WorkBuddy Deep Dive: 10 Advanced AI Productivity Features with Prompts & Workflows
This article details 10 practical WorkBuddy features including persistent-memory automation, reusable Skills, browser control with live login state, in-document AI co-writing, table analysis, custom MCP connectors, shared resource library, expert agents, WeChat remote dispatch, and one-click inspiration replication, complete with prompts and cost comparisons.
01 Automation: Persistent-Memory Cron with Experts, Skills & Work Directories
WorkBuddy's automation is more than a simple cron job. It combines persistent memory , bindable experts , skills , and work directories to turn scheduled tasks into context-aware workflows.
Daily Work Log
Runs nightly at 10 PM, aggregates the day's activity into a structured Markdown file worklog-YYYY-MM-DD.md. Data sources (all optional):
Retrieve all chat sessions from the day, group by topic into work items.
Check project directories with file changes (excluding dependencies/cache) as evidence of output.
Incorporate conclusions from any existing project health reports or daily logs.
Output follows a fixed template: # Work Log YYYY-MM-DD, ## Today's Main Thread (≤3 lines),
## Work Item Progress (goal / done / conclusion / blockers / next step), ## Open Loops, ## Tomorrow's Priorities (≤3), ## Reusable Artifacts.
Privacy rules: no real names, company names, credentials, absolute paths; paths become relative under "Workspace"; people generalized to "Collaborator"; no raw chat/code excerpts; no speculation.
If no sessions and no file changes, returns NOOP without creating a file. After writing, replies with a ≤3-line summary.
Project Health Daily
Checks local services and produces an anomaly-first brief health-YYYY-MM-DD.md with three tiers: Abnormal / Warning / Normal . Checks:
Port liveness + health endpoint probe.
Recent activity: last 3 modified files per project (excl. deps/cache).
Dependency health: lockfile exists and newer than dependency dir; missing install flagged.
Staleness: artifact dir or status file older than 3 days → "Stalled".
Non-git directories skip git commands.
Dialogue summary ≤5 lines, ordered by severity. All results are measured, not guessed; failures recorded as-is.
The author notes WorkBuddy automatically generated a dashboard for all scheduled tasks.
02 Skills: Codify Repeated Workflows into Reusable Capabilities
A Skill packages a best-practice workflow into a deterministic capability. Distinction:
Skill = tool capability (scripts + workflow). Use when AI needs a specific ability.
Expert = AI advisor (persona + methodology + toolchain). Use for a well-defined single-point problem.
Expert Team = multiple experts + collaboration flow. Use for complex multi-role tasks.
One-liner: Skill is capability, Expert is capability + experience, Expert Team is multiple experts + collaboration flow.
Skill directory structure:
my-skill/
├── SKILL.md # required, skill spec
├── scripts/ # optional, executable scripts
├── references/ # optional, reference docs read on demand
└── assets/ # optional, templates, images, fontsSKILL.mduses YAML frontmatter; name and description are mandatory. Description is the most critical field — WorkBuddy uses it to decide whether to load the skill. A vague description means the skill never triggers.
Example: a third-party "Content Ops" skill for content quality scoring. Its description was left empty "", so it only runs when explicitly @-mentioned.
Rule of thumb: If you do the same thing a third time, turn it into a Skill.
Creation template prompt: "Turn my recent workflow into a Skill in the user-level directory. Requirements: description clarifies trigger scenarios; repeated commands become scripts in scripts/; output templates go to assets/; verify with a new example that it triggers correctly."
Maintenance tip: periodically review Skills; update descriptions or delete those never triggered.
03 Browser Control: Drive Your Live Browser via Extension
Unlike headless browsers or search APIs, WorkBuddy's browser control connects via a Chrome/Edge/QQ Browser extension, inheriting your active login state, cookies, and open tabs .
Enable: Settings → Agent → Browser Control → toggle on, install extension.
Unconventional Uses ("邪修用法")
Turn any authenticated backend into a structured data source. Many SaaS backends lack export APIs or even CSV buttons. Log in, open the list page, then prompt WorkBuddy to paginate, click detail modals for hidden fields, and emit a CSV to the workspace. Example prompt: "Watch the open backend order page, keep login. Page through last 30 days, extract order ID, amount, status, order time into CSV. Click 'View Detail' for hidden fields. Never click refund/delete/submit buttons."
"Read-to-write": feed local drafts into multi-step web forms. For content publishing, product listing, or form filing, give WorkBuddy the local file path and the logged-in backend page. It fills fields one by one, uploads attachments, saves draft, and screenshots for review. Prompt: "Read draft.md, fill title/body/summary/tags into the open backend editor. Strictly: only fill and click 'Save Draft', never 'Publish', then screenshot for verification."
Zero-Python backend monitoring. Borrow the browser's live session to periodically diff a logged-in dashboard/alert page against a local snapshot; summarize changes. Avoids maintaining Playwright scripts and cookie rotation.
04 Human-AI Co-Writing: In-Place Editing Without Round-Trips
Select any text in Word, Excel, PPT, or Markdown → invoke AI to rewrite just that selection → changes appear instantly, no save/reopen cycle.
Example prompt: "Rewrite the selected passage for external release: keep meaning, strip internal jargon and colloquialisms, do not touch any other sentence."
The explicit "do not change other sentences" guard prevents unwanted rewrites.
Ideal scenarios:
Commercial copy / article finalization — preserve style, tweak locally.
