[FEATURE]: Include system prompt in export #39033

Activity

1. github-actions

   assigned

   jlongster

   on Jul 27, 2026
2. github-actions commented on Jul 27, 2026

   

   github-actionsBot

   on Jul 27, 2026 – with GitHub Actions

   Contributor

   More actions

   This issue might be a duplicate of existing issues. Please check:

   • [FEATURE]: Save dynamically generated system prompt to opencode.db so it's included in transcript exports #26376: [FEATURE] Save dynamically generated system prompt to opencode.db so it's included in transcript exports (closed, very similar — about persisting and exporting the system prompt)
   • [FEATURE]: Add TUI ability to inspect session system messages #24990: [FEATURE] Add TUI ability to inspect session system messages (open, related — about viewing/inspecting the system prompt during a session)

   Please review these issues and let us know if your request is distinct or if it should be linked to one of them.

   👎React with 👎1Reacted by Jack Stringer
3. Rexarrior commented on Jul 31, 2026

   

   Rexarrior

   on Jul 31, 2026

   More actions

   I opened a narrower PR that may help with part of this use case: #39905

   It does not persist system prompts into session exports and does not change transcript storage. Instead, it adds a lightweight local CLI debug command:

   opencode debug prompt

   The command prints the resolved effective system prompt for the current project/agent/model, including the model base prompt, environment block, instruction files, MCP instructions, available skills, and experimental.chat.system.transform plugin modifications.

   This is intended as a smaller debugging primitive for verifying what prompt layers are active, without adding DB persistence or a full context/request dump.
4. jack5github commented on Sep 29, 2026

   

   jack5github

   on Sep 29, 2026

   Author

   More actions

   This issue is not yet resolved. Even after switching to OpenCode v2 and using the export functionality provided in both the CLI and Desktop application, full visibility of what is being sent to AI models remains a problem. Users have no way of knowing what exactly is being sent to models other than the messages that are visible to them in the TUI (which are all that is included in exports).
5. opencodejr commented on Oct 28, 2026

   

   opencodejrBot

   on Oct 28, 2026

   More actions

   Thanks for the suggestion, @jack5github! You're asking for session exports to include the system prompt the model was actually given, so it's easier to see what context the model had and roughly how many tokens it used.

   Right now opencode session export only writes out the saved session messages. The system prompt is built separately for each request and isn't saved, so it can't appear in the export. Maintainers will consider this.

   Related issues you may want to follow:

   • desktop: Context tab hides system prompt or displays a system update instead #55046: the desktop Context tab's "System prompt" section shows the last saved system message, not the full system prompt sent to the model (same underlying gap)
   • [FEATURE]: split token usage inside a step (system, schemas, each tool result, reasoning, output) #58716: request to break down token usage within a step (system prompt, tool schemas, tool results, etc.)
   • cli: session export only supports JSON — add --format md #59904: request for more session export formats

   👍React with 👍1Reacted by Jack Stringer
6. Rexarrior commented on Oct 29, 2026

   

   Rexarrior

   on Oct 29, 2026

   More actions

   For reference, I already implemented a narrower debugging tool in #39905: opencode debug prompt prints the resolved system prompt for the current project, agent, and model. It does not persist per-request prompts or include them in session exports, but may be useful as a starting point. The PR was closed by automated age/reaction cleanup, not after a technical rejection.