fix(server): avoid macos path walks in terminal polling by t3-code[bot] · Pull Request #6372 · pingdotgg/t3code

t3-code · 2026-08-12T20:26:02Z

what changed

use /usr/bin/pgrep and /bin/ps for terminal subprocess polling on macos
retain bare utility names on linux and other posix platforms
cover every macos subprocess inspection path, including the command fallback probe

why

node/libuv resolves bare executable names by trying each path entry with posix_spawn on macos. because terminal subprocess activity is polled every second, this multiplied a few logical probes into hundreds of failed spawn attempts per second on machines with many terminals and a long path.

absolute paths bypass that lookup without changing the polling behavior. this removes about 94% of the reported spawn attempts in the 14-terminal, 17-entry case.

fixes#6332

verification

vp test run src/terminal/Manager.test.ts -t 'uses absolute utility paths' --maxWorkers=1
vp test run src/terminal/Manager.test.ts --maxWorkers=1
vp test run --maxWorkers=1 in apps/server: 229 files passed, 2 skipped; 2380 tests passed, 7 skipped
vp run --filter t3 typecheck
vp check apps/server/src/terminal/Manager.ts apps/server/src/terminal/Manager.test.ts

checklist

this pr is small and focused
i explained what changed and why

request provenance

requested by: @extoci

Note

Low Risk
Behavioral change is limited to how utilities are invoked on macOS during subprocess inspection; polling logic and Linux paths are unchanged.

Overview
macOS terminal subprocess polling now invokes /usr/bin/pgrep and /bin/ps instead of resolving pgrep and ps through PATH, including the comm= and args= fallback probes in posixInspectSubprocess. Linux and other POSIX hosts still use the bare utility names.

This avoids libuv’s per-PATH-entry posix_spawn lookup on macOS, which was amplifying each ~1s poll into many failed spawns when many terminals were open (fixes#6332).

A new Manager.test case asserts the first four process-runner commands per platform: absolute paths on darwin, bare names on linux.

^{Reviewed by Cursor Bugbot for commit 16eedd9. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Use absolute binary paths for `pgrep` and `ps` in terminal polling on macOS

On macOS, posixInspectSubprocess in Manager.ts now calls /usr/bin/pgrep and /bin/ps directly instead of relying on PATH resolution. On Linux, bare command names are still used. A new test in Manager.test.ts verifies the correct command is invoked per platform.

^{Macroscope summarized 16eedd9.}

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

github-actions · 2026-08-12T20:28:51Z

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	11.3 KiB	11.3 KiB	+12 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	5.5 KiB	5.5 KiB	+10 B (+0.2%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	5.9 KiB	5.9 KiB	+2 B (+0.0%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	49.7 KiB	49.7 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	16	16	0 (0.0%)	21	✅
Claude	Total thread wire	11.3 KiB	11.3 KiB	0 B (0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	5.5 KiB	5.5 KiB	−3 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	5.9 KiB	5.9 KiB	+3 B (+0.0%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	50.6 KiB	50.6 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	16	16	0 (0.0%)	21	✅

Baseline: 18918d1 · PR result: 16eedd9 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

Codex decoded thread snapshot: 94.6 KiB
Claude decoded thread snapshot: 95.4 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

macroscopeapp · 2026-08-12T20:29:38Z

Approvability

Verdict: Approved 4b71bac

Performance optimization that uses absolute paths for pgrep and ps on macOS to avoid libuv PATH search overhead. The change is small, self-contained, and well-tested with no functional impact.

No code changes detected at 16eedd9. Prior analysis still applies.

