Skip to content

Feat/interactive agent questions#29

Merged
clintberry merged 2 commits into
mainfrom
feat/interactive-agent-questions
Jun 9, 2026
Merged

Feat/interactive agent questions#29
clintberry merged 2 commits into
mainfrom
feat/interactive-agent-questions

Conversation

@clintberry

Copy link
Copy Markdown
Contributor

No description provided.

The Pi runtime sent only the bare task text; the agent's configured
system_prompt (persona + 'prefer ask_user when blocked' guidance) was dropped
— only the legacy claude executor applied it. Thread system_prompt through
Ensure → Launch and pass it via --append-system-prompt, carrying the text in
the DEUCE_SYSTEM_PROMPT env var (devpod --set-env) so arbitrary content needs
no shell quoting. Lookup failure is non-fatal (falls back to no prompt).
Every agent on the Pi path now gets a global base system prompt
(DefaultBaseSystemPrompt) prepended to its own system_prompt, instructing it to
call the ask_user tool when blocked on a human decision instead of guessing or
asking in prose — which is what surfaces the interactive typed prompt. Applied
automatically to all agents (no per-agent seeding); overridable via
DEUCE_AGENT_SYSTEM_PROMPT.
@clintberry clintberry merged commit 07c162e into main Jun 9, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant