feat(goose-acp): enable parallel sessions with isolated agent state #6392

codefromthecrypt · 2026-01-08T00:47:54Z

Summary

This PR makes it possible to run ACP agents and sessions in parallel, with critical state isolated to each agent instance.

Before, we were using global singletons for most state, which made it impossible to change configuration without ENV or subprocess isolation. Moreover, unlike platform, built-in extensions launched subprocesses which implied you need a binary entrypoint to run them.

This PR extracts the most critical state into structs and ensures consistent threading to call sites. It also reconfigures built-in extensions to run in-process, via duplex streams.

Legacy entrypoints retain global singletons despite ACP entrypoints being isolated. This was done to allow progress, particularly to implement goose-server with ACP. This is also due to the pending PR to replace the CLI. We don't want to do too much work for code that is going to be replaced.

The most important impact is the new goose-acp crate which contains isolated code for critical paths even though the CLI entrypoint that uses it will act the same.

Later in the tests section you will see how the combination of these changes allow more ACP integration tests than before to complete in less than a second.

Type of Change

Feature
Refactor / Code quality
Tests

AI Assistance

This PR was created or reviewed with AI assistance

Testing

Now that built-in are in-process like platform extensions, we don't need to build binaries to run ACP tests.
Since we parameterized the GooseAcpAgent to accept variables via AgentConfig, tests don't rely on ENV variables.

Above and other changes allow us to safely run tests in parallel, removing env-lock, serial and reset patterns.
This means despite having more ACP integration tests, our performance is incredibly faster.

Crate	MAIN	THIS PR	Speedup
goose-acp	42.22s	0.77s	55x

Works cited

What this PR does is a step towards a more modular codebase which avoids global singletons.
Some projects I studied for reference were:

hyperswitch - AppState takes storage config
lemmy - LemmyContext created once, passed to handlers
typst - Engine requires all dependencies at construction

Related Issues

feat: acp client provider #6305

codefromthecrypt

notes on the important things

codefromthecrypt · 2026-01-08T00:51:55Z

crates/goose/tests/mcp_replays/github-mcp-serverstdio

 STDIN: {"jsonrpc":"2.0","method":"notifications/initialized"}
 STDERR: time=2025-12-11T17:58:47.642-05:00 level=INFO msg="session initialized"
-STDIN: {"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"_meta":{"progressToken":0},"name":"get_file_contents","arguments":{"owner":"block","path":"README.md","repo":"goose","sha":"ab62b863c1666232a67048b6c4e10007a2a5b83c"}}}
+STDIN: {"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"_meta":{"goose-session-id":"test-session-id","progressToken":0},"name":"get_file_contents","arguments":{"owner":"block","path":"README.md","repo":"goose","sha":"ab62b863c1666232a67048b6c4e10007a2a5b83c"}}}


now that session is propagated via parameter in the case we always know it, the expectation file bumped for good reason.

codefromthecrypt · 2026-01-08T00:51:55Z

crates/goose/tests/mcp_integration_test.rs

    }
 }

+fn build_and_get_binary_path() -> PathBuf {


now that builtin extensions no longer require being launched as subprocesses (mcp xxx args), ACP doesn't need a build to run tests. This leaves only one path where we launch processes in tests, the old one for MCP replay, only used when recording. This is put back to exactly the same code from before we needed to extract a common cargo build utility.

codefromthecrypt · 2026-01-08T00:51:55Z

crates/goose/src/scheduler_trait.rs

    ) -> Result<Option<(String, DateTime<Utc>)>, SchedulerError>;
 }
+
+const UNAVAILABLE_MESSAGE: &str = "Scheduler not available. This tool only works in server mode.";


before we had a single call site we noticed scheduler in use and disabled it for CLI sessions. This extracts so it isn't optional. Unless there's a good reason to have CLI sessions fail on scheduler, I recommend in this PR or the next just letting them work and removing this workaround and/or only using it in tests

