Follow-up cards prematurely stop autonomous Freebuff runs and waste remaining session time

[Walentalien]

What happened

Freebuff can stop an autonomous run by emitting "Suggested followups" and waiting for user input even when:

• the original requested task is not finished;
• actionable work clearly remains;
• I explicitly asked it to continue autonomously without stopping for follow-ups; and
• substantial session time remains.

This is especially problematic in Freebuff because sessions are time-limited.

For example, I can ask the agent to work continuously through a repository, finish one task, immediately move to the next task, and not stop for follow-ups. It can still finish an intermediate step, show several "Suggested followups", and end the turn.

If I do not notice immediately, the remaining session time is effectively wasted while the agent is idle.

Repeating instructions such as:

"Do not stop for follow-ups."
"Continue until the session ends."
"When one task is complete, immediately start the next actionable task."

does not reliably prevent the behavior.

The codebase also appears to already recognize this failure mode.

agents/base2/base2-free-fable.ts uses:

noFollowups: true

with the comment:

// Live runs ended on follow-up cards without a written answer. Keep
// questions and delegation, but omit the optional cards and their prompt.

Meanwhile createBase2() supports noFollowups, but other Freebuff agent variants can still receive the suggest_followups tool.

Follow-up suggestions are useful for interactive conversations, but they work against explicitly autonomous/time-limited sessions.

I think there should be a supported way to prevent follow-up cards from terminating this kind of run.

Possible solutions could be:

• an Autonomous / Continuous mode;
• a user setting to disable follow-up suggestions;
• exposing the existing noFollowups behavior;
• or preventing suggest_followups from ending a turn while unfinished TODOs/actionable work remain.

Steps to reproduce

1. Start a Freebuff session in a repository with multiple independent actionable tasks.

2. Give the agent an explicit continuous-work instruction, for example:

   Work continuously through the repository until the session ends.
   Do not stop to ask for follow-ups.
   When one task is finished, immediately continue with the next actionable task.

3. Let the agent complete an intermediate task.

4. Despite the instruction above, the agent can stop and display:

   Suggested followups:
   ....

5. The suggested items are actions the agent could execute itself, but it waits for another user message instead.

6. The Freebuff session is still active / has remaining time, so the unused part of the time-limited session is wasted while the agent waits.

Expected: for an explicitly autonomous task, the agent should continue with actionable work instead of terminating the turn solely to show optional follow-up suggestions.

Actual: the turn can end on "Suggested followups" even though unfinished work remains.

Where does this happen?

Agents / prompts / agent behavior

Operating system

Linux

Version

0.0.201

Model

not relevant

Logs or screenshots

not relevant