Stacks Documentation
SkillsChangelog
TeamSponsorsPartnersPostcardwareAwesome StacksContributing

Investigate

Copy page
Copy page as Markdown for LLMs
View as Markdown
View this page as plain text
Open in ChatGPT
Copy content and open ChatGPT
Open in Claude
Copy content and open Claude
Open in Perplexity
Search with Perplexity AI
On this page 5

stacks-investigate · Engineering craft · model-invoked

Root-cause debugging, structured so the hard part comes first. No hypothesis is allowed until you have a tight feedback loop, one command you have already run that goes red on this specific bug. Then minimise the repro, rank three to five falsifiable hypotheses, instrument one variable at a time, and lock the fix down with a regression test.

When to reach for it

  • Something broken
  • Throwing
  • Failing
  • Flaky
  • slow. Builds a tight feedback loop that goes red on the bug before any hypothesis is allowed
  • Then minimises
  • Tests hypotheses
  • Fixes and locks it down with a regression test. Enforces no fixes without root cause

Inside the skill

The sections an agent reads once the skill loads.

  • Redact
  • Phase 1: build a feedback loop
  • Phase 2: reproduce and minimise
  • Phase 3: hypothesise
  • Phase 4: instrument
  • Phase 5: fix and regression test
  • Phase 6: cleanup
  • Rules

Supporting files

Reference and scripts the skill reaches for on demand, rather than loading up front.

Using it

This one is model-invoked. Your agent reaches for it on its own when the task matches, and you can also call it by name:

/stacks-investigate

Source: stacks-investigate/SKILL.md. Shadow it for one project with app/Skills/stacks-investigate/SKILL.md, then re-run buddy setup:ai. See Writing your own.