^{You can customize Macroscope's approvability policy. Learn more.}

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

Dismissing prior approval to re-evaluate 972740d

Dismissing prior approval to re-evaluate 4b71bac

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

juliusmarminge · 2026-08-15T11:31:54Z

Closing as superseded by #6377, which has merged and covers this change on the current architecture. Thanks for the contribution.

t3-code · 2026-08-12T20:26:02Z

what changed

use /usr/bin/pgrep and /bin/ps for terminal subprocess polling on macos
retain bare utility names on linux and other posix platforms
cover every macos subprocess inspection path, including the command fallback probe

why

node/libuv resolves bare executable names by trying each path entry with posix_spawn on macos. because terminal subprocess activity is polled every second, this multiplied a few logical probes into hundreds of failed spawn attempts per second on machines with many terminals and a long path.

absolute paths bypass that lookup without changing the polling behavior. this removes about 94% of the reported spawn attempts in the 14-terminal, 17-entry case.

fixes#6332

verification

vp test run src/terminal/Manager.test.ts -t 'uses absolute utility paths' --maxWorkers=1
vp test run src/terminal/Manager.test.ts --maxWorkers=1
vp test run --maxWorkers=1 in apps/server: 229 files passed, 2 skipped; 2380 tests passed, 7 skipped
vp run --filter t3 typecheck
vp check apps/server/src/terminal/Manager.ts apps/server/src/terminal/Manager.test.ts

checklist

this pr is small and focused
i explained what changed and why

request provenance

requested by: @extoci

Note

Low Risk
Behavioral change is limited to how utilities are invoked on macOS during subprocess inspection; polling logic and Linux paths are unchanged.

Overview
macOS terminal subprocess polling now invokes /usr/bin/pgrep and /bin/ps instead of resolving pgrep and ps through PATH, including the comm= and args= fallback probes in posixInspectSubprocess. Linux and other POSIX hosts still use the bare utility names.

This avoids libuv’s per-PATH-entry posix_spawn lookup on macOS, which was amplifying each ~1s poll into many failed spawns when many terminals were open (fixes#6332).

A new Manager.test case asserts the first four process-runner commands per platform: absolute paths on darwin, bare names on linux.

^{Reviewed by Cursor Bugbot for commit 16eedd9. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Use absolute binary paths for `pgrep` and `ps` in terminal polling on macOS

On macOS, posixInspectSubprocess in Manager.ts now calls /usr/bin/pgrep and /bin/ps directly instead of relying on PATH resolution. On Linux, bare command names are still used. A new test in Manager.test.ts verifies the correct command is invoked per platform.

^{Macroscope summarized 16eedd9.}

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

github-actions · 2026-08-12T20:28:51Z

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	11.3 KiB	11.3 KiB	+12 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	5.5 KiB	5.5 KiB	+10 B (+0.2%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	5.9 KiB	5.9 KiB	+2 B (+0.0%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	49.7 KiB	49.7 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	16	16	0 (0.0%)	21	✅
Claude	Total thread wire	11.3 KiB	11.3 KiB	0 B (0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	5.5 KiB	5.5 KiB	−3 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	5.9 KiB	5.9 KiB	+3 B (+0.0%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	50.6 KiB	50.6 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	16	16	0 (0.0%)	21	✅

Baseline: 18918d1 · PR result: 16eedd9 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

Codex decoded thread snapshot: 94.6 KiB
Claude decoded thread snapshot: 95.4 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

macroscopeapp · 2026-08-12T20:29:38Z

Approvability

Verdict: Approved 4b71bac

Performance optimization that uses absolute paths for pgrep and ps on macOS to avoid libuv PATH search overhead. The change is small, self-contained, and well-tested with no functional impact.

No code changes detected at 16eedd9. Prior analysis still applies.

