feat(core): add builtin /steer command

[xukp20]

Summary

This PR adds a new builtin /steer command with a narrow, explicit meaning:

• add guidance to the current in-flight task
• keep it as the same task
• do not define it as a new task/turn command

Backend mapping in this first version:

• Claude Code:
  • send a same-session user message with priority=next
• Codex app-server:
  • call native turn/steer
• Codex exec:
  • not supported yet

Why

There has been prior discussion around repurposing /btw or renaming it to /ps:

• Replace /btw with /ps as a proper builtin command #619
• feat(core): replace /btw with /ps as builtin command #620
• Fix Issue #619: Replace /btw with /ps as proper builtin command #636

After reviewing the current cc-connect behavior, Claude Code semantics, Codex app-server semantics, and both CLI profile research docs, the main conclusions are:

• Claude native /btw is a side-question flow, not same-task steering
• Codex already uses /ps as a native command
• cc-connect still needs a clear cross-backend command for:
  • adding guidance to the currently running task

This PR intentionally introduces /steer as that explicit command instead of trying to settle the final /btw / /ps naming question here.

Scope

Included:

• builtin /steer
• core command routing and help/i18n
• Claude Code steering via user(priority=next)
• Codex app-server steering via native turn/steer

Not included:

• removing or redefining the legacy hidden /btw
• adopting /ps
• Codex exec support
• Claude priority=now

Validation

Local validation completed:

• targeted unit tests for:
  • core /steer
  • Claude priority=next message shape
  • Codex app-server turn/steer request shape
• full local CI:
  • pnpm build
  • actionlint
  • golangci-lint --new-from-rev origin/main
  • go build ./...
  • go test ./... -v -race
  • go test ./... -coverprofile=coverage.out -covermode=atomic
  • go test -v -tags=smoke,no_web ./tests/e2e/...
  • go test -v -tags=regression,no_web ./tests/e2e/...
  • go test -bench=. -benchmem -tags=performance,no_web ./tests/performance/...

Related:

• Replace /btw with /ps as a proper builtin command #619
• feat(core): replace /btw with /ps as builtin command #620
• Fix Issue #619: Replace /btw with /ps as proper builtin command #636

[chenhg5]

Review Summary

/steer through optional SessionSteerer interface is well-designed — avoids tight coupling to specific agents. Claude and Codex app-server mappings with unit tests are solid, i18n and docs consistent.

---

🔴 Blocker: corrupted shellJoinArgs comment
File: agent/claudecode/session.go (shellJoinArgs)
Why: The comment demonstrating single-quote concatenation was changed from the correct 'it'\''s' to a version with incorrect characters, misrepresenting real shell escaping and misleading future maintainers.
Fix: Restore the correct ASCII escape example (or match the original line exactly) to avoid introducing curly quotes / incorrect Unicode.

---

✅ Good: cmdSteer uses type assertion to handle unsupported backends; Codex exec scope is documented.

Request changes — please fix the shellJoinArgs comment regression.

[xukp20]

Fixed. I restored the shellJoinArgs comment to the correct ASCII single-quote escape example: 'it'\''s' -> it's.\n\nValidation run locally after the change:\n- go test ./agent/claudecode -count=1\n\nPushed in commit dba1efe.

[ternary-chen]

Thanks for the careful work here. I opened #1652 as a Draft based on the current main.

It follows the same optional-capability and Codex turn/steer direction, but has a narrower and different user-facing scope: it does not add a cross-agent /steer command or Claude behavior. Instead, it adds opt-in busy_message_mode = "steer" for plain-text busy follow-ups, keeps queue as the default, falls back to the existing queue on failure, keeps attachments queued, and makes the existing /ps path use the steering capability safely.

There is clear overlap with the Codex portion of #669. I am linking the draft for coordination and am happy to reuse the established interface/naming, split the work, or close #1652 if maintainers prefer continuing this PR.