene via URL parameters, and took screenshots automatically before generating the final spreadsheet via Excel COM automation. The whole run took 25 minutes, down from the usual hour, and the output was more complete and team-readable than the shorthand-only manual version. The workflow was then distilled into a reusable Skill package with PowerShell-based xlsx parsing and screenshot scripts, so the same pipeline transfers to other projects without rewriting prompts.

Takeaways
A TRAE Work command that reads an HTML PRD, an HTML prototype, and an xlsx template can produce a complete, screenshot-annotated feature list in 25 minutes.
Manual extraction routinely misses rule details and acceptance criteria; the automated pass captured all six changes with full rule text and operation steps.
The tool self-orchestrated a local HTTP server to serve the prototype, constructed scene-specific URLs from PRD scene IDs, and auto-screenshotted each page.
Parsing the xlsx template was the slowest step because the environment lacked Python; the agent fell back to PowerShell decompression of the xlsx ZIP and direct XML reading, hitting a Chinese-path encoding bug along the way.
The resulting Skill package includes a six-stage workflow instruction and five reference scripts (HTTP server, xlsx parser, screenshot cropper, Excel COM generator, JSON data template) so the pipeline transfers across projects.
Time dropped from ~60 minutes to 25 minutes (58% reduction), but the larger gain was completeness: structured, team-readable descriptions replaced shorthand that only the author understood.
Conclusions

The quality cliff in manual repetitive tasks is real — later entries get sloppier — and an agent sidesteps that fatigue entirely, producing uniform detail across all items.

The xlsx parsing detour (no Python → MCP failure → PowerShell ZIP/XML hack → Chinese path bug) is a microcosm of agent-in-the-wild reliability: environment gaps force fallback chains that eat time and surface edge-case encoding bugs.

Packaging the workflow as a Skill with a general layer and an HTML+xlsx enhancement layer is a practical pattern for making AI-assisted pipelines reusable across projects with varying input formats.

Concepts & terms
TRAE Work Skill
A packaged, reusable workflow definition that TRAE Work can auto-trigger when a matching task is encountered. It includes a main instruction file and reference scripts, with a general layer for broad applicability and an enhancement layer for format-specific optimizations.
Open XML / xlsx internals
An xlsx file is a ZIP archive containing XML files such as sharedStrings.xml (cell text), styles.xml (formatting), and xl/drawings/ (embedded images). Parsing these directly via PowerShell's System.IO.Compression is a fallback when Python's openpyxl is unavailable.
Source: juejin.cn ↗ Google Translate ↗ Backup ↗