SIGSEGV in discord::Handler::handle_config_command after connect — 0.10.0-beta.4 release binary (macOS x86_64) #1535

Summary

openab (release asset from openab-0.10.0-beta.4, x86_64-apple-darwin) crashes with SIGSEGV ~1–2 seconds after registered global slash commands, during the post-connect command/interaction sync. 100% reproducible on every start for 3 of 6 bot instances; the other 3 instances on the same binary run fine.

Rolling back to a previous build (Sept 7 binary, also reporting 0.10.0) eliminates the crash — affected instances connect and dispatch normally.

Environment

• OS: macOS 12.7.6 (Darwin 21.6.0), x86_64
• openab: openab-0.10.0-beta.4 release asset (x86_64-apple-darwin.tgz), binary reports openab 0.10.0
• Setup: 6 Discord bot instances, each with own config + agent_cmd (Devin ACP wrappers / CLI shims)

Crash details

exception: EXC_BAD_ACCESS (SIGSEGV)
subtype: KERN_INVALID_ADDRESS at 0x00000000fffff9d0
termination: Segmentation fault: 11 (signal 11)
faultingThread: 3 (tokio worker)

Faulting thread stack (from .ips):

[0] openab :: openab_core::discord::Handler::handle_config_command::{{closure}}

(handle_config_command is the handler for the /models and /agents global slash commands per discord.rs.)

Log before crash

INFO openab: config loaded agent_cmd=... pool_max=10 discord=true
INFO openab: discord bot running
INFO openab_core::discord: discord bot connected user=Swe(Devin)
INFO openab_core::discord: registered global slash commands
<crash ~1s later — during guild command sync HTTP calls>

RUST_LOG=trace shows the process dies in the middle of Discord HTTP traffic immediately after global command registration. Same signature across 10+ .ips reports (all handle_config_command).

What we tried (did NOT help)

• Deleted /models and /agents global slash commands via Discord API (HTTP 204) — crash persisted (they get re-registered on connect anyway)
• Cleared instance state dir (.openab/) — crash persisted
• Waited >15 min for pending interactions to expire — crash persisted

Hypothesis

handle_config_command may be invoked during startup command reconciliation or on a stale queued interaction. The fault address 0xfffff9d0 looks like a null/dangling pointer + offset dereference. Instances that crash are the ones whose channels have had /models or /agents invoked recently; unaffected instances likely have no such pending interaction — but deleting the global commands did not stop it, so the trigger may live in guild command sync rather than the interaction queue.

Related

• SIGSEGV on x86_64 macOS after commit 843bb729 (observer/lobby Phase 1 scaffold) — Discord gateway crash loop #1531 — also SIGSEGV on x86_64 macOS in the Discord path, but a different signature (miniz_oxide inside serenity ShardRunner, custom build from commit 843bb72). This report is for the released beta.4 binary.

Happy to share the full .ips crash reports or run a debug build if pointed at instructions.