fix(host): remove the socket directory once it holds no socket by ThomasK33 · Pull Request #239 · coder/agent-tty

ThomasK33 · 2026-11-02T14:37:42Z

Why

The operator wants the open issues he filed in coder/agent-tty fixed and landed. #227 reports that a harness with one throwaway Home per run leaves an empty folder under /tmp/agent-tty/ per run (11,005 on one shared host) and a stale Home Registry entry.

What changed

removeSocketDirectoryIfEmpty runs rmdir on the Home's socket folder after each socket unlink. It runs in destroy (recordDestroyed), in reconcileSession after a host crash, and in the host's own shutdown. A folder that still holds another session's socket (ENOTEMPTY or EEXIST) or is already gone (ENOENT) stays as it is.
listenInSocketDirectory runs mkdir, chmod and listen up to 5 times. It retries when the folder vanished in between, because a teardown in the same Home removed it. libuv reports a bind in a missing folder as EACCES. The host and doctor's socket check both use it, and doctor also removes the folder after its check.

Scope

Covers item 1 of destroy leaves an empty socket directory and a stale home registry entry #227. Item 2 is already on main: plain gc deregisters registered Homes whose folder is gone, since feat(home-registry): Home Registry + dashboard Home picker #131. A live check on main showed the stale entry gone after gc and the Home in deregisteredHomes. gc --home and AGENT_TTY_HOME stay scoped to one Home and do not touch the registry, by design.
CLI JSON, error codes and exit codes do not change.
A host that crashes after its terminal manifest write but before it unlinks its socket still leaves that socket behind, as on main. destroy cleans it up, and changing reconcileSession for terminal manifests is a separate change.

Blast Radius

Session start-up and teardown of every session. The removal never deletes a socket, because rmdir fails on a folder that is not empty. The host's listen retry covers the one new window, where a folder vanishes between mkdir and the bind.

Verification

test/integration/socket-directory.test.ts: 8 tests fail on main and pass here. They cover destroy with two sessions in one Home, destroy --force, a host whose program exits, a reconcile after a host crash, doctor with and without a folder that vanishes before its bind, and a folder that a preload removes before the host's chmod or bind (on main, create exits 6).
Live A/B on the CLI: main gives destroy: socket dir kept (0 entries) FAIL, this branch gives destroy: socket dir removed PASS. 48 parallel create and destroy pairs in one Home all passed and left no folder.
Gates: npm run typecheck, npm run lint and npm run format:check exit 0. vitest run test/unit: 1595 tests pass. vitest run --maxWorkers=1 --retry=2 on test/integration (4 shards, 268 tests) and test/e2e (3 shards, 34 tests) pass. The e2e color-grid screenshot test, a known load flake, needed retry x2. The lifecycle skill's check.sh passes all 52 runs, and coverage.json is synced after the merge of main.

Closes #227

Generated with xum • Model: anthropic:claude-opus-5-5 • Thinking: xhigh

destroy, reconcile and the host's shutdown rmdir the Home's socket directory after they unlink a socket, and leave a directory that still holds one. The host retries its mkdir, chmod and bind when another session's destroy removed the directory in between. Refs #227

…leaves out

ThomasK33 · 2026-11-02T14:38:11Z

@codex review

ThomasK33 · 2026-11-02T14:38:15Z

@codex security review

chatgpt-codex-connector · 2026-11-02T14:38:32Z

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-02T16:49:59.693677Z	`e24a85b`	Manual request
🔒 Security Review	✅ Completed 2026-11-02T16:18:50.372244Z	`6acfc10`	Manual request

Security findings

Advisory findings (1)

🟡 Keep private socket directories unclaimable · Low

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

chatgpt-codex-connector · 2026-11-02T14:40:27Z

Codex Review: Didn't find any major issues. 👍

Reviewed commit: 98da29e50e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-11-02T14:41:39Z

🛡️ Codex Security Review

Security review completed. No security issues were found in this pull request.

Reviewed commit: 98da29e50e

View security finding report

Only the user who started this review can view the report in Codex.

ℹ️ About Codex security reviews in GitHub

This is an experimental Codex feature. Security reviews are triggered when:

You comment "@codex security review"
A regular code review gets triggered (for example, "@codex review" or when a PR is opened), and you’re opted in so security review runs alongside code review

Once complete, Codex will leave suggestions, or a comment if no findings are found.

ThomasK33 · 2026-11-02T15:21:38Z

@codex review

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: af1b7ec7ea

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

…t directory listenInSocketDirectory (lifecycle.ts) now runs the mkdir, chmod and listen retry for the host and for doctor's socket check. doctor's cleanup also removes the Home's socket directory once it is empty.

