feat: add S hotkey to apply agent prompt as system override mid-session

[lenam2001]

Feature: Apply Agent System Prompt Mid-Session (S hotkey)

Problem

The /agents dashboard in oh-my-pi currently only supports:

• Enable/disable agent (Space)
• Edit model override (Enter)
• Create new agent (N)

There is no way to apply an agent's system prompt into the current session to switch persona mid-session (e.g., librarian → coordinator → tutor).

Solution

Add hotkey S in the Agent Dashboard to apply the selected agent's system prompt into the current active session.

Files Modified

File	Change
packages/coding-agent/src/modes/components/agent-dashboard.ts	onApplySystemPrompt callback, #applySelectedAgentPrompt(), S hotkey handler, footer hint
packages/coding-agent/src/modes/controllers/selector-controller.ts	Wire onApplySystemPrompt → session.applyAgentSystemPrompt()
packages/coding-agent/src/session/agent-session.ts	Add applyAgentSystemPrompt(agentName, prompt) — updates agent.setSystemPrompt() + persists custom_message entry

Design Decisions

• Callback bridge between AgentDashboard (UI) and AgentSession (logic) to preserve separation of concerns
• custom_message entry with display: false to avoid cluttering the TUI
• Model remains unchanged — only system prompt is overridden (use Enter if model switch is also needed)

Verification

• TypeScript compile (bun tsc --noEmit): pass
• Biome check + tsgo (bun run check): pass (962 files, no errors)
• Full repo build: blocked by Rust native compilation (pi_natives) on Windows
• Unit tests: blocked by native addon load failure on Windows
• Runtime test in live omp session: pending

Agent Workflow Context

This feature enables the multi-agent orchestration pattern used in OpenCode-style configurations where agents transition through roles mid-session:

1. 1st_librarian — discovers context
2. 2nd_coordinator — plans and delegates
3. 3rd_tutor — teaches and explains

With the S hotkey, users can now fluidly switch between these personas without restarting the session.

Next Steps

• Maintainer review
• Runtime verification in live omp session: /agents → select agent → press S → confirm system prompt switch

[lenam2001]

📎 Related discussion thread opened at #813 for broader feedback on multi-agent persona switching workflows.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e0e4b16374

ℹ️ About Codex in GitHub

Codex has been enabled to automatically 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 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Persist applied prompt as the session base prompt

Updating only agent.setSystemPrompt(prompt) here makes the S hotkey ineffective for sessions that have an extension runner: before each user prompt, emitBeforeAgentStart falls back to this.#baseSystemPrompt and overwrites the applied prompt (agent-session.ts around lines 2757-2760). In that environment, users will see “Applied ... agent prompt” but the next turn runs with the old base prompt, so the feature silently fails unless #baseSystemPrompt is updated too.

Useful? React with 👍 / 👎.