feat(cli): add /output-style for preset response voices

[emal-avala]

Summary

Four selectable response styles, each injecting a different instruction block into the system prompt to shape the model's voice:

Style	Effect
default	No override (the codebase's default voice)
concise	Shorter, fewer qualifiers, skip prefaces
explanatory	Explain reasoning and trade-offs as you go
learning	Narrate each step for users new to the codebase

> /output-style learning
Response style set to 'learning'.

> /output-style
Available response styles:
  default       — no override  (active)
  concise       — shorter responses, fewer qualifiers
  explanatory   — explain reasoning and trade-offs
  learning      — narrate steps for new-to-codebase users  (active)

Usage: /output-style <name>   (alias: /style)

Why

Different pair-programming modes want different voices. A reviewer wants terse results; an onboarding hire wants explanations; a library author wants alternative designs weighed out. Putting them on a slash command makes the switch one keystroke and session-scoped.

Implementation

• New enum ResponseStyle in crates/lib/src/state/mod.rs — Default, Concise, Explanatory, Learning, each with a prompt_fragment() method
• AppState gains a response_style: ResponseStyle field (session-local, not persisted)
• build_system_prompt emits a # Active response style block when the style is non-Default — placed before # Available Tools so it stays salient in long contexts
• Heading is distinct from the existing static # Response style guideline block so they don't collide
• response_style is included in the system-prompt cache hash so flipping style takes effect on the very next turn

Aliases

• Command: /output-style, /style
• Names: off/normal → default, terse → concise, explain → explanatory, teach/teacher → learning

Test plan

• cargo fmt --all — clean
• cargo clippy --workspace --all-targets -- -D warnings — clean
• cargo test -p agent-code-lib --lib state — 11/11 pass (6 new for ResponseStyle)
• cargo test -p agent-code-lib --lib system_prompt — 4/4 pass (2 new)
  • system_prompt_omits_style_block_for_default
  • system_prompt_injects_non_default_style_fragment (iterates all 3 non-default styles, asserts fragment appears before # Available Tools)

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.