When the AI gets stuck
As you write, you hit spots where the AI workflow itself stalls. It’s not the post’s content that’s stuck, it’s the flow that stopped. Here are a few you’ll meet often, and the way out.
When the AI-smell gate fails twice
Section titled “When the AI-smell gate fails twice”When the AI-smell gate fails right before publishing, the post gets rewritten automatically. But if it rewrites twice and still fails, the orchestrator stops and hands off to the user. It doesn’t rewrite endlessly.
At this point, read the post yourself once. Usually the same cause repeats. The intro opens with the conclusion every time, every paragraph is a similar length, or first-person experience is empty and got filled with generalities. Touch up one paragraph yourself, or give a concrete direction like “restart the intro with something I actually went through,” and the score drops. Pointing at the stuck cause works faster than a vague “rewrite it.”
When it asks for experience details
Section titled “When it asks for experience details”Before drafting, there’s a spot where the AI stops and asks “what error did you see,” “where did you get stuck.” This isn’t stuck; it’s a deliberately planted spot. Without a concrete failure or data, the AI asks the user rather than inventing experience.
Write down what you remember. If you don’t recall the exact error message, “it was roughly this situation” is enough. That detail builds the post’s trust and clears the AI-smell gate along with it. If you wave it off here with “just write whatever,” you get a post filled with generalities and it gets blocked at the gate again.
When publishing stops at credentials
Section titled “When publishing stops at credentials”If the dry-run worked but the actual publish doesn’t, it’s usually the credentials. The publish script runs only through dry-run if credentials.json or token.json is missing, and reports which file is missing. It doesn’t invent a Blogger URL.
In that case, go back to the OAuth item in prerequisites and the credential step in publish. token.json is created on first publish via browser consent, so if the consent screen didn’t appear, check first that credentials.json is in place.
When a session broke and you don’t know how far you got
Section titled “When a session broke and you don’t know how far you got”If you reopened days later or work doesn’t carry over after a /clear, the session-start hook (session-start.sh) pulls the previous work and recent history automatically. If you’re still confused, ask Claude Code “summarize what I’ve done so far.” It pulls together the briefs and staging posts in _workspace/ and progress.md to tell you how far you’ve come.
When you finish for the day, writing the next starting point into active.md makes this spot much smoother next time. It’s also the spot stop-reminder.sh alerts you about when it’s empty.
If it still doesn’t resolve
Section titled “If it still doesn’t resolve”Copying the error message verbatim, pasting it into Claude Code, and asking “why is this happening?” is often the fastest. If that still doesn’t work, find the same symptom in troubleshooting, and if it’s not there either, open a GitHub issue. Write down which step, which command, the full error, and environment info together.
A collection of common sticking points is in troubleshooting.