^{You can customize Macroscope's approvability policy. Learn more.}

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

Dismissing prior approval to re-evaluate 972740d

Dismissing prior approval to re-evaluate 4b71bac

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

juliusmarminge · 2026-08-15T11:31:54Z

Closing as superseded by #6377, which has merged and covers this change on the current architecture. Thanks for the contribution.

t3-code · 2026-08-12T20:26:02Z

what changed

use /usr/bin/pgrep and /bin/ps for terminal subprocess polling on macos
retain bare utility names on linux and other posix platforms
cover every macos subprocess inspection path, including the command fallback probe

why

node/libuv resolves bare executable names by trying each path entry with posix_spawn on macos. because terminal subprocess activity is polled every second, this multiplied a few logical probes into hundreds of failed spawn attempts per second on machines with many terminals and a long path.

absolute paths bypass that lookup without changing the polling behavior. this removes about 94% of the reported spawn attempts in the 14-terminal, 17-entry case.

fixes#6332

verification

vp test run src/terminal/Manager.test.ts -t 'uses absolute utility paths' --maxWorkers=1
vp test run src/terminal/Manager.test.ts --maxWorkers=1
vp test run --maxWorkers=1 in apps/server: 229 files passed, 2 skipped; 2380 tests passed, 7 skipped
vp run --filter t3 typecheck
vp check apps/server/src/terminal/Manager.ts apps/server/src/terminal/Manager.test.ts

checklist

this pr is small and focused
i explained what changed and why

request provenance

requested by: @extoci

Note

Low Risk
Behavioral change is limited to how utilities are invoked on macOS during subprocess inspection; polling logic and Linux paths are unchanged.

Overview
macOS terminal subprocess polling now invokes /usr/bin/pgrep and /bin/ps instead of resolving pgrep and ps through PATH, including the comm= and args= fallback probes in posixInspectSubprocess. Linux and other POSIX hosts still use the bare utility names.

This avoids libuv’s per-PATH-entry posix_spawn lookup on macOS, which was amplifying each ~1s poll into many failed spawns when many terminals were open (fixes#6332).

A new Manager.test case asserts the first four process-runner commands per platform: absolute paths on darwin, bare names on linux.

^{Reviewed by Cursor Bugbot for commit 16eedd9. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Use absolute binary paths for `pgrep` and `ps` in terminal polling on macOS

On macOS, posixInspectSubprocess in Manager.ts now calls /usr/bin/pgrep and /bin/ps directly instead of relying on PATH resolution. On Linux, bare command names are still used. A new test in Manager.test.ts verifies the correct command is invoked per platform.

^{Macroscope summarized 16eedd9.}

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

github-actions · 2026-08-12T20:28:51Z

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	11.3 KiB	11.3 KiB	+12 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	5.5 KiB	5.5 KiB	+10 B (+0.2%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	5.9 KiB	5.9 KiB	+2 B (+0.0%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	49.7 KiB	49.7 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	16	16	0 (0.0%)	21	✅
Claude	Total thread wire	11.3 KiB	11.3 KiB	0 B (0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	5.5 KiB	5.5 KiB	−3 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	5.9 KiB	5.9 KiB	+3 B (+0.0%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	50.6 KiB	50.6 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	16	16	0 (0.0%)	21	✅

Baseline: 18918d1 · PR result: 16eedd9 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

Codex decoded thread snapshot: 94.6 KiB
Claude decoded thread snapshot: 95.4 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

macroscopeapp · 2026-08-12T20:29:38Z

Approvability

Verdict: Approved 4b71bac

Performance optimization that uses absolute paths for pgrep and ps on macOS to avoid libuv PATH search overhead. The change is small, self-contained, and well-tested with no functional impact.

No code changes detected at 16eedd9. Prior analysis still applies.

^{You can customize Macroscope's approvability policy. Learn more.}

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

Dismissing prior approval to re-evaluate 972740d

Dismissing prior approval to re-evaluate 4b71bac

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

juliusmarminge · 2026-08-15T11:31:54Z

Closing as superseded by #6377, which has merged and covers this change on the current architecture. Thanks for the contribution.

t3-code · 2026-08-12T20:26:02Z

what changed

use /usr/bin/pgrep and /bin/ps for terminal subprocess polling on macos
retain bare utility names on linux and other posix platforms
cover every macos subprocess inspection path, including the command fallback probe

why

node/libuv resolves bare executable names by trying each path entry with posix_spawn on macos. because terminal subprocess activity is polled every second, this multiplied a few logical probes into hundreds of failed spawn attempts per second on machines with many terminals and a long path.

