ai-coding-minesIndexGitHub

Claude Code · AI Agents

15 entries

Traps that appear when you delegate work to an agent. Most take the form of what the agent reported doing differing from what is actually left behind.


A hypothesis without a control group was always wrong

Case A — after deploy, only the new route 404s. Five hypotheses (keystore, source maps, manifest, infra outage, lost commit) were all assertions beyond the observed range, and all

Agent reporting discipline

Two things kept collapsing.