Library — DistroSource

Everything digital. One source.

567 resources

Project · Intermediate · 8h
Project: A tested prompt library for one job
Build five reusable prompts for one role's recurring tasks, each with variables, a reference run and a ten-case test set.
- Choose one role (support, sales, marketing, ops). Identify its five most repeated writing tasks. For each, produce a prompt template with variables, a filled reference run, and a ten-case test set with rubrics and a recorded pass rate. Deliver as a folder someone in that role could use on Monday.
- Milestone 1: Choose and draft
- Milestone 2: Reference and test
ChatGPTClaude
Learning path · Beginner · 5h
Prompts that hold up
From a one-line request to a reusable prompt with variables, tests and a maintenance routine.
- Write prompts that produce consistent, checkable output for a defined task, and maintain them as models change.
- Skills: Structuring a prompt: role, context, steps, format, constraints, Using examples well, Writing variables and a test set, Diagnosing bad output
Time 5hChatGPTClaude
Cheat sheet · Beginner · 5 min
Agent building blocks cheat sheet
The parts of a production agent and the rule for each. 4 sections, 16 entries.
- Prompt
- Tools
- Control
Entries 16ClaudeChatGPT
Checklist · Beginner · 10 min
Agent evaluation before release checklist
What an evaluation run must show before an agent is released or updated.
- An agent update without an evaluation is a guess deployed to users.
- ☐ A fixed test set of thirty or more inputs including refuse cases
- ☐ Expected outputs or a rubric per case
Items 10ClaudeChatGPT
Guide · Intermediate · 10 min
Agent architecture choices
The four architectural decisions every agent forces and how to make them for a first production agent.
- Agent frameworks differ in surface but force the same four decisions: how the loop ends, how tools fail, where state lives, and who approves side effects. This guide makes each decision for a first production agent and explains the trade-off.
- How the loop ends
- How tools fail
ClaudeChatGPT
Playbook · Advanced · 5h
Multi-step agent with approvals: playbook
Implement an agent that plans multi-step work and pauses for a person before any side effect.
- Multi-step tasks completed with every side effect approved by a person.
- Step 1: Separate planning from acting
- Step 2: Propose, do not act
Steps 5Deliverables 3ClaudeChatGPT
Playbook · Intermediate · 2h 30m
Add tool calling to an existing assistant: playbook
Give a chat assistant one real tool — a database lookup — safely, with validation and logging.
- The assistant answers account questions from live data without ever writing to it.
- Step 1: Choose one read-only tool
- Step 2: Validate the arguments
Steps 5Deliverables 3ClaudeChatGPT
Playbook · Advanced · 5h
Build a research agent with tools: playbook
Implement a read-only research agent that searches, reads and reports with sources, within a budget.
- A report with sources on any question, under a fixed cost, with no side effects.
- Step 1: Write the system prompt as a contract
- Step 2: Define the tools strictly
Steps 5Deliverables 3ClaudeChatGPT
Tool · Beginner · 3 min
JSON formatter
Validate and pretty-print a JSON payload before you use it.
- Pretty-prints, minifies or sorts keys in any JSON payload and reports the first syntax error with its position.
ChatGPTClaude
Tool · Beginner · 3 min
SQL formatter
Make a long query readable before it goes into review.
- Upper-cases keywords and breaks a query into one clause per line so reviewers can read it.
ChatGPTClaude
Checklist · Beginner · 10 min
Before sharing an assistant-produced analysis checklist
The checks between a generated number and a decision.
- A confident wrong number travels fast. Run this before sending.
- ☐ The definition of each metric is written down
- ☐ The query was run by you and the code is saved
Items 10ChatGPTClaude
Code recipe · Advanced · 15 min
Cohort retention table in SQL
Compute monthly retention by signup cohort in one query.
- Retention by cohort is the chart everyone wants and the query everyone gets subtly wrong; this version counts each subscriber once per month.
Language SQLChatGPTClaude
AI tool guide · Beginner · 20 min