absolute paths bypass that lookup without changing the polling behavior. this removes about 94% of the reported spawn attempts in the 14-terminal, 17-entry case.

fixes#6332

verification

vp test run src/terminal/Manager.test.ts -t 'uses absolute utility paths' --maxWorkers=1
vp test run src/terminal/Manager.test.ts --maxWorkers=1
vp test run --maxWorkers=1 in apps/server: 229 files passed, 2 skipped; 2380 tests passed, 7 skipped
vp run --filter t3 typecheck
vp check apps/server/src/terminal/Manager.ts apps/server/src/terminal/Manager.test.ts

checklist

this pr is small and focused
i explained what changed and why

request provenance

requested by: @extoci

Note

Low Risk
Behavioral change is limited to how utilities are invoked on macOS during subprocess inspection; polling logic and Linux paths are unchanged.

Overview
macOS terminal subprocess polling now invokes /usr/bin/pgrep and /bin/ps instead of resolving pgrep and ps through PATH, including the comm= and args= fallback probes in posixInspectSubprocess. Linux and other POSIX hosts still use the bare utility names.

This avoids libuv’s per-PATH-entry posix_spawn lookup on macOS, which was amplifying each ~1s poll into many failed spawns when many terminals were open (fixes#6332).

A new Manager.test case asserts the first four process-runner commands per platform: absolute paths on darwin, bare names on linux.

^{Reviewed by Cursor Bugbot for commit 16eedd9. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Use absolute binary paths for `pgrep` and `ps` in terminal polling on macOS

On macOS, posixInspectSubprocess in Manager.ts now calls /usr/bin/pgrep and /bin/ps directly instead of relying on PATH resolution. On Linux, bare command names are still used. A new test in Manager.test.ts verifies the correct command is invoked per platform.

^{Macroscope summarized 16eedd9.}

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

github-actions · 2026-08-12T20:28:51Z

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	11.3 KiB	11.3 KiB	+12 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	5.5 KiB	5.5 KiB	+10 B (+0.2%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	5.9 KiB	5.9 KiB	+2 B (+0.0%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	49.7 KiB	49.7 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	16	16	0 (0.0%)	21	✅
Claude	Total thread wire	11.3 KiB	11.3 KiB	0 B (0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	5.5 KiB	5.5 KiB	−3 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	5.9 KiB	5.9 KiB	+3 B (+0.0%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	50.6 KiB	50.6 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	16	16	0 (0.0%)	21	✅

Baseline: 18918d1 · PR result: 16eedd9 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

Codex decoded thread snapshot: 94.6 KiB
Claude decoded thread snapshot: 95.4 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

macroscopeapp · 2026-08-12T20:29:38Z

Approvability

Verdict: Approved 4b71bac

Performance optimization that uses absolute paths for pgrep and ps on macOS to avoid libuv PATH search overhead. The change is small, self-contained, and well-tested with no functional impact.

No code changes detected at 16eedd9. Prior analysis still applies.

^{You can customize Macroscope's approvability policy. Learn more.}

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

Dismissing prior approval to re-evaluate 972740d

Dismissing prior approval to re-evaluate 4b71bac

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

juliusmarminge · 2026-08-15T11:31:54Z

Closing as superseded by #6377, which has merged and covers this change on the current architecture. Thanks for the contribution.

t3-code · 2026-08-12T20:26:02Z

what changed

use /usr/bin/pgrep and /bin/ps for terminal subprocess polling on macos
retain bare utility names on linux and other posix platforms
cover every macos subprocess inspection path, including the command fallback probe

why

node/libuv resolves bare executable names by trying each path entry with posix_spawn on macos. because terminal subprocess activity is polled every second, this multiplied a few logical probes into hundreds of failed spawn attempts per second on machines with many terminals and a long path.

