Twenty-five copy-paste scenarios for getting real work done with Web Agent. Each entry maps to bundled skills and the tools those skills expect. Prompts stay in English — paste them as-is into chat.
Surface this playbook assumes: 49 built-in tools (skill CRUD via unified skill tool), 19 bundled skills. Full catalog: README § Tooling · README § Skills.
Filter by category: Research · Memory · Planning · Automation · Workspace · Debug · Multimodal · Delivery · UX · Safety · Meta
| Category | Use case | Bundled skill(s) | Key tools |
|---|---|---|---|
| Research | Find niche creators / competitors | /open-web-research |
web_search, web_fetch, write_file, artifact_present |
| Research | Academic paper / citation dig | /research-pack |
web_search, web_fetch, write_file, artifact_present |
| Research | Extract a table or JSON from a page | /structured-extraction |
web_fetch, write_file, artifact_present |
| Meta | Discover installable skills online | /find_skills |
web_search, web_fetch, skill |
| Memory | Save a durable preference | /memory-layers |
memory_save, memory_recall |
| Memory | Capture rolling session context | /memory-layers |
session_memory_append, session_memory_list |
| Memory | Mirror memory into Obsidian-style vault | /memory-layers |
wiki_setup, wiki_sync, wiki_search |
| Memory | Find something from an old chat | /memory-layers |
session_search |
| Planning | Spec-first feature plan (no execution yet) | /plan |
read_file, grep, write_file, artifact_present |
| Planning | Plan and execute multi-step work | /task-execution |
todo_write, write_file, artifact_present |
| Automation | Daily digest while tab is open | /heartbeat-cron |
cron_register, cron_list, web_search, web_fetch |
| Workspace | Bootstrap a new side project folder | /project-scaffold |
make_dir, write_file, browse_workspace |
| Workspace | Reorganize files safely | /workspace-safety, /browser-runtime-map |
browse_workspace, move_file |
| Debug | Hypothesis-first bug hunt | /systematic-debugging |
read_file, grep, file_diff, run_shell |
| Debug | Shell / npx failed in WebContainer |
/browser-runtime-map |
read_file, web_fetch, grep |
| Multimodal | Read a screenshot or diagram | /multimodal-ingest |
vision_analyze, write_file |
| Multimodal | Summarize a YouTube tutorial | /multimodal-ingest |
youtube_transcribe, write_file, artifact_present |
| Delivery | Present a finished report in-app | /artifact-delivery |
write_file, artifact_present |
| Delivery | Email a deliverable | /artifact-delivery |
write_file, email, artifact_present |
| Delivery | Flowchart for a plan or report | /chart |
artifact_present |
| UX | Disambiguate a vague ask | /clarify |
(none) |
| Safety | Checkpoint before bulk delete | /workspace-safety |
browse_workspace, delete_file |
| Safety | Pasted API key by mistake | /artifact-delivery |
Redact in replies; memory_save only when user asks |
| Meta | Improve Web Agent itself | /web-agent-skill |
read_file, grep, skill, memory_save |
Find niche creators / competitors — open-web discovery with verified fetches
When you need a shortlist of people, channels, or companies in a niche — not academic papers.
Try this prompt:
Find 8 YouTube creators in the UAE who regularly post about AI agents or coding assistants.
Verify channel pages with fetches, label each as confirmed/likely/not regional, and save a markdown table under work/research/uae-creators.md — then show it to me.
Bundled skills: /open-web-research (+ /clarify if scope unclear; /structured-extraction for row shaping; /artifact-delivery for preview)
Tools that fire: web_search, web_fetch, write_file, artifact_present
Academic paper / citation dig — arXiv and Semantic Scholar paths
When you want papers, citations, or a mini literature review — not general creator discovery.
Try this prompt:
Find 5 recent arXiv papers (2024–2026) on retrieval-augmented generation for code assistants.
Summarize each in 3 bullets, include PDF links, and save work/research/rag-code-assistants.md with a references section.
Bundled skills: /research-pack (+ /artifact-delivery for preview)
Tools that fire: web_search, web_fetch, write_file, artifact_present
Extract a table or JSON from a page — fetch, parse, normalize
When a webpage has rows you want as CSV or JSON — not a narrative summary.
Try this prompt:
Fetch https://example.com/pricing (or any public pricing page I paste next) and extract plan name, price, and key limits into a deduped JSON array.
Save work/extract/pricing.json and show me a preview table.
