agent: V2 base prompt has no question-tool guidance, so agents ask in prose
Nobody has claimed this yet.
Assessment
- Difficulty
- 2/5
- Estimated time
- 1-3 hours
- Newbie friendliness
- 70/100
- Issue type
- Bug
- Clarity
- Clearly specified
- Activity status
- Active
- Tech stack
- typescript
- Domain
- ai-infra-agents
Research direction
Find the V2 base system prompt in the codebase, likely in a file defining agent prompts or tool usage policies. Compare it with the V1 prompt to locate the missing guidance about the question tool. Add a line instructing the model to use the question tool for user decisions. Test by running a plan or build agent and verifying that end-of-turn questions now trigger the tool call.
Written by the indexing model from the issue text.
Description
Description
OpenCode V2's base system prompt never mentions the question tool, while the V1 prompt told the model to use it. Plan/build agents therefore end turns with prose question lists instead of the interactive form, even though the tool is available, permitted, and works in the desktop client. The tool's own schema description only covers asking "during execution", so end-of-turn decisions (e.g. "which workstream should start?") are asked as text.
Plugins
None related.
OpenCode version
2.0.15 (desktop app; also checked 2.0.9, 2.0.12, 2.0.14)
Steps to reproduce
- Use a
planorbuildagent with a model that relies on explicit prompt cues. - Ask for a plan that ends with a decision only the user can make.
- Observe the final message: the questions are prose; the
questiontool is never called.
Expected behavior
The agent asks through the question tool, as the V1 system prompt instructed: "When the user's request is vague, use the question tool to clarify before reading files or making changes."
Actual behavior
Prose questions, no tool call. Extracting the embedded prompts from opencode-cli.exe:
- V1 prompt,
# Tool usage policy: contains the question-tool line above. - V2 base prompt (7,854 chars): zero mentions of the question tool. The only in-band cues are the tool's schema description ("...questions during execution...") and the plan-mode reminder ("Discuss the plan with the user directly in the conversation.").
The tool itself is healthy: it is registered with an allow rule for plan/build (deny for built-in subagents), and calls render and complete in the desktop client.
Additional context
- Suggested fix: restore a line to the V2 base prompt, e.g. "When a decision is genuinely the user's — a direction choice, a plan-review fork, a confirmation that gates the next step — ask through the
questiontool instead of ending the turn with a prose question list." - This is a prompt regression in V2, not a client-version regression: all four shipped CLI versions checked contain both prompts.
Operating System
Windows 11 (10.0.26200, win32 x64)
Terminal
Desktop app
- Dominant language
- TypeScript
- Stars
- 209k
- Forks
- 27.5k
- Avg merge
- 9h 42m
- Merged PRs (30d)
- 389
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from anomalyco/opencode
-
Difficulty 2/5 1-3 hours Newbie friendliness 75/100
-
Difficulty 2/5 1-3 hours Newbie friendliness 75/100
-
Difficulty 2/5 1-3 hours Newbie friendliness 75/100
-
Difficulty 2/5 1-3 hours Newbie friendliness 75/100
-
Difficulty 2/5 1-3 hours Newbie friendliness 75/100
All issues in anomalyco/opencode
Similar issues
-
bug(cli): hapi doctor inline-media prints a fabricated B:\ helper-script path in packaged installs Open
Difficulty 2/5 1-3 hours Newbie friendliness 70/100
-
Crush Open
Difficulty 1/5 Under an hour Newbie friendliness 85/100
catppuccin/catppuccin#3125 ·
-
Add a SECURITY.md Open
Difficulty 1/5 Under an hour Newbie friendliness 90/100
ElementsProject/cln-application#167 · 1 comment · 1 reaction ·
-
Difficulty 2/5 1-3 hours Newbie friendliness 75/100
Quantco/pnpm-licenses#17 ·
-
Difficulty 2/5 1-3 hours Newbie friendliness 75/100