absolute paths bypass that lookup without changing the polling behavior. this removes about 94% of the reported spawn attempts in the 14-terminal, 17-entry case.

fixes#6332

verification

vp test run src/terminal/Manager.test.ts -t 'uses absolute utility paths' --maxWorkers=1
vp test run src/terminal/Manager.test.ts --maxWorkers=1
vp test run --maxWorkers=1 in apps/server: 229 files passed, 2 skipped; 2380 tests passed, 7 skipped
vp run --filter t3 typecheck
vp check apps/server/src/terminal/Manager.ts apps/server/src/terminal/Manager.test.ts

checklist

this pr is small and focused
i explained what changed and why

request provenance

requested by: @extoci

Note

Low Risk
Behavioral change is limited to how utilities are invoked on macOS during subprocess inspection; polling logic and Linux paths are unchanged.

Overview
macOS terminal subprocess polling now invokes /usr/bin/pgrep and /bin/ps instead of resolving pgrep and ps through PATH, including the comm= and args= fallback probes in posixInspectSubprocess. Linux and other POSIX hosts still use the bare utility names.

This avoids libuv’s per-PATH-entry posix_spawn lookup on macOS, which was amplifying each ~1s poll into many failed spawns when many terminals were open (fixes#6332).

A new Manager.test case asserts the first four process-runner commands per platform: absolute paths on darwin, bare names on linux.

^{Reviewed by Cursor Bugbot for commit 16eedd9. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Use absolute binary paths for `pgrep` and `ps` in terminal polling on macOS

On macOS, posixInspectSubprocess in Manager.ts now calls /usr/bin/pgrep and /bin/ps directly instead of relying on PATH resolution. On Linux, bare command names are still used. A new test in Manager.test.ts verifies the correct command is invoked per platform.

^{Macroscope summarized 16eedd9.}

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

github-actions · 2026-08-12T20:28:51Z

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	11.3 KiB	11.3 KiB	+12 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	5.5 KiB	5.5 KiB	+10 B (+0.2%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	5.9 KiB	5.9 KiB	+2 B (+0.0%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	49.7 KiB	49.7 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	16	16	0 (0.0%)	21	✅
Claude	Total thread wire	11.3 KiB	11.3 KiB	0 B (0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	5.5 KiB	5.5 KiB	−3 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	5.9 KiB	5.9 KiB	+3 B (+0.0%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	50.6 KiB	50.6 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	16	16	0 (0.0%)	21	✅

Baseline: 18918d1 · PR result: 16eedd9 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

Codex decoded thread snapshot: 94.6 KiB
Claude decoded thread snapshot: 95.4 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

macroscopeapp · 2026-08-12T20:29:38Z

Approvability

Verdict: Approved 4b71bac

Performance optimization that uses absolute paths for pgrep and ps on macOS to avoid libuv PATH search overhead. The change is small, self-contained, and well-tested with no functional impact.

No code changes detected at 16eedd9. Prior analysis still applies.

^{You can customize Macroscope's approvability policy. Learn more.}

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

Dismissing prior approval to re-evaluate 972740d

Dismissing prior approval to re-evaluate 4b71bac

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

juliusmarminge · 2026-08-15T11:31:54Z

Closing as superseded by #6377, which has merged and covers this change on the current architecture. Thanks for the contribution.

t3-code · 2026-08-12T20:26:02Z

what changed

use /usr/bin/pgrep and /bin/ps for terminal subprocess polling on macos
retain bare utility names on linux and other posix platforms
cover every macos subprocess inspection path, including the command fallback probe

why

node/libuv resolves bare executable names by trying each path entry with posix_spawn on macos. because terminal subprocess activity is polled every second, this multiplied a few logical probes into hundreds of failed spawn attempts per second on machines with many terminals and a long path.