Bundled skills: /structured-extraction (+ /artifact-delivery for preview)
Tools that fire: web_fetch, write_file, artifact_present
Discover installable skills online — registry search and install path
When you want to find community skills from online registries before installing one.
Try this prompt:
/find_skills markdown wiki knowledge base
Search skills.sh, SkillsMP, and Cursor marketplace — return the top 5 by installs or stars with links. Do not install yet; just rank them.
Bundled skills: /find_skills (+ /clarify if query ambiguous; /web-agent-skill if installing)
Tools that fire: web_search, web_fetch, skill
Improve Web Agent itself — safe self-evolution in this repo
When you are hacking on Web Agent runtime, skills, or capabilities — not general app work in your workspace.
Try this prompt:
Read how bundled skills are indexed in src/agent/runtime/memory/skills.ts and suggest one surgical improvement to the skills context block. Do not edit bundled SKILL.md files — propose a patch plan only.
Bundled skills: /web-agent-skill (+ /memory-layers for where to store lessons)
Tools that fire: read_file, grep, skill, memory_save
Save a durable preference — facts that survive reloads
When a preference should persist across sessions — not a one-off session note.
Try this prompt:
Remember this for future turns: my default formatter is Prettier with semi: true and singleQuote: false.
Tell me exactly what key you stored and how I can verify it next session.
Bundled skills: /memory-layers
Tools that fire: memory_save, memory_recall
Capture rolling session context — lightweight in-session notes
When you want a decision or status captured for this conversation — not a permanent fact.
Try this prompt:
We're exploring option B for the dashboard layout. Append a short session note with that decision and what we ruled out — I don't need this as a permanent fact yet.
Bundled skills: /memory-layers
Tools that fire: session_memory_append, session_memory_list
Mirror memory into Obsidian-style vault — PARA markdown for humans
When you want browseable wiki files synced from runtime memory — default vault under .webagent/knowledge-vault/.
Try this prompt:
/wiki_setup
/wiki_sync all
/wiki_search deployment
Show me what landed in the vault index and one snippet from search.
Bundled skills: /memory-layers
Tools that fire: wiki_setup, wiki_sync, wiki_search
Find something from an old chat — search archived sessions
When you remember discussing something but not which session — keyword search across history.
Try this prompt:
Search my past conversations for mentions of "Prettier" or "formatter" and quote the most relevant snippet with the session date if available.
Bundled skills: /memory-layers
Tools that fire: session_search
Spec-first feature plan (no execution yet) — `/plan` before code
When you want a reviewable markdown spec saved under plans/ — implementation comes on a follow-up message.
Try this prompt:
/plan Add a dark-mode toggle to the settings panel with persisted preference and a short README note
Follow-up (next message):
Execute the plan you just wrote.
Bundled skills: /plan command (+ /artifact-delivery for plan preview)
Tools that fire: read_file, grep, write_file, artifact_present
Break a stacked request into todos — decompose before tool fan-out
When one message contains several deliverables — ordered todos before action.
Try this prompt:
I need you to: (1) list top-level files in the workspace, (2) grep for TODO comments, (3) write a one-page summary markdown. Break this into todos first and show me the list before doing step 1.
Bundled skills: /task-execution (+ /chart if ≥4 steps)
Tools that fire: todo_write, skill
Execute an approved multi-step plan — gated step-by-step run
When todos already exist and you said "go ahead" — full report at the end.
Try this prompt:
Execute the plan you just wrote. Stop if a step fails and give me a partial report.
Bundled skills: /task-execution (+ /systematic-debugging on failure; /artifact-delivery for report)
Tools that fire: todo_write, read_file, write_file, grep, artifact_present
Daily digest while tab is open — heartbeat cron, not host crontab
When you want a recurring job while the app tab stays open — digests, reminders, periodic search.
Try this prompt:
Register a daily heartbeat job at 9:00 AM (my timezone) that web_searches "AI agent browser news", fetches the top 2 results, and appends a 5-bullet digest to work/digests/daily.md. List the job with cron_list when done.
Bundled skills: /heartbeat-cron (+ /browser-runtime-map for step tools; /artifact-delivery for delivery options)
Tools that fire: cron_register, cron_list, web_search, web_fetch, write_file
Bootstrap a new side project folder — scaffold before writes
When starting a new app, demo, or spike — isolated folder under projects/ or work/.
Try this prompt:
Create projects/expense-tracker/ with a minimal README, package.json stub, and src/ folder. Show me the tree when done.
