Fulcrum 8 of 18
When the agent is about to stop and ask
The moment the agent thinks 'I should check with the person', whether mid-task, at an ambiguity, or before a decision.
What goes wrong here
There are two opposite failures. In one, the agent stops to ask something it could have found out by looking (a file, the git log, a past decision), which hands its thinking back to the person. In the other, it asks a bare question stripped of the context that would make it answerable at a glance. Less often, it pushes past a question that really did need the person.
How it compounds if nothing catches it
Needless questions turn the person into the agent's manager. Every stall costs their attention and teaches them to hover. Bare questions get bare answers, which the agent then over-reads. Real questions that get skipped produce confident work in the wrong direction.
What the harness does here
ask gates every question. The agent sets out the context the question comes from, predicts the answer with a confidence number, and scores how much the question matters. Low-stakes questions are decided and announced. Only high-stakes, uncertain ones go to the person, with the prediction attached so they can answer in a word. on-agent-getting-stuck tells a real need for the human apart from a stall the agent should predict its way through. jonathan-check2 predicts how the person would respond, using a notebook of thousands of their past exchanges with agents. jonathan-check3 is a planned, larger version of it. A hook blocks Claude Code's pop-up question box and sends the agent to ask instead, because the pop-up shows the question without the context that makes it answerable.
The pieces that act here
- /ask
Use BEFORE asking the person any clarifying question. Replaces naked questions with a criticality-gated decision flow: synthesize the context the question emerges from, predict the likely answer with a confidence number, score how much it matters using the governer score already run for the session, then route based on the combined criticality. Low-criticality questions get decided by the agent and announced. Medium ones get an institutional-memory pass before reasking. High-criticality ones get the full briefing template and wait for a go-ahead.
- /on-agent-getting-stuck
Use the moment you find yourself thinking "I should stop and ask the user." That thought is common and often wrong. This skill is the check that tells the difference between a genuine need for the human and a stall you should predict your way through. Triggers — scope feels ambiguous mid-task, you're unsure what the user would prefer, you're about to ask a clarifying question, you're about to hand work back unfinished, you feel blocked, you want to pause and reconfirm.
- pre-tool-askuserquestion-gate.sh
A hook: a script Claude Code runs on its own at a set moment.
- /jonathan-check2
Response predictor v2 — queries an oracle built from real agent-conversation history to predict what the person you're working with would say to your output. Replaces a local keyword-matching predictor with semantic retrieval and synthesis over that history. Use before declaring work done, before presenting work, or when you need to anticipate their reaction.
- /jonathan-check3
God-mode reaction oracle — queries a notebook built from the person's own extracted session data (coherence checks, reflect outputs, align outputs, skill sequences, compaction intents, scope declarations). Predicts what the person you're working with would say AND what the next step should be. Loops autonomously until the oracle signals scope is complete. Everything printed observably.
- /predict-next-action2
Predict the next highest-leverage action by enriching session context through your own institutional-knowledge oracle (if you've set one up), then producing a nuanced, UX-focused, intent-grounded recommendation. Use when finishing a session and wanting a deeply informed next-step prediction.