11/26/2024•4 min

Graplia Prompting Guide

promptinggraplian8n

Graplia Prompting Guide

From prompt to reliable automation — without burning your credits.

Reminder: the more precise your prompt is (trigger, parameters, errors, outputs), the less the AI has to guess — and the fewer credits you'll consume.

1) Graplia in a Nutshell

Graplia turns a prompt into an executable workflow. You describe what needs to happen (when to trigger it, which apps, which parameters, what to do on failure, where to send the result), and Graplia orchestrates the nodes, manages the connections, and delivers the output.

2) The 4 Building Blocks to Specify in Your Prompt

A. The Trigger

Clearly state when and how to start:

  • Manual, from the dashboard.
  • Scheduled (e.g., “every day at 08:00 Europe/Paris”).
  • App Event (new email, GitHub push on main…).
  • Webhook/API (if triggered from another system).

B. The Nodes and Their Parameters

List the apps and provide the minimum necessary info to avoid ambiguity:

  • GitHub: repositories (or URLs), branch, authors/teams, event types (merged PRs, tags, etc.).
    Ex. “Analyze org/project-front and org/project-api (or https://github.com/org/project-front), branch main, authors @alice, @bob.”
  • Mail (Gmail/IMAP): label/folder, sender(s), time period, item cap.
    Ex. “INBOX, since yesterday, from:partners@example.com, max 30.”
  • Databases/Documents (Notion/Drive/etc.): target space/base/table, expected fields/properties.
  • Transformations: output format (structured JSON, Markdown table, CSV).
  • Connections/OAuth: name of the connection to use (e.g. Github_Prod, Gmail_Sales).
  • Manually Editable Parameters: expose them in your prompt (“REPO”, “MAX_ITEMS”, “DATE_START”) for easy tweaks without regenerating.

C. Error Handling Strategy

Specify what to do if:

  • Connection fails: number of retries, delay between attempts, who to notify.
  • No data (e.g., no emails): send “Nothing to process” (don’t treat as an error).
  • Rate-limit/quota: retry, reduce scope, reschedule.
  • Unexpected schema or permissions: skip faulty items, log them, alert.

D. Output and Confirmations

Clarify:

  • Format: JSON with a clear schema, or a sorted Markdown table.
  • Destination: Notion (database + properties), Drive (folder/file), Git (repo/path), etc.
  • Confirmation: email(s) and/or Telegram (channel ID) with a summary (volume, success/failures).
  • Human validation for sensitive actions: wait for an “OK” before executing.

3) Quick Method to Write Your Prompt

  1. State the goal in one sentence.
  2. Define when it triggers.
  3. Specify scope and connections (repos, labels, dates, caps, accounts).
  4. Describe the transformation (sorting/grouping rules, format).
  5. Indicate where the output goes and whether validation is required.
  6. Explain error handling (retries, notifications, “zero result” case).
  7. Mark editable parameters for manual adjustments.
  8. Mention security constraints (least privilege, destructive actions require approval).

4) Examples (Bad vs. Good Natural Prompts)

Example 1 — Daily GitHub PR Report

Bad Prompt

“Make a daily GitHub commit report.”

Good Natural Prompt

“Every morning at 08:00 (Europe/Paris), I want a recap of merged PRs on org/project-front and org/project-api (branch main). Use the Github_Prod connection. Include only authors @alice and @bob and limit to 50 PRs in the past 24h. For each PR, return JSON with id, title, author, labels, file_count, link, sorted by label and then by file count (descending). Write the table at the top of the Notion page Daily Eng Report. If GitHub responds ‘rate-limited’, retry twice with 2-minute intervals, then alert ops@example.com. At the end, send a summary email to eng-leads@example.com and post it to Telegram channel -1001234567890.”

Why It’s Good: Clear trigger, scope, filters, output format, destination, and error handling.

Example 2 — Partner Emails Digest

Bad Prompt

“Summarize my important emails.”

Good Natural Prompt

“Every day at 09:00 (Europe/Paris), create a digest of emails from partners@example.com received yesterday in the INBOX of account Gmail_Sales, up to 30 messages. For each email, capture the sender, subject, up to three key points, and whether action is required. Publish the digest to Telegram channel -1002233445566 and also save a CSV in Drive/Sales/Digests. If there are no emails, send ‘Nothing to process’ instead of raising an error. If Gmail connection fails, retry twice, then notify sales-ops@example.com.”

Why It’s Good: Explicit criteria, capped scope, clear output, zero-result handling, and failure escalation.

Example 3 — Drive Cleanup with Human Validation

Bad Prompt

“Delete outdated files on Drive.”

Good Natural Prompt

“I’ll trigger this manually when needed. Scan Drive/Projects/Archives to list files without access in the last 180 days, but exclude the Legal folder. First, give me a Markdown table (top 100 by size) and a JSON {path, size, last_access}. Then, wait for my approval on Telegram -10099887766: I want to approve or cancel. Don’t delete anything without my OK. If permissions are missing, list the affected files and specify what delegation is required.”

Why It’s Good: Risky action properly gated by validation, precise scope, usable output, and permission handling.

5) Express Check-List

  • Trigger: manual / scheduled (time + TZ) / event / webhook
  • Connections & scope: accounts, repos/folders/databases, time period, caps
  • Manually editable parameters: REPO, MAX_ITEMS, DATE_START, etc.
  • Transformation: sorting/grouping rules, output format (JSON/MD/CSV)
  • Output: exact destination + field schema
  • Confirmations: emails and/or Telegram (channel ID)
  • Errors: retries, “zero result”, rate-limit, permissions, schema
  • Security: least privilege, approval before destructive actions

6) Structured Template (Copy, Paste & Fill)

  • Goal (1 sentence): …
  • Trigger: manual / scheduled (date+time, TZ) / event (app+event) / webhook
  • Apps & connections to use: …
  • Scope: targeted repos/folders/databases + filters (branch/labels/authors, senders, time period)
  • Editable variables before execution: REPO=…, DATE_START=…, LABELS=…, etc.
  • Transformation & expected format: sorting/grouping rules + (JSON schema / Markdown table / CSV)
  • Output destination: (Notion database+properties / Drive path / Git repo+path / other)
  • Confirmations & notifications: emails=… ; Telegram channel/ID=… ; summary content (volume, successes/failures, links)
  • Error handling: retries (count+delay), behavior if zero results, rate-limit/quota, schema/permissions; who to alert and how
Back to articles