Bundled skills: /project-scaffold
Tools that fire: make_dir, write_file, browse_workspace
Reorganize files safely — checkpoint mindset before moves
When cleaning up uploads or old drafts — list and verify before bulk moves or deletes.
Try this prompt:
List everything under uploads/ and work/. Propose a safe reorganization (moves only, no deletes yet) and explain what you'd checkpoint first.
Bundled skills: /workspace-safety, /browser-runtime-map
Tools that fire: browse_workspace, move_file
Hypothesis-first bug hunt — reproduce before random edits
When something fails, flakes, or regresses — structured diagnosis, not guess-and-patch.
Try this prompt:
The chat input stops accepting paste after I upload an image. Form one hypothesis, name the smallest read-only checks you'd run first (grep/read_file), and only then suggest a fix.
Bundled skills: /systematic-debugging
Tools that fire: read_file, grep, file_diff, run_shell
Shell / npx failed in WebContainer — pick the right tool
When curl, git clone, or npx fails — use file and HTTP tools instead of host shell.
Try this prompt:
run_shell failed with "command not found" when I asked you to curl an API. Fetch the same URL with web_fetch instead and explain what works in this browser runtime vs what doesn't.
Bundled skills: /browser-runtime-map
Tools that fire: read_file, web_fetch, grep
Read a screenshot or diagram — vision before reasoning
When the input is an image path in the workspace or an upload — OCR, UI state, architecture diagrams.
Try this prompt:
I uploaded a screenshot to uploads/. Analyze it with vision: list every visible error message and button label verbatim, then save work/notes/screenshot-audit.md.
Bundled skills: /multimodal-ingest
Tools that fire: vision_analyze, write_file
Summarize a YouTube tutorial — transcript-first
When you paste a YouTube link — full transcript with timestamps before summarizing.
Try this prompt:
Summarize this tutorial: https://www.youtube.com/watch?v=dQw4w9WgXcQ
Pull the transcript, extract 8 timestamped takeaways, and present a short markdown report.
Bundled skills: /multimodal-ingest (+ /artifact-delivery for preview)
Tools that fire: youtube_transcribe, write_file, artifact_present
Present a finished report in-app — View popup, not inline dump
When the deliverable is markdown or a file — use the artifact bar for preview/download.
Try this prompt:
Write a one-page weekly status report to work/reports/week.md (Goals / Done / Blockers / Next) and present it with artifact_present — don't paste the full body in chat.
Bundled skills: /artifact-delivery
Tools that fire: write_file, artifact_present
Email a deliverable — outbound via Resend when configured
When you need the report emailed — requires Resend configured in settings.
Try this prompt:
Draft work/reports/summary.md with a 5-bullet project update, present it, and if email is configured send it to my address with subject "Weekly update".
Bundled skills: /artifact-delivery (+ /memory-layers if keys mentioned)
Tools that fire: write_file, email, artifact_present
Flowchart for a plan or report — Mermaid via artifact preview
When a diagram clarifies steps, flows, or architecture — GitHub-style Mermaid in artifact view.
Try this prompt:
Draw a Mermaid flowchart of my morning routine automation (wake → digest → todo review → deep work) and present it so I can view the diagram in the artifact popup.
Bundled skills: /chart (+ /artifact-delivery)
Tools that fire: artifact_present
Disambiguate a vague ask — structured choices, no guessing
When intent is ambiguous — one clarification block with options in the UI.
Try this prompt:
Help me organize my stuff.
Or explicitly:
/clarify I want to improve my workflow but I'm not sure where to start
Bundled skills: /clarify
Tools that fire: (none — emits <<<CLARIFY>>> block)
Checkpoint before bulk delete — list and stat before destructive ops
When you want cleanup — export or list scope before delete_file or wide refactors.
Try this prompt:
I want to delete everything under work/scratch/. List what's there with sizes, confirm nothing outside that tree would be touched, then delete only if the list matches my intent.
Bundled skills: /workspace-safety
Tools that fire: browse_workspace, delete_file
Pasted API key by mistake — redact in chat; save only if asked
When you accidentally paste a key — redact it from replies and artifacts. Use memory_save only when the user explicitly asks you to remember it (browser-local storage).
Try this prompt:
I pasted this by mistake: sk-test-1234567890abcdef — redact it from your reply and tell me how to rotate safely. Do not save it unless I ask.
Bundled skills: /artifact-delivery, /memory-layers
Tools that fire: (redaction guidance; optional memory_save when user requests)
- test-prompts.md — contributor smoke prompts
- README.md — product overview and tool catalog
- CAPABILITIES.md — adding skills and capabilities