Collection guide

Stop writing from memory. Have the AI you use gather and organise what today left behind in Slack, GitHub and your calendar — then review it yourself before saving it to Effica Grove.

The one prompt worth remembering

Draft today's daily report.
Fetch the collection guide from Effica Grove, decide for yourself which of my
connections match the recommended capabilities, and gather my activity for that day.
Organise it into "what I did / what I noticed / where I got stuck / tomorrow's hypothesis",
and report which connections you consulted. Do not confirm the report yet.

How it works

The AI gathers, you review, Effica Grove keeps.

SaaS / tools

What you left behind in Slack, GitHub, Calendar, Gmail, Linear, Notion and Figma

Your AI

Gathering, merging and organising. The reasoning runs on your AI's side

Your review

Read what the AI proposes and fix it. Nothing becomes a confirmed record without this

Effica Grove

Saved along the template and persisted to your own GitHub

Choose the AI you use

The steps and prompts below change with your choice

The AI you use

Collecting with ChatGPT

Have it search the SaaS you connected as connectors, straight from the conversation.

Web / Desktop · Plus and above

1
Open Settings → Connectors
Open Settings → Connectors in ChatGPT. If it is not there, it is disabled by your plan, your region or an admin setting.
2
Check what you have already connected
You do not need to connect anything new for this report. Whatever your current connectors can reach is enough.
3
Send the prompt with an explicit date
Always write the date. "Today" alone can land on the wrong day depending on the hour.
4
Review it before saving
The AI's output is a proposal. Read it and fix it before handing it to the save step.

Example prompts for ChatGPT

Do not name services. Which connections to use is for the AI to decide from the collection guide. What you must always include is the date, and the narrowing to your own involvement.

Gather

Gather my activity for today from my connected connectors,
using only those matching Effica Grove's collection guide.
Also report which connections you used.

Narrow down

From what you gathered, keep only the work I was involved in myself.
Drop other people's work and automated notifications, and
summarise what others said as work-related facts rather than quoting them.

Shape it

Sort what you organised into the sections of
the daily report template.

The collection rules reach your AI automatically

If Effica Grove is connected, your AI calls get_collection_guide before it starts writing and receives, in machine-readable form, which sources it may draw on and what to collect from them. There is no need to copy any of the below into your prompt. Effica Grove never calls your other SaaS on your behalf — the gathering happens through your AI's own connections.

Recommended sources (given as capabilities)

They are passed as a capability and a list of items to collect, not as service names. A connector not listed here is still fine if it has the same capability.

sourceControl
Implementation
e.g. GitHub / GitLab / Bitbucket
commits · pull_requests · reviews · issues
teamCommunication
Communication
e.g. Slack / Microsoft Teams / Discord
messages_by_me · threads_participated · mentions
calendar
Meetings
e.g. Google Calendar / Outlook Calendar
events_today · participants
taskManagement
Tasks
e.g. Linear / Jira / GitHub Projects
done · in_progress · assigned · commented
documentManagement
Documents
e.g. Notion / Google Drive / Confluence
created_documents · updated_documents
design
Design
e.g. Figma
updated_files · comments · reviews

Rules the AI follows

They are passed in both Japanese and English, and the ids never change.

Collect only activity the user was involved in; do not record colleagues’ or the team’s work as the user’s own.
collect-only-user-activity
Do not use connections outside the recommended capabilities, or gather information unrelated to work.
no-unrelated-connectors
Skip recommended sources that are not connected; do not force a substitute source in their place.
skip-unavailable-sources
Do not carry over the raw text of direct messages, private channels, or email.
no-raw-private-messages
Record what others said as the work-related facts the user needs, not as verbatim quotations.
summarize-not-quote
Give the organization’s AI usage policy, connection permissions, and administrator settings precedence over these rules.
respect-org-policy
Save a generated summary only after the user has confirmed it; never save unconfirmed.
confirm-before-save

Five ways to collect — the least work is at the top

Only move further down when the SaaS you use has no official connector.

#	Method	What it is	Examples of what you get	Setup	Freshness	What is exposed
1	Connectors / apps	The official ChatGPT / Claude integrations. The AI reads the SaaS directly	Your PRs, reviews and issues / messages and threads you joined / the day's meetings and attendees / pages you updated	A few clicks	Always current	What the connection asked for
2	MCP connectors	Reach a service with no official connector through an existing or self-built MCP server	Fetched by repository, channel and date range — down to diffs and state changes	Needs building	Always current	What the MCP server exposes
3	API / CLI	Run gh, git and friends from Claude Code or Codex	gh pr list --author @me gh issue list --author @me git log --since="today"	An authenticated machine	Always current	Whatever your token can reach
4	Local files	Have it read Markdown, JSON or CSV from disk or Git	Changes inside a given folder, notes you appended	Just point at the file	As fresh as the file	Only the files you hand over
5	Export files	The fallback for a SaaS with neither a connector nor an API (CSV, ZIP and so on)	Only the records as of the moment you exported	Export each time	As of the export	Only the exported file

What you can actually get varies with the connection's implementation, permissions and plan.

What the mechanism guarantees as well

Even if the AI gets it wrong, none of these can happen on the Effica Grove side.

An AI cannot confirm a report

There is no way to confirm a report over MCP. Confirming happens in Effica Grove, or from code you wrote yourself.

The target date is always explicit

The saving tools require a target date (YYYY-MM-DD) and never fill in "today". That stops a late-night session from landing on the wrong day.

Only daily reports and diary entries are writable

Weekly reports and above are generated by GitHub Actions. An AI cannot rewrite the higher-level reports directly.

Records written by an AI are marked

Anything written over MCP is labelled, so you can tell afterwards where it came from.

"Handing over" and "keeping" are different things

Data you can fetch is not automatically data you should keep in a daily report. The same content is treated differently at each of the two stages.

Handing over (temporary)

SaaS → your AI

Other people's words, personal data and confidential material can be mixed in. How it is handled follows your AI provider's terms and your organisation's policy.

Keeping (permanent)

What you reviewed → your GitHub

Persisted to a private repository and possibly shown to others later. Real names and confidential material do not belong here.

Read the full record-keeping criteria

If you have not connected an AI yet

Saving what you gathered to Effica Grove requires the setup in the external connection guide first.

To the external connection guide