absolute paths bypass that lookup without changing the polling behavior. this removes about 94% of the reported spawn attempts in the 14-terminal, 17-entry case.

fixes#6332

verification

vp test run src/terminal/Manager.test.ts -t 'uses absolute utility paths' --maxWorkers=1
vp test run src/terminal/Manager.test.ts --maxWorkers=1
vp test run --maxWorkers=1 in apps/server: 229 files passed, 2 skipped; 2380 tests passed, 7 skipped
vp run --filter t3 typecheck
vp check apps/server/src/terminal/Manager.ts apps/server/src/terminal/Manager.test.ts

checklist

this pr is small and focused
i explained what changed and why

request provenance

requested by: @extoci

Note

Low Risk
Behavioral change is limited to how utilities are invoked on macOS during subprocess inspection; polling logic and Linux paths are unchanged.

Overview
macOS terminal subprocess polling now invokes /usr/bin/pgrep and /bin/ps instead of resolving pgrep and ps through PATH, including the comm= and args= fallback probes in posixInspectSubprocess. Linux and other POSIX hosts still use the bare utility names.

This avoids libuv’s per-PATH-entry posix_spawn lookup on macOS, which was amplifying each ~1s poll into many failed spawns when many terminals were open (fixes#6332).

A new Manager.test case asserts the first four process-runner commands per platform: absolute paths on darwin, bare names on linux.

^{Reviewed by Cursor Bugbot for commit 16eedd9. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Use absolute binary paths for `pgrep` and `ps` in terminal polling on macOS

On macOS, posixInspectSubprocess in Manager.ts now calls /usr/bin/pgrep and /bin/ps directly instead of relying on PATH resolution. On Linux, bare command names are still used. A new test in Manager.test.ts verifies the correct command is invoked per platform.

^{Macroscope summarized 16eedd9.}

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

github-actions · 2026-08-12T20:28:51Z

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	11.3 KiB	11.3 KiB	+12 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	5.5 KiB	5.5 KiB	+10 B (+0.2%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	5.9 KiB	5.9 KiB	+2 B (+0.0%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	49.7 KiB	49.7 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	16	16	0 (0.0%)	21	✅
Claude	Total thread wire	11.3 KiB	11.3 KiB	0 B (0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	5.5 KiB	5.5 KiB	−3 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	5.9 KiB	5.9 KiB	+3 B (+0.0%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	50.6 KiB	50.6 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	16	16	0 (0.0%)	21	✅

Baseline: 18918d1 · PR result: 16eedd9 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

Codex decoded thread snapshot: 94.6 KiB
Claude decoded thread snapshot: 95.4 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

macroscopeapp · 2026-08-12T20:29:38Z

Approvability

Verdict: Approved 4b71bac

Performance optimization that uses absolute paths for pgrep and ps on macOS to avoid libuv PATH search overhead. The change is small, self-contained, and well-tested with no functional impact.

No code changes detected at 16eedd9. Prior analysis still applies.

^{You can customize Macroscope's approvability policy. Learn more.}

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

Dismissing prior approval to re-evaluate 972740d

Dismissing prior approval to re-evaluate 4b71bac

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

juliusmarminge · 2026-08-15T11:31:54Z

Closing as superseded by #6377, which has merged and covers this change on the current architecture. Thanks for the contribution.

t3-code · 2026-08-12T20:26:02Z

what changed

use /usr/bin/pgrep and /bin/ps for terminal subprocess polling on macos
retain bare utility names on linux and other posix platforms
cover every macos subprocess inspection path, including the command fallback probe

why

node/libuv resolves bare executable names by trying each path entry with posix_spawn on macos. because terminal subprocess activity is polled every second, this multiplied a few logical probes into hundreds of failed spawn attempts per second on machines with many terminals and a long path.

absolute paths bypass that lookup without changing the polling behavior. this removes about 94% of the reported spawn attempts in the 14-terminal, 17-entry case.

