Dynamic system prompt templating #3202
Open
+369
−3
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
- Add Template utility for processing bash commands in ! syntax - Support file:// URLs in agent prompts for loading external files - Enable dynamic content generation in system prompts using same logic as commands - Add comprehensive tests covering templating, file loading, and error handling - Backward compatible with existing static prompts
- Extend Template utility to support both shell (!) and file (@file) templating - Handle non-git repositories by falling back to Instance.directory when worktree is root - Add comprehensive tests for file references, directory listings, and combined templating - Support tilde paths (~/.file) and graceful error handling for missing files - Enable full command-style templating capabilities in agent system prompts
- Inline file:// URL processing and templating directly in resolveSystemPrompt - Remove unnecessary resolveAgentPrompt helper function and _internal export - Focus tests on Template utility rather than internal implementation details - Maintain full functionality while improving code organization - Clean up test structure and remove code smells
- Remove file:// URL handling as it's already processed elsewhere in the pipeline - Keep only the Template.process call for bash commands and file references - Simplify implementation to focus on core templating functionality
Collaborator
|
This is a good start, I will branch off and change it to be a little more in line with what i had in mind, thanks for PR |
Flux159
commented
Oct 20, 2025
| agent, | ||
| system: input.system, | ||
| }) | ||
|
|
Author
There was a problem hiding this comment.
Just an fyi for your branch, I've been using this locally for a few days with opencode run "" --print-logs and found it useful to add some logs here to help debug system prompts:
log.info(`Agent name: ${agent.name}`);
log.info("Agent parameters", {
name: agent.name,
temperature: agent.temperature,
topP: agent.topP,
options: agent.options,
});
log.info(`Resolved System Prompt: ${system.join("\n")}`);
| })(), | ||
| })()), | ||
| ) | ||
| system.push(...(await SystemPrompt.environment())) |
Author
There was a problem hiding this comment.
There's a separate issue around allowing environment to be optional in system prompts #1894 (saw this when logging) - code pointer should be here
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary:
Adding the ability to parse markdown templating for agent system prompts. Specifically adds ability to parse
!and@references inside of agent prompts.This PR adds a helper util/template.ts that handles both bash & file reference templates, then uses this inside of session/prompt.ts. The rest is just automated testing for the functionality.
This is specifically for #3195
Test plan:
Manually tested by making a custom agent inside of
.opencode/agent/test.mdwith the following contents:Then opening opencode dev mode:
bun dev, switching to the test agent via/agents, then asking "What is your magic phrase" and it responded with myusername-magicas expected.