fix: use prctl FFI to set process name as 'agent' in top/ps

[konard]

Summary

Fixes #144 — Our Agent CLI should be identified in top as agent not bun command.

Root Cause

The previous fix (PR #145) used process.title = 'agent' and process.argv0 = 'agent', but Bun does not implement the process.title setter unlike Node.js. Setting these properties only changes in-process JavaScript values but does not call prctl(PR_SET_NAME) under the hood like Node.js does.

Evidence from experiments:

// After process.title = 'agent':
/proc/PID/comm: bun          # Still shows 'bun'
ps COMMAND: bun               # Still shows 'bun'

Solution

Use Bun's FFI to call prctl(PR_SET_NAME, "agent") on Linux, which sets /proc/<pid>/comm — the kernel-level process name visible in top, ps, and htop.

On macOS and Windows, no FFI is needed:

• macOS: When installed via bun install -g, the symlink is named agent, so macOS already shows agent in ps/top
• Windows: Task Manager always shows the executable name

Verification

After the fix on Linux:

/proc/PID/comm: agent         # Now shows 'agent'
ps COMMAND: agent              # Now shows 'agent'

Changes

• js/src/cli/process-name.ts — New module with setProcessName() using Bun FFI prctl(PR_SET_NAME) on Linux
• js/src/index.js — Replaced ineffective process.title/process.argv0 with setProcessName('agent') import
• js/tests/process-name.test.js — Autonomous test that spawns a bun process, calls setProcessName, and verifies the kernel comm name equals agent (Linux-only, skips on macOS/Windows)
• .github/workflows/js.yml — Added process-name.test.js to CI test runner
• js/.changeset/fix-process-name-ffi.md — Changeset for patch release
• experiments/ — Research scripts documenting the investigation

Test plan

• bun test tests/process-name.test.js passes locally (Linux)
• bun test tests/json-standard-unit.test.js passes (no regression)
• ESLint + Prettier checks pass
• CI passes on Ubuntu (Linux assertions pass)
• CI passes on macOS (tests gracefully skip with message)
• CI passes on Windows (tests gracefully skip with message)

🤖 Generated with Claude Code

[konard]

🤖 Solution Draft Log

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost estimation:

• Public pricing estimate: $7.762179
• Calculated by Anthropic: $5.306180 USD
• Difference: $-2.455999 (-31.64%)
  📎 Log file uploaded as Gist (974KB)
  🔗 View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.