fix: correct architect use of plan exit to save the plan file correctly

[johnnyeric]

Issue

N/A

Context

Fix CLI plan follow-up after plan_exit for Architect/plan-style flows. The plan file was saved successfully, but the CLI could stop without showing the “Ready to implement?” prompt when the follow-up needed to read the saved plan file.

Implementation

Bind the Kilo-owned plan follow-up entrypoint before running the async follow-up flow. This preserves the active instance context across the Effect.promise boundary, so PlanFollowup.ask can still resolve the saved plan path when it falls back to reading the plan file from disk.
Kept the fix in Kilo-owned code instead of changing shared opencode logic.

Screenshots / Video

How to Test

Reviewer test steps

• Tested different scenarios on the architect agent to build a plan and continue implementation on a new session
• Tested CLI and Extension

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 45b1a88153

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[kilo-code-bot]

Code Review Summary

Status: 1 Issue Found | Recommendation: Address before merge

Overview

Severity	Count
CRITICAL	0
WARNING	0
SUGGESTION	1

Issue Details (click to expand)

SUGGESTION

File	Line	Issue
.changeset/	—	No changeset added for this user-facing bug fix

Observations

Fix correctness (InstanceState.bind): Calling InstanceState.bind(PlanFollowup.ask) before the Effect.promise boundary correctly preserves the active instance context so Instance.current calls inside PlanFollowup.ask work after the async boundary. The bind function gracefully returns fn unchanged when no context is present, so there is no regression risk.

Architect condition: The condition input.agent.options?.id !== "architect" correctly gates plan-file injection on the options.id field as defined by Agent.Info. The access is type-safe since Agent.Info.options is Record<string, unknown>.

InstanceState.bind for Session.plan: The InstanceState.bind(() => Session.plan(input.session, Instance.current))() pattern is correct — it captures the current instance context synchronously before the async boundary, then immediately invokes the bound function. No issue here.

Architect prompt text: The differentiated system-reminder for the architect agent (requiring the user to explicitly choose "Finalize and save the plan" or "Continue refining" before writing the plan file) is a reasonable guard to prevent premature plan saves. The relaxed limit text for architect (allowing other files when the user explicitly asks) is also a sound change.

Missing changeset (unchanged from previous review): This PR fixes a user-visible regression. A .changeset/*.md with patch severity should be added. No changeset file was added in either of the two commits on this PR.

Files Reviewed (1 file)

• packages/opencode/src/kilocode/session/prompt.ts — 1 suggestion (in summary)

Fix these issues in Kilo Cloud

---

_{Reviewed by claude-4.6-sonnet-20260217 · 494,586 tokens}

_{Review guidance: REVIEW.md from base branch main}

[johnnyeric]

Now architect agent will create multiple files if asked.