Show workspace bootstrap context in Codex prompt snapshots#76302
Show workspace bootstrap context in Codex prompt snapshots#76302pashpashpash merged 1 commit intomainfrom
Conversation
|
Codex review: needs maintainer review before merge. Summary Reproducibility: yes. Current main has a focused Codex app-server test for forwarding non- Next step before merge Security Review detailsBest possible solution: Keep the PR open for normal maintainer review and CI, preserving the narrow snapshot-only representation of the already-shipped Codex bootstrap forwarding contract. Do we have a high-confidence way to reproduce the issue? Yes. Current main has a focused Codex app-server test for forwarding non- Is this the best way to solve the issue? Yes. For this PR’s goal, extending the existing prompt snapshot generator through the Codex test API What I checked:
Likely related people:
Remaining risk / open question:
Codex review notes: model gpt-5.5, reasoning high; reviewed against dd43caa27aa1. |
The Codex runtime prompt snapshots now show the model prompt, OpenClaw instructions, user input, and dynamic tools. One missing piece was OpenClaw's workspace bootstrap context: after the recent Codex bootstrap forwarding work, files like SOUL.md, TOOLS.md, and HEARTBEAT.md travel through Codex config.instructions, but the happy-path fixtures did not make that layer visible.
This PR adds dummy workspace bootstrap files to the snapshot generator and renders the resulting config.instructions layer in the reconstructed model-bound prompt. It also records the simulated file paths in snapshot metadata and asserts that the generated Telegram snapshot includes the bootstrap placeholders. AGENTS.md stays documented as native Codex-owned context rather than being duplicated in these fixtures.