ChatGPT data analysis for AI data analysis
Use the data analysis feature to explore a file with code you can read, not answers you must trust.
- Uploading a CSV to ChatGPT's data analysis mode lets it write and run Python to profile, chart and summarise the data; the code it runs is visible, which is what makes the output checkable.
- Workflow: First look at a file
- Workflow: A chart with the code
ChatGPTClaude
Automation recipe · Intermediate · 45 min
Anomaly detection on a daily metric: automation recipe
Build anomaly detection on a daily metric as a n8n automation: post an alert when a daily metric moves outside its normal range, with the recent values attached.
- Trigger: Schedule: daily at 07:00.
- 1. Postgres
- 2. Code
Platform n8nActions 4ChatGPTClaude
Workflow · Beginner · 20 min
Weekly metrics narrative workflow
Write the narrative for a weekly metrics email from computed numbers, never from raw ones.
- When: Monday morning.
- 1. Compute in the sheet
- 2. Narrate from the table
Steps 4Tools Spreadsheet, Any assistantChatGPTClaude
Workflow · Intermediate · 1h
Data quality audit workflow
Audit a table for the problems that silently corrupt analysis and fix the definitions first.
- When: Before any analysis on a table you have not used in a month.
- 1. Profile each column
- 2. Check the joins
Steps 4Tools Any assistant, Database accessChatGPTClaude
Workflow · Intermediate · 45 min
Question to query to chart workflow
Answer a business question with a query and a chart, with the assistant writing and you verifying.
- When: A stakeholder asks a question the dashboards do not answer.
- 1. Define before querying
- 2. Write the query with a check
Steps 5Tools Any assistant, Database access, SpreadsheetChatGPTClaude
Checklist · Beginner · 10 min
Assistant-written SEO content before publishing checklist
The checks that keep generated pages from being thin, wrong or duplicated.
- Generated pages are easy to make and easy to make badly. This list is the difference.
- ☐ The page answers the searcher's questions in order
- ☐ It contains something the top results do not
Items 10ChatGPTClaude
Code recipe · Intermediate · 15 min
Sitemap and canonical audit script
Crawl your sitemap and report pages whose canonical points elsewhere or that return errors.
- A sitemap listing pages that canonicalise elsewhere or 404 wastes crawl budget and hides real problems; a small script finds them in minutes.
Language JavaScriptChatGPTClaude
AI tool guide · Beginner · 20 min
Google Search Console for AI SEO
Use Search Console as the source of truth for what actually ranks and gets clicks.
- Search Console shows the queries your pages appear for, impressions, clicks and positions, plus indexing and page-experience issues; every SEO decision should start from its data rather than from a third-party estimate.
- Workflow: Find near-misses
- Workflow: Indexing check
ChatGPTClaude
Automation recipe · Advanced · 45 min
Ranking drop alert with diagnosis: automation recipe
Build ranking drop alert with diagnosis as a n8n automation: when a tracked page drops more than five positions, fetch the current results and post a diagnosis.
- Trigger: Schedule: daily at 06:00 reading the rank tracker export.
- 1. HTTP Request
- 2. Code
Platform n8nActions 5ChatGPTClaude
Workflow · Intermediate · 45 min
Structured data for a page type workflow
Add valid, honest structured data to a page type and verify it.
- When: A new page type launches or an existing one lacks schema.
- 1. Pick the type
- 2. Generate from real fields
Steps 4Tools Any assistant, A schema validatorChatGPTClaude
Workflow · Intermediate · 1h
Monthly decay review workflow
Find pages losing traffic each month and decide refresh, merge or leave.
- When: First week of the month.
- 1. Rank the losers
- 2. Diagnose each
Steps 4Tools Analytics export, Any assistantChatGPTClaude
Workflow · Intermediate · 1h 30m
Keyword to brief to draft workflow
Move from a target query to a searcher-led brief to a draft, with the results page as the evidence at each step.
- When: A query is chosen for a new page.
- 1. Read the results, not the keyword
- 2. Write the brief
Steps 5Tools Any assistant, Search results, AnalyticsChatGPTClaude

Search