A pasted secret is a burned secret
grep -rlE "sk_live_| AKIA[A-Z0-9]{16}| ghp_[A-Za-z0-9]{36}" ~/ . claude/ projects/ 2>/ dev/ null | wc -lGet the free fixes for code AI gets wrong
Each one shows what the AI wrote, what it should have written, and how to check your own app
What is going on
Chat transcripts persist outside your security boundary. Claude Code stores every session as plaintext JSONL on disk, synced wherever your home folder syncs. Commits co-authored by a coding agent leak secrets at about twice the baseline rate. The window is not private because it feels private.
Where it bit
Four real tokens pasted into an AI chat while debugging, two of them twice. Three were rotatable. One was not: a legacy API app the provider no longer issues, so rotating it means losing the capability for good. The sharper lesson was not "do not paste," it was: know which of your credentials are replaceable before you leak one.
The practice
Refer to secrets by env-var name in chat, never by value. Treat anything pasted as compromised and rotate. Keep an inventory with a column for "re-issuable." Shorten session retention with cleanupPeriodDays and deny .env reads in the agent's permissions.
Get this check as a script you can run tonight

Make your coding agent prove it didn't cheat
Coding agents special-case, skip or edit tests so the check goes green. A five-sentence prompt Anthropic tested took Claude Opus 4 from cheating in 47% of runs to 5%- 3 min read
How to pick who fixes your vibe-coded app
Your app was built fast with AI and now something is wrong. Here is who can fix it, what they should do before they quote, the questions to ask and the red flags to walk away from - 3 min read
What an app check covers
An app check is an engineer reading the app your AI tools wrote, ranking what will break and writing the fix for each problem. Here is every area it covers, how the five days run and what you keep at the end
If this check came back with more than you expected, that is worth a conversation