- Concurrency soak (exit criterion): 12 concurrent gated runs, every decision attempted twice concurrently, explicit resumes racing the durable sweeper — exactly one execution per approval, grants consumed at most once, every decision audited, zero stuck runs, zero unaudited executions. (Testkit pool raised to 20 connections; the 5-connection pool starved the storm.) - axe a11y sweep (exit criterion): serious+critical violations fail CI on login, shell, chat, computer home, settings app, all global pages, and the wizard. Two real violations found and fixed: aria-label on a plain div (wizard progress -> role=group) and a button directly inside a <dl> (settings -> plain bordered list). - tools/bundler (exit criterion): keygen / assemble / verify CLI — copies artifacts, writes manifest.json + sha256 checksums.txt + a detached ed25519 signature; verification is fully offline (keyless signing is internet-dependent and disqualified). Tests: round trip, tampered artifact caught by hash, tampered checksum list caught by signature, wrong key refused, missing artifact reported. 147 Rust + 63 frontend tests + 27 Playwright journeys (incl. 4 a11y). Co-Authored-By: Claude Fable 5 <[email protected]>
207 lines
6.7 KiB
Rust
207 lines
6.7 KiB
Rust
//! P6 security soak (spec §17 exit): under concurrent load with racing
|
|
//! decisions and duplicated resume attempts, ZERO unaudited gated
|
|
//! executions — every outbox row maps to exactly one approved decision,
|
|
//! every grant is consumed at most once, and nothing executes twice.
|
|
|
|
use std::sync::Arc;
|
|
use std::time::Duration;
|
|
|
|
use tc_domain::{
|
|
AccessPolicy, Agent, AgentId, AgentStatus, Role, RunState, User, UserId, Workspace, WorkspaceId,
|
|
};
|
|
use tc_llm::ScriptedProvider;
|
|
use tc_runtime::{RunEventBody, Runtime, RuntimeConfig};
|
|
use tc_safety::{approvals, Decision, ResumeReady};
|
|
|
|
const SCENARIOS: &str = r#"
|
|
[[scenario]]
|
|
marker = "[[scenario:gated-email]]"
|
|
|
|
[[scenario.turns]]
|
|
events = [
|
|
{ type = "tool_use", name = "email.send", input = { to = "[email protected]", subject = "Soak", body = "Load test." } },
|
|
]
|
|
|
|
[[scenario.turns]]
|
|
events = [
|
|
{ type = "text", text = "Done." },
|
|
]
|
|
"#;
|
|
|
|
const RUNS: usize = 12;
|
|
|
|
#[tokio::test]
|
|
async fn concurrent_gated_runs_never_execute_unaudited_or_twice() {
|
|
let pool = tc_testkit::test_pool().await;
|
|
let ws = Workspace {
|
|
id: WorkspaceId::new(),
|
|
name: "Acme".into(),
|
|
plan: "team".into(),
|
|
};
|
|
tc_db::repo::workspaces::insert(&pool, &ws).await.unwrap();
|
|
let owner = User {
|
|
id: UserId::new(),
|
|
workspace_id: ws.id,
|
|
email: format!("{}@acme.test", UserId::new()),
|
|
role: Role::Owner,
|
|
display_name: "Owner".into(),
|
|
created_at: time::OffsetDateTime::UNIX_EPOCH,
|
|
};
|
|
tc_db::repo::users::insert(&pool, &owner).await.unwrap();
|
|
let agent = Agent {
|
|
id: AgentId::new(),
|
|
workspace_id: ws.id,
|
|
name: "Scout".into(),
|
|
job_title: "Analyst".into(),
|
|
system_prompt: String::new(),
|
|
avatar: String::new(),
|
|
accent: String::new(),
|
|
wallpaper: String::new(),
|
|
managed_by: owner.id,
|
|
status: AgentStatus::Online,
|
|
};
|
|
tc_db::repo::agents::insert(&pool, &agent, &AccessPolicy::default())
|
|
.await
|
|
.unwrap();
|
|
tc_db::repo::credits::add_lot(&pool, ws.id, 10_000, "soak")
|
|
.await
|
|
.unwrap();
|
|
|
|
let rt = Runtime::new(
|
|
pool.clone(),
|
|
Arc::new(ScriptedProvider::from_toml(SCENARIOS).unwrap()),
|
|
RuntimeConfig::basic("scripted", 1024),
|
|
);
|
|
// The durable sweeper races our explicit resume calls on purpose.
|
|
rt.spawn_resume_sweeper(Duration::from_millis(25));
|
|
|
|
// Launch all runs concurrently, each in its own session.
