share what ditto found in your logs

[ohad6k]

if you run emulo, drop one surprising thing it found about how you work.

no secrets, no private code, no full you.md needed.

use this shape:

agent: claude / codex / cursor / other
sessions mined: <number>
most accurate trait:
most uncomfortable trait:
one thing your agent should now stop doing:

i'm especially curious whether it finds things you never wrote in CLAUDE.md, AGENTS.md, cursor rules, or custom instructions.

[theconsultant]

That was fun.. and pretty accurate!

Agent: Claude Code
Sessions analyzed: 147 (~2,300 messages, ~4 months)
Most accurate trait: Manual-publish is trust-gated — I let the agent prep everything (throttle deletes, draft the CSV, snapshot DNS) but stop it at the send/write step and do that by hand, until a specific pipeline earns enough trust that I let it write directly. Never wrote this down anywhere.
Most uncomfortable trait: "Automated almost everything except the one step where he'd have to trust it… the bottleneck isn't the work, it's that a single misread scope once cost him enough that he now pays a permanent tax in dry-runs and manual sends to keep reality observable."
One thing the agent should stop: Re-deriving data a source of truth already owns, and paraphrasing when I asked for the actual list ("show me the list" → "show me an actual list" is in my logs verbatim).

[ohad6k]

This is exactly the kind of thing I hoped it would catch. The trust-gated manual-publish step is a real working pattern, not a preference you'd ever think to write into CLAUDE.md. That's the whole point, it reads what you did, not what you remembered to document.

The "show me the list" to "show me an actual list" verbatim catch is my favorite part. It found the exact moment the pattern fired in your own words.

If you want to trade notes on what it's pulling from people's logs, we started a discord: https://discord.gg/VNnMq2U5r. Would love to see more of these.