I'm not sure I completely follow; the reason the CLI didn't have the scheduler is that a CLI session just only runs that session so it is never going to run scheduled jobs. we could still allow the CLI to set-up scheduled jobs, but you'd need the desktop to actually trigger them if that makes sense

codefromthecrypt · 2026-01-08T00:51:55Z

crates/goose/src/session/session_manager.rs


-static SESSION_STORAGE: OnceCell<Arc<SessionStorage>> = OnceCell::const_new();
+static SESSION_STORAGE: LazyLock<Arc<SessionStorage>> =
+    LazyLock::new(|| Arc::new(SessionStorage::new(Paths::data_dir())));


in order to propagate and also still have a singleton I looked around at about 10 rust projects and this lazy lock pattern seemed the best, as old sites still get a global singleton until/unless updated, and new can propagate through in all cases the actual connect stuff is deferred to first.

codefromthecrypt · 2026-01-08T00:51:55Z

crates/goose/src/permission/permission_confirmation.rs

    AllowOnce,
    Cancel,
    DenyOnce,
+    AlwaysDeny,


we were missing this even though our permissions system allows persistence of deny. This is needed for ACP and the motivation was testing permissions which had a bug and they now are testable without clobbering or ENV

codefromthecrypt · 2026-01-08T00:51:55Z

crates/goose-acp/tests/server_test.rs

+        .unwrap();
+}
+
+#[test_case(Some(PermissionOptionKind::AllowAlways), ToolCallStatus::Completed, "user:\n  always_allow:\n  - lookup__get_code\n  ask_before: []\n  never_allow: []\n"; "allow_always")]


this is the new test and notice no ACP test requires serial anymore, and none use ENV or clobber eachother's database or permissions files

codefromthecrypt · 2026-01-08T00:51:56Z

crates/goose/src/config/permission.rs


+const PERMISSION_FILE: &str = "permission.yaml";
+
+static PERMISSION_MANAGER: LazyLock<Arc<PermissionManager>> =


permission manager needs to be a singleton where used, so I used exact same pattern as session storage. before depending on call site reads might come from memory which could have problems when not shared. now it is always shared and pushed down consistently

Copilot

Pull request overview

This PR enables parallel ACP agent sessions with isolated state by refactoring global singletons into instance-based configuration. Key changes include:

Introduces AgentConfig struct to encapsulate session manager, permission manager, mode, and scheduler
Refactors SessionManager from static methods to instance-based with lazy initialization
Refactors PermissionManager from mutable singleton to thread-safe Arc-wrapped instance
Creates new goose-acp crate for isolated ACP server implementation
Moves built-in MCP extensions to in-process execution via duplex streams
Adds session_id parameter to MCP client trait methods for proper session tracking
Achieves 55x speedup in ACP integration tests through parallel execution

Reviewed changes

Copilot reviewed 79 out of 88 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
crates/goose/src/agents/agent.rs	Introduces AgentConfig, refactors Agent to use instance-based dependencies
crates/goose/src/session/session_manager.rs	Converts from static methods to instance-based with lazy pool initialization
crates/goose/src/config/permission.rs	Makes thread-safe with RwLock, converts to instance-based
crates/goose/src/agents/extension_manager.rs	Updates context to hold SessionManager, supports in-process builtins
crates/goose/src/agents/mcp_client.rs	Adds session_id parameter to call_tool and makes trait methods have default impls
crates/goose-acp/src/server.rs	New isolated ACP server implementation (not shown in diff)
crates/goose/tests/*	Updates tests to use isolated instances, removes serial execution
crates/goose-mcp/src/lib.rs	Adds builtin extension registry for in-process spawning
crates/goose-server/src/routes/*	Updates all routes to use SessionManager from AppState
crates/goose-cli/src/*	Updates CLI to use instance-based managers

michaelneale · 2026-01-08T01:13:28Z

crates/goose-bench/Cargo.toml


 [dependencies]
-anyhow = "1.0"
+anyhow = { workspace = true }


nice - this does what I think it does which makes good sense.

michaelneale · 2026-01-08T01:48:32Z

so far working really well!

michaelneale · 2026-01-08T03:45:58Z

I also like that it is half as many processes for usual workloads. I want to stamp this but need to work out how to get enough consensus but LGTM - I will try next running it bundled

codefromthecrypt · 2026-01-08T04:12:00Z

ps confirmed in zed the missing "reject always" works and updates my permissions file properly

