/TASK GROUPS
What do you want
your AI to do?
Small, ordered workflows with a clear result. Each group explains why its instructions belong together.
Use the Markdown with Codex, Claude, open-weight models or another AI. Each task lists the tools and setup it needs.
Find your setup/CHOOSE YOUR TASK
Instructions that work together.
Pick an outcome. Get the inputs, the steps in order, and a way to check the result.
Monitor a source and explain what changed
Build a baseline, detect meaningful changes, and produce an updated dashboard with an alert decision.
See the workflow4 CONNECTED STEPSGive an AI a usable project memory
Compile supplied documents into a source-linked wiki and retrieve the right evidence for a question.
See the workflow5 CONNECTED STEPSBring a source into an AI data product
Implement ingestion whose records survive bad refreshes and are actually queryable by the intended AI consumer.
See the workflow5 CONNECTED STEPSProduce reports whose claims trace to data
Generate a report or chart-backed answer using computed facts and auditable relationships.
See the workflow5 CONNECTED STEPSBuild consistent branded emails
Turn approved content into a branded email preview with checked layout, links and factual content.
See the workflow5 CONNECTED STEPSMake an AI coding change with working checks
Map a requested change, implement its controls and prove the actual user path handles the target failure.
See the workflow3 CONNECTED STEPSProve an application can recover
Produce a recovery runbook backed by an isolated restoration and successful application writes.
See the workflow5 CONNECTED STEPSRepair a listing value without losing the correction
Prepare and execute a scoped repair, verify the rendered result, and prove it survives upstream synchronization.
See the workflow4 CONNECTED STEPSPublish exactly the content that was approved
Build a publication workflow in which corrections remain reviewable and delayed publishers cannot restore obsolete or revoked content.
See the workflow4 CONNECTED STEPSAudit historical experiments without hiding failed trials
Produce an experiment audit that preserves every trial and tests whether historical signals used only available information.
See the workflow4 CONNECTED STEPSRetrieve evidence and diagnose empty results
Choose a suitable retrieval route and distinguish missing evidence from broken extraction or ignored filters.
See the workflow4 CONNECTED STEPSKnow whether an automated job really worked
Instrument an existing job so missed runs, invalid output, resource problems and unresolved alerts are visible separately.
See the workflow