# Conflicts: # .agents/skills/verify-session-lifecycle/coverage.json

ThomasK33 · 2026-11-02T16:13:09Z

@codex review

ThomasK33 · 2026-11-02T16:13:20Z

@codex security review

chatgpt-codex-connector · 2026-11-02T16:16:00Z

Codex Review: Didn't find any major issues. Bravo.

Reviewed commit: 6acfc10b8f

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector

🛡️ Codex Security Review

Here are some automated security review suggestions for this pull request.

Reviewed commit: 6acfc10b8f

ℹ️ About Codex security reviews in GitHub

This is an experimental Codex feature. Security reviews are triggered when:

You comment "@codex security review"
A regular code review gets triggered (for example, "@codex review" or when a PR is opened), and you’re opted in so security review runs alongside code review

Once complete, Codex will leave suggestions, or a comment if no findings are found.

An unusual rmdir error (EBUSY, EPERM) at host shutdown or after doctor's socket check is logged at debug level instead of crashing the host after its final manifest write.

…rmdir

ThomasK33 · 2026-11-02T16:47:59Z

@codex review

chatgpt-codex-connector · 2026-11-02T16:49:58Z

Codex Review: Didn't find any major issues. Delightful!

Reviewed commit: e24a85b3cb

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

ThomasK33 added 6 commits November 2, 2026 14:14

test: cover socket directory removal and a vanished directory at bind

f81f46e

test: cover a socket directory that vanishes before the chmod

1f53364

docs(lifecycle-skill): describe the socket directory steps the model …

68f77ea

…leaves out

chore(lifecycle-skill): sync coverage after the socket directory change

5a9f405

refactor(host): keep only the libuv note on the socket retry

98da29e

test: pin the socket directory removal of destroy --force

af1b7ec

chatgpt-codex-connector Bot reviewed Nov 2, 2026

View reviewed changes

Comment thread src/host/lifecycle.ts Resolved

Comment thread src/host/lifecycle.ts Resolved

Comment thread src/host/hostMain.ts Resolved

ThomasK33 added 4 commits November 2, 2026 15:33

test: cover the socket directory of doctor's socket check

798058c

Merge remote-tracking branch 'origin/main' into fix/227-socket-dir

57dcf06

# Conflicts: # .agents/skills/verify-session-lifecycle/coverage.json

chore(lifecycle-skill): sync coverage after merging main

6acfc10

chatgpt-codex-connector Bot reviewed Nov 2, 2026

View reviewed changes

Comment thread src/host/lifecycle.ts Resolved

ThomasK33 added 4 commits November 2, 2026 16:40

test: a host whose socket directory rmdir fails still exits cleanly

7024660

fix(host): remove the socket directory best-effort at shutdown

7d4d6ad

An unusual rmdir error (EBUSY, EPERM) at host shutdown or after doctor's socket check is logged at debug level instead of crashing the host after its final manifest write.

chore(lifecycle-skill): sync coverage after the best-effort shutdown …

ef9e0bd

…rmdir

Merge remote-tracking branch 'origin/main' into fix/227-socket-dir

e24a85b

ThomasK33 added this pull request to the merge queue Nov 2, 2026

Merged via the queue into main with commit 58e5a6a Nov 2, 2026
12 checks passed

ThomasK33 deleted the fix/227-socket-dir branch November 2, 2026 17:29

This was referenced Nov 2, 2026

fix(batch): default a wait step to the 30 s wait timeout #250

Merged

host sockets: a crashed host's socket outlives a terminal manifest; use an owner-private socket root #252

Closed

github-actions Bot mentioned this pull request Nov 2, 2026

chore(release): 0.6.0 #172

Open

ThomasK33 mentioned this pull request Nov 3, 2026

fix(host): keep sockets under a per-uid root and remove a dead host socket #288

Merged

Conversation

ThomasK33 commented Nov 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Why

What changed

Scope

Blast Radius

Verification

Uh oh!

ThomasK33 commented Nov 2, 2026

Uh oh!

ThomasK33 commented Nov 2, 2026

Uh oh!

chatgpt-codex-connector Bot commented Nov 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codex Review Summary

Security findings

Advisory findings (1)

Uh oh!

chatgpt-codex-connector Bot commented Nov 2, 2026

Uh oh!

chatgpt-codex-connector Bot commented Nov 2, 2026

🛡️ Codex Security Review

Uh oh!

ThomasK33 commented Nov 2, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ThomasK33 commented Nov 2, 2026

Uh oh!

ThomasK33 commented Nov 2, 2026

Uh oh!

chatgpt-codex-connector Bot commented Nov 2, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

🛡️ Codex Security Review

Uh oh!

Uh oh!

ThomasK33 commented Nov 2, 2026

Uh oh!

chatgpt-codex-connector Bot commented Nov 2, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

ThomasK33 commented Nov 2, 2026 •

edited

Loading

chatgpt-codex-connector Bot commented Nov 2, 2026 •

edited

Loading