The fastest, cheapest, most efficient open-source agent harness. Run more than 500 agents on a $10 VPS.
Rust agent-orchestrationai-agentsai-assistantdesktopllmlocal-firstmcppersonal-aiprivacyrustsecond-braintauri
test(regression): update stack overflow test paths
Steven Enamakela098fcc17db3
diff --git a/tests/composio_list_tools_stack_overflow_regression.rs b/tests/composio_list_tools_stack_overflow_regression.rsindex 5169296e27..9c04e75675 100644--- a/tests/composio_list_tools_stack_overflow_regression.rs+++ b/tests/composio_list_tools_stack_overflow_regression.rs@@ -21,7 +21,7 @@//! ← Config::load_or_init_with_env_lookup//! ← config::ops::load_config_with_timeout//! ← ComposioListToolsTool::execute-//! ← subagent_runner::run_inner_loop / run_typed_mode / run_subagent+//! ← subagent runner inner loop / run_typed_mode / direct child host//! ← SkillDelegationTool::execute (delegate_to_integrations_agent, since removed)//! ← Agent::execute_tool_call / execute_tools / turn//! ← web_chat::run_chat_task@@ -55,7 +55,7 @@//! rebuild the upper chat-channel layers (`web_chat:://! run_chat_task → Agent::turn → execute_tools → <delegation tool>`)//! without standing up an HTTP + Socket.IO stack. We drive the production-//! path from `run_subagent` downward — i.e. everything below a delegation+//! path from direct child execution downward — i.e. everything below a delegation//! tool's `execute` — on a small tokio worker stack (`SUBAGENT_TOWER_STACK_BYTES`).//!//! **Caveat — what this test does and does not catch.** Because the@@ -70,7 +70,7 @@//! * **What this test does NOT catch:** a re-introduction of the//! specific stack-overflow trigger (re-inlining the TOML parse onto//! the worker). The production crash needed the full tower above-//! `run_subagent` to push 2 MB over.+//! direct child execution to push 2 MB over.//!//! The actual stack-overflow fix is in//! [`crates/openhuman-core/src/config/schema/load.rs`](`parse_config_with_recovery`)@@ -93,7 +93,7 @@//! frame growth in the tower is caught long before it threatens production,//! * `OPENHUMAN_WORKSPACE` pointed at a tempdir with a representative//! `config.toml` so the TOML parser does real work,-//! * `run_subagent` on a test-owned worker exactly like a delegation tool+//! * direct child execution on a test-owned worker exactly like a delegation tool//! does, with a stubbed `ChatModel`//! that emits one `composio_list_tools` tool call on iteration 1//! and stops on iteration 2.@@ -128,7 +128,7 @@ use tinytools_agent::dialect::ToolCallFormat;/// `AGENT_WORKER_STACK_BYTES` (16 MiB, which stays the real budget).////// Measured in an unoptimised build (#6379): before the async-frame work the-/// sub-agent tower (`run_subagent` → `run_subagent_direct` → config load →+/// sub-agent tower (direct host lifecycle → child execution → config load →/// `Config::save`) needed between 3.5 and 4 MiB; it overflowed at 3 MiB. It/// now peaks near 1.6 MiB (passes at 1664 KiB, overflows at 1536 KiB). 3 MiB/// leaves ~85% headroom for unrelated growth while still failing on the old@@ -399,7 +399,7 @@ async fn drive_subagent() {.expect("vision worker built-in must exist").clone();// Give the worker a test-owned identity and a narrow, neutral surface.- // Inherit keeps the stub model above on the real run_subagent path.+ // Inherit keeps the stub model above on the direct child host path.def.id = "stack_regression_worker".into();def.display_name = Some("Stack regression worker".into());def.when_to_use = "Read the supplied request for this stack regression.".into();