fixes#6332

verification

vp test run src/terminal/Manager.test.ts -t 'uses absolute utility paths' --maxWorkers=1
vp test run src/terminal/Manager.test.ts --maxWorkers=1
vp test run --maxWorkers=1 in apps/server: 229 files passed, 2 skipped; 2380 tests passed, 7 skipped
vp run --filter t3 typecheck
vp check apps/server/src/terminal/Manager.ts apps/server/src/terminal/Manager.test.ts

checklist

this pr is small and focused
i explained what changed and why

request provenance

requested by: @extoci

Note

Low Risk
Behavioral change is limited to how utilities are invoked on macOS during subprocess inspection; polling logic and Linux paths are unchanged.

Overview
macOS terminal subprocess polling now invokes /usr/bin/pgrep and /bin/ps instead of resolving pgrep and ps through PATH, including the comm= and args= fallback probes in posixInspectSubprocess. Linux and other POSIX hosts still use the bare utility names.

This avoids libuv’s per-PATH-entry posix_spawn lookup on macOS, which was amplifying each ~1s poll into many failed spawns when many terminals were open (fixes#6332).

A new Manager.test case asserts the first four process-runner commands per platform: absolute paths on darwin, bare names on linux.

^{Reviewed by Cursor Bugbot for commit 16eedd9. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Use absolute binary paths for `pgrep` and `ps` in terminal polling on macOS

On macOS, posixInspectSubprocess in Manager.ts now calls /usr/bin/pgrep and /bin/ps directly instead of relying on PATH resolution. On Linux, bare command names are still used. A new test in Manager.test.ts verifies the correct command is invoked per platform.

^{Macroscope summarized 16eedd9.}

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

github-actions · 2026-08-12T20:28:51Z

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	11.3 KiB	11.3 KiB	+12 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	5.5 KiB	5.5 KiB	+10 B (+0.2%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	5.9 KiB	5.9 KiB	+2 B (+0.0%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	49.7 KiB	49.7 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	16	16	0 (0.0%)	21	✅
Claude	Total thread wire	11.3 KiB	11.3 KiB	0 B (0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	5.5 KiB	5.5 KiB	−3 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	5.9 KiB	5.9 KiB	+3 B (+0.0%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	50.6 KiB	50.6 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	16	16	0 (0.0%)	21	✅

Baseline: 18918d1 · PR result: 16eedd9 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

Codex decoded thread snapshot: 94.6 KiB
Claude decoded thread snapshot: 95.4 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

macroscopeapp · 2026-08-12T20:29:38Z

Approvability

Verdict: Approved 4b71bac

Performance optimization that uses absolute paths for pgrep and ps on macOS to avoid libuv PATH search overhead. The change is small, self-contained, and well-tested with no functional impact.

No code changes detected at 16eedd9. Prior analysis still applies.

^{You can customize Macroscope's approvability policy. Learn more.}

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

Dismissing prior approval to re-evaluate 972740d

Dismissing prior approval to re-evaluate 4b71bac

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

juliusmarminge · 2026-08-15T11:31:54Z

Closing as superseded by #6377, which has merged and covers this change on the current architecture. Thanks for the contribution.

t3-code · 2026-08-12T20:26:02Z

what changed

use /usr/bin/pgrep and /bin/ps for terminal subprocess polling on macos
retain bare utility names on linux and other posix platforms
cover every macos subprocess inspection path, including the command fallback probe

why

node/libuv resolves bare executable names by trying each path entry with posix_spawn on macos. because terminal subprocess activity is polled every second, this multiplied a few logical probes into hundreds of failed spawn attempts per second on machines with many terminals and a long path.

absolute paths bypass that lookup without changing the polling behavior. this removes about 94% of the reported spawn attempts in the 14-terminal, 17-entry case.

fixes#6332

verification

vp test run src/terminal/Manager.test.ts -t 'uses absolute utility paths' --maxWorkers=1
vp test run src/terminal/Manager.test.ts --maxWorkers=1
vp test run --maxWorkers=1 in apps/server: 229 files passed, 2 skipped; 2380 tests passed, 7 skipped
vp run --filter t3 typecheck
vp check apps/server/src/terminal/Manager.ts apps/server/src/terminal/Manager.test.ts

checklist

this pr is small and focused
i explained what changed and why

request provenance

requested by: @extoci

Note

Low Risk
Behavioral change is limited to how utilities are invoked on macOS during subprocess inspection; polling logic and Linux paths are unchanged.

