conon.ai

conon.ai/Tips

Cap Research With A Deliverable

'Look into this' with no boundary is how an agent reads fifty files and stalls — the docs list boundless research among the five classic failures. Scope every investigation to a named artifact: 'a one-page summary naming the three files to change'. The deliverable is the fence.

ProAGENTSClaude Code

Why it works

'Look into this' with no boundary is how an agent reads fifty files and stalls — the docs list boundless research among the five classic failures. The fence is a named deliverable: 'a one-page summary naming the three files to change' converts wandering into work, because the agent can tell when it's DONE. Research without an output shape has no done.

How to do it

  1. Before any investigation, name the artifact: summary, ranked list, table, decision memo.
  2. Cap the input: 'check these two directories', 'the last month only', 'top 10 by size'.
  3. Give it a stop condition: 'enough to recommend one option, not to be exhaustive.'
  4. Phrase it as production, not exploration: 'produce X from Y' beats 'explore Y'.
  5. Received the artifact? Follow-up rounds get the same treatment — new artifact, new cap.
  6. For big investigations, route through a subagent AND name the deliverable — clean context and a fence.

Copy this

prompt
Investigate why the export step is slow. Deliverable: a one-page summary with the top 3 suspected causes ranked, the evidence for each, and the single file you'd change first. Look only in the pipeline folder. Stop when you can rank three.

What to watch for

  • 'Be thorough' is an anti-instruction for research — thoroughness without a deliverable is precisely the failure mode; say what DONE looks like instead.
  • An artifact too big ('a full report') recreates the problem in the output; one page is a feature, not a limit.

Sources

Original: conon.ai

Read next

No Verification, No Green Light
The docs' hardest rule: if you can't verify an output yourself, don't let the agent proceed until IT has. Make self-checking part of done — run it, screenshot it, list what was checked. For anything client-facing, 'it looked right' has never once been a defense.
Claude Code
Tell Compaction What To Keep
When a long conversation gets squeezed, the agent decides what survives — and may drop exactly what you needed. One line in CLAUDE.md fixes it forever: 'when compacting, always preserve the list of changed files and my requirements'. Works in every session, no reminders, and /compact accepts the same guidance inline.
Claude Code
Compact Early, On Your Terms
Auto-compaction fires when the window is nearly full — which is after quality already sagged, and the automatic summary guesses at what mattered. Compact earlier and by hand, or ask the agent to lower its auto-compact threshold, so the squeeze happens while the session is still sharp.
Claude Code
Artifacts Demo It, Hosting Ships It
'Publish this as a shareable page' gets you a live URL in a minute — it republishes to the same address as you iterate, which is perfect for showing a client work in progress. Know the box you're in: one page, no backend, 16MB, private or public. Forms, logins, and payments are your cue to move to real hosting.
Claude Code

We build and operate the systems behind tips like this one — agents, automations and internal tools, shipped into production. See what the studio does →