test: comprehensive shell passthrough test suite + extract into library

[emal-avala]

Summary

Extracts shell passthrough logic into a testable library module and adds 40 tests across 3 layers.

What changed

New module: crates/lib/src/services/shell_passthrough.rs (519 lines)

Extracts the ! prefix logic from repl.rs into reusable, testable functions:

• capture_lines() — reads from any Read impl, captures to buffer with 50KB limit
• run_and_capture() — spawns bash subprocess, streams output, returns CapturedOutput
• build_context_message() — builds UserMessage(is_meta: true) from captured output

Refactored: crates/cli/src/ui/repl.rs (92 lines → 19 lines)

The ! handler now calls the library module. Same behavior, 73 fewer lines.

Test coverage

25 unit tests (in shell_passthrough.rs):

Category	Tests	What they verify
capture_lines	6	Under limit, truncation, empty, single line, callbacks after truncation, exact boundary
build_context_message	5	Header format, truncation suffix, empty→None, multiline, special chars
run_and_capture	10	stdout, stderr, mixed, multiline, empty, exit codes, cwd, callbacks, truncation, invalid cmd
Full pipeline	3	echo→message, empty→None, truncated→suffix

11 integration tests (in shell_passthrough_integration.rs):

Category	Tests	What they verify
Message system	3	Valid UserMessage, alternation compatibility, JSON roundtrip
Subprocess	5	cwd, exit codes, binary output, 500-line capture, complete lines on truncation
Content format	3	Header format, suffix position, special character preservation

4 E2E bash tests (section L in e2e-tests.sh):

Test	What it verifies
L1	Tool output appears in /messages conversation history
L2	Multi-turn context retention with tool output
L3	Large output handling (1000 lines, no crash)
L4	stderr captured alongside stdout

Test results

621 tests total, 0 failures, 0 ignored
cargo clippy: 0 warnings
cargo fmt: clean

Test plan

• cargo fmt --all -- --check — clean
• cargo clippy -- -D warnings — zero warnings
• cargo test — 621 tests pass (40 new)
• E2E bash tests (L1-L4) require API key — tested in CI via release-e2e.yml

🤖 Generated with Claude Code

[chatgpt-codex-connector]

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