$ cat ~/.config/goose/permission.yaml
user:
  always_allow: []
  ask_before: []
  never_allow:
  - kiwi__search-flight

codefromthecrypt · 2026-01-08T04:16:53Z

ps verified there's no bug with us about the double check vs double x on deny. I'll raise a PR with zed to fix that, right now the only icon they have for double is approve

codefromthecrypt · 2026-01-08T05:20:05Z

opened zed-industries/zed#46329 on zed for the unrelated UI glitch I noticed testing this

Copilot

Pull request overview

Copilot reviewed 79 out of 88 changed files in this pull request and generated no new comments.

codefromthecrypt · 2026-01-08T06:23:10Z

@angiejones @alexhancock FYI I tested in webstorm (jetbrains) and it works. One note is that current version of Jetbrains doesn't respond to permissions requests like Zed, etc. So, don't change the GOOSE_MODE default from auto until that's not the case.

codefromthecrypt · 2026-01-08T06:25:06Z

also checked 3rd party MCP also works in jetbrains same permissions caveat which isn't new to this PR and probably already solved, just not released in latest jetbrains.

codefromthecrypt · 2026-01-09T05:34:25Z

rebased no code changes since last review otherwise

Copilot

Pull request overview

Copilot reviewed 79 out of 88 changed files in this pull request and generated no new comments.

Copilot

Pull request overview

Copilot reviewed 77 out of 86 changed files in this pull request and generated no new comments.

codefromthecrypt · 2026-01-14T03:09:29Z

this PR takes quite a lot of effort to rebase every couple days. let me know if there is anything I can do to get this in so my "goose time" isn't dominated by rebasing.

Signed-off-by: Adrian Cole <[email protected]>

Copilot

Pull request overview

Copilot reviewed 77 out of 86 changed files in this pull request and generated no new comments.

baxen

LGTM, a lot of great changes in here for the architecture to enable this

A lot of my comments i think can be followups, along with eventually getting the rest of the global state out like you mentioned (should make the desktop app multi chat less confusing too!)

baxen · 2026-01-14T06:32:00Z

crates/goose/src/agents/mcp_client.rs

        &self,
        name: &str,
        arguments: Option<JsonObject>,
+        session_id: &str,


I get why we need session id now in so many downstream tools (no more consistent way to lookup the process's current session). I like how it becomes _meta in the MCP spec. Can be a followup, but i'd cnsider having a "meta" arg in call_tool instead of session_id and always add session id into it

good point. when we get around to otel we'll need to implicitly add trace IDs, but maybe we wouldn't if we were passing meta (since we could do it manually just like we are for session ID)

baxen · 2026-01-14T06:33:05Z

crates/goose/src/config/permission.rs

+    pub fn new(config_dir: PathBuf) -> Self {
+        let permission_path = config_dir.join(PERMISSION_FILE);
+        let permission_map = if permission_path.exists() {
+            let file_contents = fs::read_to_string(&permission_path).unwrap();


nit: unwrap -> expect (old code had it)

baxen · 2026-01-14T06:35:21Z

crates/goose/src/config/permission.rs

            serde_yaml::from_str(&file_contents).unwrap_or_else(|_| HashMap::new())
        } else {
-            HashMap::new() // No config file, create an empty map
+            fs::create_dir_all(&config_dir).unwrap();


nit: missing why we need to create dir here? not a serious concern though

we have a choice to ensure the directory is there in the beginning or later when we persist. Not all code paths to this point ensure the config dir is there.. I wanted to put a TODO why but I have noticed not everyone likes to have TODOs.. if it were a TODO, I would say TODO: ensure all XDG paths are ensured at all entrypoints

baxen · 2026-01-14T06:42:30Z

crates/goose/src/session/session_manager.rs

-    }
+    fn create_pool(path: &Path) -> Pool<Sqlite> {
+        if let Some(parent) = path.parent() {
+            fs::create_dir_all(parent).unwrap();


