feat: add /goal command for autonomous task completion #31762

Summary

Add a /goal slash command that sets a completion condition and keeps the agent working across turns until the condition is met. This is inspired by Claude Code's /goal command and OpenAI Codex's Goal Mode.

Motivation

Currently, users must manually prompt the agent after each step for multi-step tasks. A /goal command would allow users to define a completion condition once and let the agent work autonomously until the goal is achieved.

Proposed Behavior

Commands

• /goal <condition> - Set a new goal and start working toward it
• /goal (no args) - Show current goal status
• /goal clear - Clear the active goal

How it works

1. User sets a goal with a completion condition
2. Agent works toward the goal in the current turn
3. After each turn, an evaluator model checks if the condition is met
4. If not met, the evaluator's reason is fed back and the agent continues
5. If met, the goal clears and an achieved entry is logged

Evaluator

• Uses a configurable small/fast model (default: session's configured model)
• Only judges from the conversation transcript (does not run tools)
• Returns yes/no decision with a short reason
• Conditions should be phrased around observable output

Goal State

• Stored per-session in the database
• Tracks: condition, status (active/achieved/cleared), iterations, tokens, timestamps
• Emits events for UI updates

Example Usage

/goal All tests in test/auth pass and lint is clean

The agent will then:

1. Run the tests
2. Check lint
3. If tests fail, fix them and re-run
4. Continue until both conditions are met

Design Decisions

• Independent from todos: Goal is a session-level objective, separate from task lists
• Configurable evaluator: Users can specify a custom evaluator model
• Session-scoped: Goals are per-session, not persistent across sessions
• V1 command system: Matches existing /init and /review pattern

Testing

• Unit tests for SessionGoal service
• Unit tests for goal evaluator
• Integration tests for /goal command
• Tests for auto-continuation behavior

References

• Claude Code /goal: https://code.claude.com/docs/en/goal
• Codex Goal Mode: https://developers.openai.com/codex/use-cases/follow-goals