Overview
macOS terminal subprocess polling now invokes /usr/bin/pgrep and /bin/ps instead of resolving pgrep and ps through PATH, including the comm= and args= fallback probes in posixInspectSubprocess. Linux and other POSIX hosts still use the bare utility names.

This avoids libuv’s per-PATH-entry posix_spawn lookup on macOS, which was amplifying each ~1s poll into many failed spawns when many terminals were open (fixes#6332).

A new Manager.test case asserts the first four process-runner commands per platform: absolute paths on darwin, bare names on linux.

^{Reviewed by Cursor Bugbot for commit 16eedd9. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Use absolute binary paths for `pgrep` and `ps` in terminal polling on macOS

On macOS, posixInspectSubprocess in Manager.ts now calls /usr/bin/pgrep and /bin/ps directly instead of relying on PATH resolution. On Linux, bare command names are still used. A new test in Manager.test.ts verifies the correct command is invoked per platform.

^{Macroscope summarized 16eedd9.}

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

github-actions · 2026-08-12T20:28:51Z

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	11.3 KiB	11.3 KiB	+12 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	5.5 KiB	5.5 KiB	+10 B (+0.2%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	5.9 KiB	5.9 KiB	+2 B (+0.0%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	49.7 KiB	49.7 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	16	16	0 (0.0%)	21	✅
Claude	Total thread wire	11.3 KiB	11.3 KiB	0 B (0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	5.5 KiB	5.5 KiB	−3 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	5.9 KiB	5.9 KiB	+3 B (+0.0%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	50.6 KiB	50.6 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	16	16	0 (0.0%)	21	✅

Baseline: 18918d1 · PR result: 16eedd9 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

Codex decoded thread snapshot: 94.6 KiB
Claude decoded thread snapshot: 95.4 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

macroscopeapp · 2026-08-12T20:29:38Z

Approvability

Verdict: Approved 4b71bac

Performance optimization that uses absolute paths for pgrep and ps on macOS to avoid libuv PATH search overhead. The change is small, self-contained, and well-tested with no functional impact.

No code changes detected at 16eedd9. Prior analysis still applies.

^{You can customize Macroscope's approvability policy. Learn more.}

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

Dismissing prior approval to re-evaluate 972740d

Dismissing prior approval to re-evaluate 4b71bac

Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

juliusmarminge · 2026-08-15T11:31:54Z

Closing as superseded by #6377, which has merged and covers this change on the current architecture. Thanks for the contribution.

fix(server): avoid macos path walks in terminal polling
96b5f88
Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

github-actionsBot added the size:S 10-29 changed lines (additions + deletions). label Aug 12, 2026

macroscopeappBot previously approved these changes Aug 12, 2026
View reviewed changes

docs(server): explain macos utility paths
972740d
Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

github-actionsBot added the vouch:trusted PR author is trusted by repo permissions or the VOUCHED list. label Aug 12, 2026

macroscopeappBot previously approved these changes Aug 12, 2026
View reviewed changes

macroscopeappBot approved these changes Aug 12, 2026
View reviewed changes

test(server): preserve posix utility lookup behavior
16eedd9
Co-authored-by: Exotic <118054752+extoci@users.noreply.github.com>

t3-codeBotforce-pushed the fix/macos-terminal-poller-paths branch from 4b71bac to 16eedd9Compare August 12, 2026 20:40

juliusmarminge closed this Aug 15, 2026