- Home
- Skills
- Productivity
- Follow-Through
Works with the AI Agents you already use
Follow-Through
Turns meetings, email threads and chats into one tracked list of who promised what by when, keeps it current, and drafts the reminders for your approval.
$19
Follow-Through
Example session with this skill installed
Here are my notes from Wednesday's standup (4 Nov). Who committed to what, and by when?
- Read your context and instructions
- Compiled the follow-through
9 items from the 4 Nov standup – 3 need you
Needs you
- No owner: "Call the venue about the deposit" – Sami said "someone needs to call the venue". Who takes it?
- Date unclear: Lina's deck – "next Friday" said on a Wednesday could be 6 Nov or 13 Nov. Which?
- No date: Lina, send the contract to legal. Worth asking for one.
Commitments
| Who | What | Due | Said |
|---|---|---|---|
| Dana | Fix the onboarding email typo | Fri 6 Nov | "I'll fix it by Friday" |
| Omar → Dana | Send Q3 numbers | Mon 30 Nov | "Sure, end of month." |
| Omar (soft) | Look at the churn dashboard | – | "I'll try to look at it" |
Decisions
- Printing goes to vendor B.
Open questions
- Does legal need to review the new terms?
Connects securely to your tools. The creator never sees your data.
About this skill
Meetings produce promises. Most of them die in the notes.
Follow-Through makes your AI agent a reliable keeper of commitments. Give it meeting notes, a transcript, an email thread or a chat export and it returns who promised what, by when, with the words that prove it. It keeps one register up to date as new messages arrive, and it drafts the chasing without ever sending anything on its own.
What it does
- Only what was said. Every item carries a short quote from the source. No invented owners, no invented deadlines. "Someone should call the venue" is flagged as unowned, not pinned on whoever spoke last.
- Dates done properly. "By Friday" is resolved against the day it was said, not today. "Next Friday" said on a Wednesday is flagged as ambiguous, with both candidates, instead of silently picked.
- One list, no duplicates. New material updates the existing register: done, moved, reassigned, dropped. Old dates and owners stay in the history. Rows the new material did not touch are left alone.
- Chasing with judgement. One message per person, a question (not a reminder) for tentative promises, nothing for items already delivered, your own overdue items shown first. Drafts only: nothing is sent without your go-ahead.
- A weekly digest whose counts add up.
What is in the package
SKILL.md– the working method (capture, dates, register, reconcile, report, chase, digest)scripts/register.py– a standard-library Python tool that keeps the register as a CSV you can open in Excel or Google Sheets, resolves spoken dates, refuses likely duplicates and validates the file- Three reference guides: commitment language and edge cases; how transcripts, notes, email and chat each mislead; reminder templates by relationship
- Commercial licence
Tested, with honest numbers
Six scenarios, each run three times with the skill and three times without, scored by independent graders: 192 of 192 checks with the skill, 179 of 192 without. The unassisted model is already competent at this job; it got owners and dates right either way. The skill's gain is in traceability and discipline: evidence quotes beside every item, untouched rows left untouched, clear counts of what changed. Graders also noted, outside the score, that unassisted runs added commitments nobody had agreed to and wrote reminder drafts with deadlines the user never set. Costs: replies are 30 to 50% longer, about 15% more tokens and 25% more time per task.
Does not send messages or post to task tools on its own; it drafts and waits for approval. Tested on English sources of up to about 60 lines; very long transcripts and other languages were not tested. It cannot guarantee every commitment in a source is captured: review the list before relying on it.
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
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 2 days ago
Requires
Permissions
File Scopes
No network access. The script reads and writes only the register file you point it at. The skill never sends messages without explicit approval.