- Home
- Skills
- Productivity
- Pin4Html — Live co-editing for HTML reports
Works with the AI Agents you already use
Pin4Html — Live co-editing for HTML reports
Mark up your agent's HTML report in the browser — rewrite, delete, verify, move. It fixes each note while you keep reading, and the page updates in place.
Free
Pin4Html — Live co-editing for HTML reports
Example session with this skill installed
I've finished reading the Q3 growth review in the browser and left 7 marks. Fix the fix-now ones while I keep going, then handle the rest when I say go ahead.
- Read your context and instructions
- Compiled the pin4html
- Generated the document
Summary
Handled the 5 fix-now marks on showcase-en.html as they came in. The page updated in place after each edit; you never reloaded. 2 marks were left as later and are waiting for "go ahead".
Fix-now marks
| # | Type | What you marked | What I did | Status |
|---|---|---|---|---|
| 1 | rewrite | "grew 34% quarter over quarter" | Used your replacement: "grew 34% quarter over quarter (36,000 → 48,200)". Also added the absolute numbers to the summary card so the two match. | resolved |
| 2 | comment | Pin on the Monthly churn 3.1% KPI card | Split the card into Student / Personal / Team. Personal and Team show flat at 2.2%, so the rise is visibly the student plan. | resolved |
| 4 | layout | Box over the right third of Figure 1 | Shaded the August promo window and labelled it "Student plan launch", so the jump explains itself. | resolved |
| 5 | delete | "Team plan growth came almost entirely from existing customers adding seats." | Removed it from Revenue by plan. There's no expansion section yet; tell me if you want one and I'll move it there. | resolved |
| 8 | comment | "Keep the Team plan pricing unchanged through Q4." | Note says Sales agrees, so no change. | resolved |
Waiting for "go ahead"
| # | Type | Mark |
|---|---|---|
| 6 | question | "This is a seasonal effect…" — Do we have last year's data to back this up? |
| 7 | add | Add after the annual-billing line: "Pilot with 2,000 students in November; target churn below 2.5%." |
For #6 I won't assert seasonality without data. If last year's churn numbers aren't in the report, I'll soften the sentence to "may be seasonal" and ask you for the figures rather than invent them.
Next
- Keep marking;
watchis running again and will pick up new fix-now marks when you pause. - Say go ahead to process #6 and #7.
showcase-en.html
PDF · document
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
Your agent writes a report. You read it in the browser and right-click whatever's wrong: rewrite this, delete that, verify this number, move this chart. Marks with "fix now" reach the agent within seconds. It edits the file, replies to each mark, and the page updates in place. No reload, no copy-pasting feedback into chat.
What makes it different
- Live loop, not export. Feedback flows back while you're still reading, so review and fixing happen at the same time.
- Seven typed marks: comment, rewrite, delete, add, verify, question, layout. The agent knows exactly what you want, not just where.
- Pins and box selection for charts and layout, plus text selection for wording.
- Consistent edits. When a number or fact changes, the agent updates every other occurrence in the report and tells you where.
- Safe concurrent editing. You and the agent can write at the same time; nothing gets lost.
- Never touches your report with annotator code. The script is injected by a local server.
- Zero dependencies. One Python script, one JS file, Python standard library only.
- Local only. Server binds to 127.0.0.1. No telemetry, no external calls.
- English and Traditional Chinese UI.
How it works
- Start: the agent runs
serveon the report, which opens it in your browser with the annotation layer. - Mark up: select text or right-click; Alt+D boxes an area. Leave "fix now" checked, or uncheck to batch for later.
- Watch: a background
watchloop hands new fix-now marks to the agent when you pause. - Fix and reply: the agent edits the source HTML and replies to each mark; the page ticks them off and updates in place.
Use cases
- Reviewing reports, research briefs and technical docs your agent produced
- Iterating on landing page copy and layout with instant visual updates
- A "verify" pass on claims and numbers before you share a draft
Known limitations
Needs a local Python 3 and an agent that can run background commands. Annotations live in a .pins.json file next to the report. Without the server (offline review file), there's no live loop; reviewers hand marks over with "Copy for AI".
Open source (MIT): https://github.com/youllook/Pin4Html
Needs a local Python 3 and an agent that can run background commands on the same machine as the browser. Without the server (offline review file) there is no live loop; reviewers hand marks over with "Copy for AI". Marks anchor to text and DOM selectors, so heavy structural rewrites of the HTML can leave older unhandled marks pointing nowhere; re-select them with a right-click.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 9 days ago
- Free with an Agensi account
Requires
Permissions
Allowed Hosts
Terminal: runs a local Python server (bound to 127.0.0.1 only) and a background watch command. Read/Write files: reads the report, writes <report>.pins.json and edits the report when fixing marks. Browser: opens the local review page. No internet access is used; cdn.jsdelivr.net is only referenced if the user opts into `inject --cdn`.