nit: could use an expect here for a better error message (we had a lot of access issues with some peoples xdg dirs)

baxen · 2026-01-14T06:51:22Z

crates/goose-mcp/src/lib.rs

+    pub spawn_server: SpawnServerFn,
+}
+
+fn spawn_and_serve<S>(


this is great, really nice to clean up the processes we were leaving here

baxen · 2026-01-14T06:52:18Z

crates/goose-acp/Cargo.toml

+goose = { path = "../goose" }
+goose-mcp = { path = "../goose-mcp" }
+rmcp = { workspace = true }
+sacp = "10.1.0"


discussed in discord, will bring this to the official rust sdk once they finish their refactor

codefromthecrypt · 2026-01-14T11:32:46Z

thanks for the look @baxen I added a commit which I think got the bulk of things

Signed-off-by: Adrian Cole <[email protected]>

Copilot

Pull request overview

Copilot reviewed 77 out of 86 changed files in this pull request and generated 2 comments.

Comments suppressed due to low confidence (1)

crates/goose-acp/src/server.rs:1

Avoid cloning session_manager on every call. Store a reference in GooseAcpAgent struct or use Arc::clone(&self.agent.config.session_manager) for clarity that you're cloning the Arc, not the SessionManager.

use anyhow::Result;

Copilot · 2026-01-14T11:36:38Z

crates/goose/src/agents/agent.rs

+        let provider_name = provider.get_name().to_string();
+        let model_config = provider.get_model_config();
+
        let mut current_provider = self.provider.lock().await;
-        *current_provider = Some(provider.clone());
+        *current_provider = Some(provider);


Extract provider metadata before acquiring lock to minimize lock hold time. Move lines 1468-1469 after line 1472 since provider is moved into the lock.

Copilot · 2026-01-14T11:36:38Z

crates/goose/src/session/session_manager.rs

-        let storage = Self { pool };
-        storage.run_migrations().await?;
-        Ok(storage)
+    async fn pool(&self) -> Result<&Pool<Sqlite>> {


Method name 'pool' is ambiguous - it returns a reference to the pool after initialization. Rename to 'get_pool' or 'ensure_initialized_pool' to clarify its purpose.

Suggested change

async fn pool(&self) -> Result<&Pool<Sqlite>> {

async fn get_pool(&self) -> Result<&Pool<Sqlite>> {

DOsinga · 2026-01-14T12:58:36Z

oops this totally slipped my mind. I'll have a look too since this is kinda big?

DOsinga · 2026-01-14T18:55:03Z

I went through this once. it's clearly an improvement and we should possibly just merge as is, but after this I think the architecture would be left in a slightly confusing state.

This is solid progress. The dependency injection via AgentConfig is the right move, and making the core components instantiable (vs global singletons) unblocks library mode. Let's merge this forward.

What Works Well

AgentConfig pattern - bundling SessionManager, PermissionManager, scheduler into config for DI
Agent.with_config() - enables multiple independent instances
Threading session_id through methods - makes Agent more session-agnostic
Test improvements - no more #[serial] or unsafe reset hacks

What's Confusing

AgentManager + Agent both have SessionManager

pub struct AgentManager {
sessions: LruCache<String, Arc>, // session_id -> Agent
session_manager: Arc, // Also has session manager
}

pub struct Agent {
config: AgentConfig {
session_manager: Arc, // Agent also has it
}
}

AgentManager finds the Agent for a session_id, but Agent also needs SessionManager. This double indirection is confusing - why does AgentManager need to know about sessions at all if it's just caching agents?

What goes in AgentConfig vs on Agent?

pub struct AgentConfig {
session_manager, permission_manager, scheduler, goose_mode // In config
}

pub struct Agent {
provider: SharedProvider, // NOT in config
extension_manager: Arc, // NOT in config
config: AgentConfig,
}

Why are provider and extension_manager outside AgentConfig? What's the principle for what goes where?

Two competing patterns

goose-server: Uses AgentManager, creates 1 Agent per session
goose-acp: Bypasses AgentManager, uses 1 Agent for all sessions

Both work, but it's unclear when to use which pattern or what Agent's relationship to sessions is meant to be.