Add a formula column in a spreadsheet — no need to rebuild the whole sheet.
Swap title/data on a single PPT slide — leave other slides untouched.
Unify terminology across a long doc — edit piece by piece with immediate visual feedback.
05 Table Analysis & Visualization: Understand Your Existing Data
Given an Excel/CSV (exported orders, survey results, finance details), WorkBuddy reads, cleans, analyzes, and charts — no need to generate new tables.
Prompt example: "Read this sheet. Tell me: row count, columns with missing values, date format consistency. Then pivot by 'Month × Category', highlight top 3 MoM changes, render a chart. All conclusions strictly from data; unanswerable items listed as 'Needs Additional Data'."
Capabilities: batch formulas, cross-sheet references, replace manual calc with formulas; deduplication, column splitting, date/amount normalization; pivot, cross-tab, sort/filter; generate presentation-ready charts directly.
06 Connectors + Custom MCP: Bridge Agent to External World
Connectors let the agent read email, browse docs, query project boards, send meeting invites. Four categories:
Data Query — pull from external sources (DB, knowledge base).
Service Call — invoke third-party APIs (send email, create calendar event, create task).
File Management — access cloud storage / filesystems (read cloud drive, upload attachments).
Messaging — integrate with IM (send WeCom / Lark messages).
Custom MCP is the real differentiator. Official integrations cover only mainstream apps; custom MCP lets you connect private assets: local note vault, NAS photo library, internal dashboard, personal micro-API. Example: the author mounted Typora notes via a custom MCP to read articles externally.
Prompt to add a custom MCP: "Add a custom MCP service to the config file, merge only the new entry, don't overwrite existing. Output the added snippet for review, do not start the service yet."
Note: newly added MCPs require manual "Trust" in the Connector Management page under "Custom Connectors" before activation.
07 Resource Library: Shared Workspace for Human & Agent
Not just a network drive — the shared disk between you and the agent . Task outputs (docs, sheets, HTML) and user uploads coexist. Three value layers:
Basic — store/retrieve artifacts.
Advanced — co-editing: multi-user + AI revision + paragraph comments.
Pro — build personal workbenches: HTML + CSV → lightweight web app; publish local files as public links.
Prompt: "Save this task's artifacts to the Resource Library, organize by project directory. Publish the HTML artifact as an online link and give me the URL."
The author emphasizes layer 3: many vibe-coded prototypes stay local; publishing a link turns them into shareable tools.
Distinction: Resource Library manages deliverables; Knowledge Base manages long-term, repeatedly retrieved knowledge.
08 Experts / Expert Teams: Persona + Methodology + Toolchain
An Expert is not just a persona — it bundles persona + methodology + toolchain . A finance expert and a legal expert approach the same material differently: what to look for, how to judge, output format.
For a beginner building a WeChat Mini Program, an expert lowers the entry barrier and saves token-cost trial-and-error.
Expert Team = multiple experts collaborating (e.g., MVP Dev Team: Project Director, PM, UI, DevOps, Frontend QA). Equivalent to hiring a small agency paid only in tokens.
Cost alert: Expert Teams run multiple experts in parallel with multi-round model interactions, consuming 3–5× the credits of a single expert .
Author's conservative routing:
Clear single-point problem → Single Expert.
Deliver a full system, multi-role needed → Expert Team.
Just need a fixed AI capability → Skill, not Expert.
Binding an expert into automation: same cron, different expert = different "person" showing up on schedule (minus performance reviews).
Current issue: many experts/teams have overlapping responsibilities.
09 WeChat Assistant: Remote Task Dispatch
Bind the assistant in settings → send commands via WeChat → WorkBuddy executes on your computer.
Example end-to-end flow: scheduled content fetch → store in knowledge base → deduplicate & merge into "New Intelligence Brief" → render as HTML → push to WeChat.
Limitation: some content sources lack webhook callbacks for new articles, so real-time push is impossible; polling is the only option. This is an upstream limitation, not WorkBuddy's.
10 One-Click Inspiration Replication
WorkBuddy's Inspiration Community hosts real, runnable works (personal dashboards, data boards, interactive pages, analysis templates). "One-click replicate" clones the entire working artifact.
Solves cold-start: replicate first, tweak later — almost always faster and cheaper than prompting from scratch. Even if half the page gets rewritten, a runnable reference cuts iteration cost dramatically.
Additional Daily-Used Features
Deep Research — multi-source retrieval + cross-verification → direct report.
Three Modes — Craft (execute directly), Plan (think first), Ask (advise only).
Multimodal Generation — images, video, 3D models.
Agent Email — send/receive as you, with attachments.
Quick Reference Table
Automation — memory-enabled cron with expert/skill binding → recurring, time-triggered tasks.
Skills — solidify workflow into capability → when you've done it the third time.
Browser Control — borrow your live login to drive pages → cross-page scraping, auto-form-fill, backend patrol.
Human-AI Co-Writing — select & edit in-place → local edits on existing docs while preserving overall style.
Table Analysis & Viz — comprehend your data → have data, need insights & charts.
Connectors + MCP — let agent reach external world → read mail, browse docs, send messages.
Resource Library — shared human-agent disk → artifact persistence, collaboration, link publishing.
Experts / Expert Teams — capability+experience / multi-role collab → single issue → expert; big project → team.
WeChat Assistant — dispatch & collect via WeChat → away from computer but need results.
Inspiration Replication — clone proven works → don't know where to start, want a fast custom version.
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
