Things that are "obvious" but easy to get wrong. Check here before thrashing.

• LONG-RUNNING PROCESS never exits → don't wait for it (see workflow-long-running-processes).
• INTERACTIVE PROMPT (apt, npm init, ssh host-key, pip on conflict) hangs waiting for input → use non-interactive flags (-y, --yes, --no-input, DEBIAN_FRONTEND=noninteractive) or echo input.
• WRONG WORKING DIR: each terminal tool call may reset cwd. Always cd into the project, or use absolute paths. If a file "doesn't exist," check pwd first.
• VENV NOT ACTIVE: pip install then python can use different environments. Use the venv's python explicitly, or activate in the same command chain with &&.
• PORT ALREADY IN USE: a failed prior run left a server bound. ss -ltn | grep <port>, kill the old PID before restarting.
• TEST THAT NEEDS A RUNNING SERVER: start server backgrounded FIRST, then run the test, then kill the server. Don't run the test against nothing.
• A COMMAND THAT PRINTS NOTHING may have succeeded (many unix tools are silent on success). Check the exit code ($?), don't assume silence = failure and retry.
• BUILD/INSTALL IS SLOW not hung: npm install, cargo build, pip compile can take minutes. Give them a real timeout (300s+), don't kill at 30s and retry.
• INFINITE RETRY: if the same command failed twice identically, the THIRD identical attempt will also fail. Change something or invoke workflow-stuck-detect-reset.