Configure and run Situational Awareness

Edit the SA Directive, read the per-agent uncertainty badge, interpret the wave situation report, and handle fixation warnings.

Situational Awareness is on by default. This page covers what you can control and what you see: the SA Directive, the situation summary added at the start of each run, the per-agent uncertainty badge, the warnings passed to later agents, the wave situation report, and the fixation warning.

The SA Directive

The SA Directive is a short instruction given to every agent, alongside the Wow Directive. It tells the agent to stay aware of the situation throughout the task instead of rushing to the finish.

Toggle and edit: Settings → Workflow → SA Directive. The toggle defaults to enabled. The text is editable.

The default directive (restored when you clear the field while enabled) is:

Maintain situational awareness throughout this task. Before claiming completion, ask: what changed since I started, what do I actually know vs. assume, what's likely next, and what's still unknown? Name uncertainty explicitly — prefer 'I don't know yet' over fabrication. Attribute every load-bearing claim to its source (a file you read, a tool result you observed, an instruction from the user, or your own inference). Surface gaps and contradictions instead of papering over them.

Why it is separate from the Wow Directive: Wow asks is this output worth showing off?. SA asks does this output honestly say what the agent doesn't know?. Keeping them apart lets you tune one without losing the other.

Where it is saved: with the app, not in the workflow file. It applies to every workflow on this computer.

The situation summary at the start of a run

When a run starts, Cerevisor adds a short summary of the current situation under the directive. Your own text stays untouched; the summary is worked out fresh for every run and is not saved. It names:

  • The current state label and how fresh Cerevisor's picture is.
  • Up to three unknowns currently on file.
  • How many disagreements between you and Cerevisor are open.

So every agent starts with the actual situation, not a generic reminder. The same summary appears in the System Log (a "Run-start situation" card) and as a chip in the run window. If Cerevisor cannot put the summary together, agents get the plain directive. Situational Awareness never blocks a run.

The per-agent uncertainty badge

When an agent finishes, Cerevisor shows how sure that result is as the Agent Situation Badge on the agent's card. It brings together:

  • How confident the agent said it was, and the one-line reason it gave.
  • Whether Cerevisor's check of the output passed it, rejected it, or did not run, and why.
  • Short tags such as "no confidence marker", "verifier rejected" or "high confidence".

Badges that deserve your attention: any verifier rejected result, any agent that finished with no confidence marker at all, and several low-confidence results on the same role.

The badge resets at the start of each new run.

Warnings passed to later agents

Uncertainty is not only shown to you: it changes what later agents are told. When an agent's result was rejected by the check, or the agent itself reported low confidence, every agent that builds on its output gets a one-line warning to verify important claims first, for example "Researcher" self-reported 0.40 confidence (thin evidence). Verify load-bearing claims before building on them. Healthy results carry no warning, so the signal stays rare and meaningful.

The Wave Situation Report

After every wave finishes, Cerevisor writes a Wave Situation Report and shows it in the run window between waves.

The report covers:

  • Which wave it was and when it finished.
  • Each agent: its status (completed, failed or skipped) and how sure its result is.
  • Files produced this wave, grouped by agent.
  • Files the wave was expected to produce but didn't, so you can see gaps before later waves build on them.
  • Notes: what changed, what is still unknown, and what the next wave should pay attention to.

At the top, the report counts the agents that completed, failed and were skipped, the agents that said they were confident but did not pass the check, and the agents that never said how sure they were.

The report is also passed to the next wave's agents, so each wave starts from the same read of what is done and what is missing.

Fixation warnings

During Loop & Enhance, Cerevisor compares each round's plan with earlier rounds. When a round repeats an earlier one, it raises a fixation warning. The Situational Awareness panel lists recent warnings, newest first, with the round that repeated and the earlier round it matched.

A fixation warning changes the state label to Fixation risk. The chat builder takes that as a cue to suggest a change of approach instead of another round.

What resets between runs

When a new run starts, the live picture is cleared: the uncertainty badges, the wave situation reports, the fixation warnings and the run-start chip (which is then filled in again for the new run).

The SA Directive itself, its text and its toggle, is not affected.

Where to go next

Back to docs