feat(rivetkit-agent-os): phase 3 process — 11 arms (exec/spawn/wait/kill/stop/list/all/tree/get/writeStdin/closeStdin) + ExecResultDto

[abcxff]

• 18/24 driver tests green (spawn+waitProcess, listProcesses, killProcess across all 6 native cells).
• 6 exec failures all from same root cause: agent-os sidecar's resolve_software uses node_modules/{package} which doesn't traverse pnpm's .pnpm hoisting. Phase 2.5 software mapping gap, not arm bugs.

[abcxff]

Stack for rivet-dev/rivet

• feat(rivetkit-agent-os): phase 3 process — 11 arms (exec/spawn/wait/kill/stop/list/all/tree/get/writeStdin/closeStdin) + ExecResultDto #5197 unkrsszt · 2026-06-08 18:29:29 👈
• test(rivetkit-agent-os): partial-failure driver test for readFiles batch DTO (Option<content>, Option<error>) #5196 pzzzpmtr · 2026-06-08 18:29:28
• feat(rivetkit-agent-os): phase 3 filesystem — mkdir/readdir/exists/move/deleteFile/readFiles/writeFiles/readdirRecursive arms + batch DTOs #5195 onyykwlk · 2026-06-08 18:29:27
• test(rivetkit-agent-os): config parsing tests + documented JS↔Rust software shape mismatch (phase 3 prep) #5194 ouyvrlxt · 2026-06-08 18:29:26
• feat(rivetkit-agent-os): phase 2 — stat action arm + structured-payload wrapper tests for cross-encoding parity #5193 slkvonmt · 2026-06-08 18:29:25
• feat(rivetkit): phase 1c — agentOs() native ladder + engine driver branch + driver-suite e2e green #5192 nxzrussy · 2026-06-08 18:29:24
• feat(rivetkit-napi): phase 1b — NapiActorFactory.fromAgentOs static constructor + writeFile arm + e2e factory round-trip #5191 mtlnunov · 2026-06-08 18:29:22
• feat(rivetkit-agent-os): phase 1a — rust crate skeleton + readFile arm + dispatcher e2e against real sidecar #5190 wrxkrssv · 2026-06-08 18:29:21
• feat(rivetkit): JSON_COMPAT_UINT8_ARRAY byte-payload encoding parity with TS (encode + decode + cross-language tests) #5189 ukssumln · 2026-06-08 18:29:20
• chore: phase 0 prereqs (rivetkit workspace member, serde_bare align, wasm skip.agentOs default) #5188 qvnkvkxw · 2026-06-08 18:29:19
• docs: PLAN2 + TODOLIST + RIVETKIT_RUST_FIX (clean slate, post-wipe) #5187 yqrwozvn · 2026-06-08 18:29:18
• main

Check out this stack: forklift get https://github.com/rivet-dev/rivet/pull/5197
Pull/update this stack: forklift sync
Publish local edits: forklift submit
Merge when ready: forklift merge

[claude]

Code Review - Phase 3 Process Arms

The implementation cleanly follows the filesystem.rs pattern. Here are the findings:

Correctness / Reliability

exec has no timeout: exec calls vm.exec(command, ExecOptions::default()) with default options. If ExecOptions has no deadline, a hung command blocks the actor indefinitely. Consider adding a configurable timeout arg or documenting that the caller is responsible.

listProcesses is dispatched as infallible: the arm calls list_processes and replies ok with no error path. Only safe if AgentOs::list_processes is genuinely infallible. Worth confirming at the agent_os_client level.

Style Issues

Import inside function body (process.rs, write_process_stdin): StdinInput is imported with a use statement inside the function body. CLAUDE.md says to always add imports at the top of the file. filesystem.rs:139 has the same pre-existing pattern but that does not justify continuing it.

processTree arm is missing outer braces unlike all other arms. Minor inconsistency.

Inline comment "// No args." in the listProcesses arm violates CLAUDE.md guidance to only comment when the WHY is non-obvious.

Known Gaps

6/24 driver tests fail due to pnpm hoisting - the sidecar resolve_software does not traverse pnpm .pnpm hoisting. Well-documented in the PR body and updated TS comment. No action needed here.

Minor Notes

ExecResultDto only renames exit_code to exitCode; stdout and stderr are fine as-is.
spawn being sync is acceptable for a brief fork/exec syscall.
map_err(anyhow::Error::from) is correct where the underlying type does not impl Intoanyhow::Error.

Summary

The two things worth addressing before merge are the exec timeout concern (functional) and the inline import (CLAUDE.md compliance). The missing braces on processTree and the no-args comment are trivial. The 6-test gap is pre-acknowledged and scoped to Phase 2.5, not these arms.