Provider in multiple places

Agent.provider (mutable via update_provider)
Session.provider (string stored in DB)
GooseAcpConfig.provider (for initialization)

What's the source of truth?

In general though this does feel like a good improvement. longer term though I think we might want to deprecate the whole notion of agents as a thing. we have sessions and an extension manager. then when you want to do something you can start an agent loop with a session and your preferences (goose mode, number of turns?). it then asks the (global?) extension manager to get it the extensions it needs. for stateless mcp, it can just reuse what is already there and that should be the general case. maybe sometimes it needs to start an extra one if one is busy. but other than that, the agent loop can just be a function without any agent state. thoughts?

codefromthecrypt · 2026-01-14T22:21:53Z

@DOsinga I agree things aren't yet coherent and there are still some mixed responsibilities.. there are a lot of call sites to old constructors and in some cases agent == agent+session

I was told that we are killing the old CLI which is a large part of some copy/paste stuff, and after that it would be goose-server+acp only, let's do another major arch pass when that's in!

* main: fixed 0 token in openrouter steaming (#6493) feat(goose-acp): enable parallel sessions with isolated agent state (#6392) copilot instruction to flag prelease docs (#6504) docs: acp mcp support (#6491) feat: add flatpak support for linux (#6387) fix(code_execution): serialize record_result output as JSON (#6495) perf(google): avoid accumulating thoughtSignatures across conversation history (#6462) fix(openai): make tool_call arguments optional and fix silent stream termination (#6309) fix: Improve error messages for invalid tool calls (#6483)

…ased * 'main' of github.com:block/goose: Fix popular topics not starting chat when clicked (#6508) fix[desktop]: deeplink ui repeat on refresh (#6469) fixed test compilation on main branch (#6512) fix: correctly parse extension name from tool call for MCP apps (#6482) fixed 0 token in openrouter steaming (#6493) feat(goose-acp): enable parallel sessions with isolated agent state (#6392) copilot instruction to flag prelease docs (#6504) docs: acp mcp support (#6491) feat: add flatpak support for linux (#6387)

* 'main' of github.com:block/goose: (28 commits) chore(deps): bump aiohttp from 3.13.0 to 3.13.3 in /scripts/provider-error-proxy (#6539) chore(deps): bump brotli from 1.1.0 to 1.2.0 in /scripts/provider-error-proxy (#6538) docs: temp correction for agent directory (#6544) chore: upgrade rmcp (#6516) docs: clarify directory in /documentation readme (#6541) Release 1.20.0 Standalone mcp apps (#6458) don't add escaping to the command field (#6519) Fix popular topics not starting chat when clicked (#6508) fix[desktop]: deeplink ui repeat on refresh (#6469) fixed test compilation on main branch (#6512) fix: correctly parse extension name from tool call for MCP apps (#6482) fixed 0 token in openrouter steaming (#6493) feat(goose-acp): enable parallel sessions with isolated agent state (#6392) copilot instruction to flag prelease docs (#6504) docs: acp mcp support (#6491) feat: add flatpak support for linux (#6387) fix(code_execution): serialize record_result output as JSON (#6495) perf(google): avoid accumulating thoughtSignatures across conversation history (#6462) fix(openai): make tool_call arguments optional and fix silent stream termination (#6309) ...

Copilot AI review requested due to automatic review settings January 8, 2026 00:47

Copilot started reviewing on behalf of codefromthecrypt January 8, 2026 00:48 View session

codefromthecrypt mentioned this pull request Jan 8, 2026

feat(goose-acp): enable parallel sessions with isolated agent state #6363

Closed

4 tasks

codefromthecrypt commented Jan 8, 2026

View reviewed changes

Copilot AI reviewed Jan 8, 2026

View reviewed changes

michaelneale reviewed Jan 8, 2026

View reviewed changes

Copilot AI review requested due to automatic review settings January 8, 2026 05:47

Copilot started reviewing on behalf of codefromthecrypt January 8, 2026 05:47 View session

Copilot AI reviewed Jan 8, 2026

View reviewed changes

codefromthecrypt force-pushed the feat/scacp-10-library-mode branch from 40697bb to 67df032 Compare January 9, 2026 05:33

codefromthecrypt force-pushed the feat/scacp-10-library-mode branch from 67df032 to 5cd195d Compare January 9, 2026 05:46

Copilot AI review requested due to automatic review settings January 9, 2026 05:46

Copilot started reviewing on behalf of codefromthecrypt January 9, 2026 05:47 View session

Copilot AI reviewed Jan 9, 2026

View reviewed changes

codefromthecrypt force-pushed the feat/scacp-10-library-mode branch from 5cd195d to 8060238 Compare January 9, 2026 06:23

Copilot AI review requested due to automatic review settings January 14, 2026 03:01

codefromthecrypt force-pushed the feat/scacp-10-library-mode branch from 8060238 to 9fce44a Compare January 14, 2026 03:01

Copilot started reviewing on behalf of codefromthecrypt January 14, 2026 03:02 View session

Copilot AI reviewed Jan 14, 2026

View reviewed changes

codefromthecrypt force-pushed the feat/scacp-10-library-mode branch from 9fce44a to 8f937b7 Compare January 14, 2026 03:07

codefromthecrypt requested a review from baxen January 14, 2026 03:08

codefromthecrypt force-pushed the feat/scacp-10-library-mode branch from 8f937b7 to 69383fe Compare January 14, 2026 03:09

feat(goose-acp): enable parallel sessions with isolated agent state

835aab0

Signed-off-by: Adrian Cole <[email protected]>

codefromthecrypt force-pushed the feat/scacp-10-library-mode branch from 69383fe to 835aab0 Compare January 14, 2026 03:13

Copilot AI review requested due to automatic review settings January 14, 2026 03:13

Copilot started reviewing on behalf of codefromthecrypt January 14, 2026 03:13 View session

Copilot AI reviewed Jan 14, 2026

View reviewed changes

baxen approved these changes Jan 14, 2026

View reviewed changes

codefromthecrypt force-pushed the feat/scacp-10-library-mode branch from d534580 to 5226528 Compare January 14, 2026 11:35

feedback

05a43fd

Signed-off-by: Adrian Cole <[email protected]>

Copilot AI review requested due to automatic review settings January 14, 2026 11:35

codefromthecrypt force-pushed the feat/scacp-10-library-mode branch from 5226528 to 05a43fd Compare January 14, 2026 11:35

Copilot AI reviewed Jan 14, 2026

View reviewed changes

baxen merged commit 7d4a6bd into main Jan 14, 2026
20 checks passed

baxen deleted the feat/scacp-10-library-mode branch January 14, 2026 22:19

rabi mentioned this pull request Jan 15, 2026

fix(code_execution): update test after PlatformExtensionContext API change #6510

Closed

9 tasks


		const PERMISSION_FILE: &str = "permission.yaml";

		static PERMISSION_MANAGER: LazyLock<Arc<PermissionManager>> =

	async fn pool(&self) -> Result<&Pool<Sqlite>> {
	async fn get_pool(&self) -> Result<&Pool<Sqlite>> {

feat(goose-acp): enable parallel sessions with isolated agent state #6392

feat(goose-acp): enable parallel sessions with isolated agent state #6392

Uh oh!

Conversation

codefromthecrypt commented Jan 8, 2026

Summary

Type of Change

AI Assistance

Testing

Works cited

Related Issues

Uh oh!

codefromthecrypt left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Choose a reason for hiding this comment

Uh oh!

michaelneale commented Jan 8, 2026

Uh oh!

michaelneale commented Jan 8, 2026

Uh oh!

codefromthecrypt commented Jan 8, 2026

Uh oh!

codefromthecrypt commented Jan 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codefromthecrypt commented Jan 8, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

codefromthecrypt commented Jan 8, 2026

Uh oh!

codefromthecrypt commented Jan 8, 2026

Uh oh!

codefromthecrypt commented Jan 9, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

codefromthecrypt commented Jan 14, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

baxen left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

codefromthecrypt commented Jan 8, 2026 •

edited

Loading