|
|
let mut run_ids = Vec::new();
|
|
let mut handles = Vec::new();
|
|
for i in 0..RUNS {
|
|
let session = tc_db::repo::sessions::create(&pool, agent.id, ws.id, &format!("Soak {i}"))
|
|
.await
|
|
.unwrap();
|
|
let rt = rt.clone();
|
|
handles.push(tokio::spawn(async move {
|
|
let started = rt
|
|
.send_message(session.id, "send it [[scenario:gated-email]]")
|
|
.await
|
|
.unwrap();
|
|
let mut rx = started.events;
|
|
while let Ok(envelope) = rx.recv().await {
|
|
if matches!(
|
|
envelope.event,
|
|
RunEventBody::RunSuspended { .. } | RunEventBody::Error { .. }
|
|
) {
|
|
break;
|
|
}
|
|
}
|
|
started.run_id
|
|
}));
|
|
}
|
|
for handle in handles {
|
|
run_ids.push(handle.await.unwrap());
|
|
}
|
|
|
|
// All suspended with pending approvals.
|
|
let pending = approvals::list_pending(&pool, ws.id).await.unwrap();
|
|
assert_eq!(pending.len(), RUNS);
|
|
let outbox_before = sqlx::query_scalar::<_, i64>("SELECT count(*) FROM outbox")
|
|
.fetch_one(&pool)
|
|
.await
|
|
.unwrap();
|
|
assert_eq!(outbox_before, 0, "nothing may execute while pending");
|
|
|
|
// Decide concurrently: even-indexed approve, odd-indexed reject — and
|
|
// every decision is attempted TWICE concurrently (double-click storm),
|
|
// plus an explicit duplicate resume racing the sweeper.
|
|
let mut deciders = Vec::new();
|
|
for (i, approval) in pending.iter().enumerate() {
|
|
let decision = if i % 2 == 0 {
|
|
Decision::Approve
|
|
} else {
|
|
Decision::Reject
|
|
};
|
|
for _ in 0..2 {
|
|
let pool = pool.clone();
|
|
let rt = rt.clone();
|
|
let id = approval.id;
|
|
let run_id = approval.run_id;
|
|
let user = owner.id;
|
|
deciders.push(tokio::spawn(async move {
|
|
let decided = approvals::decide(&pool, id, user, decision).await;
|
|
// Exactly one of the two concurrent attempts wins.
|
|
let _ = rt
|
|
.resume_run(ResumeReady {
|
|
run_id,
|
|
approval_id: id,
|
|
approved: decision == Decision::Approve,
|
|
})
|
|
.await;
|
|
decided.is_ok()
|
|
}));
|
|
}
|
|
}
|
|
let mut wins = 0;
|
|
for decider in deciders {
|
|
if decider.await.unwrap() {
|
|
wins += 1;
|
|
}
|
|
}
|
|
assert_eq!(wins, RUNS, "each approval decided exactly once");
|
|
|
|
// Every run reaches Completed.
|
|
for run_id in &run_ids {
|
|
let mut done = false;
|
|
for _ in 0..200 {
|
|
let run = tc_db::repo::runs::get(&pool, *run_id).await.unwrap();
|
|
if run.state == RunState::Completed {
|
|
done = true;
|
|
break;
|
|
}
|
|
tokio::time::sleep(Duration::from_millis(25)).await;
|
|
}
|
|
assert!(done, "run {run_id} never completed");
|
|
}
|
|
|
|
// THE invariant: executions == approvals approved, exactly.
|
|
let approved = RUNS / 2;
|
|
let outbox = sqlx::query_scalar::<_, i64>("SELECT count(*) FROM outbox")
|
|
.fetch_one(&pool)
|
|
.await
|
|
.unwrap();
|
|
assert_eq!(
|
|
outbox as usize, approved,
|
|
"execute once per approval, never more"
|
|
);
|
|
|
|
// Every approved grant consumed exactly once; rejected ones have none.
|
|
let consumed =
|
|
sqlx::query_scalar::<_, i64>("SELECT count(*) FROM execution_grants WHERE consumed = true")
|
|
.fetch_one(&pool)
|
|
.await
|
|
.unwrap();
|
|
let total_grants = sqlx::query_scalar::<_, i64>("SELECT count(*) FROM execution_grants")
|
|
.fetch_one(&pool)
|
|
.await
|
|
.unwrap();
|
|
assert_eq!(consumed as usize, approved);
|
|
assert_eq!(total_grants as usize, approved);
|
|
|
|
// Every decision audited.
|
|
let audited = sqlx::query_scalar::<_, i64>(
|
|
"SELECT count(*) FROM audit_log WHERE event_type IN ('approval.approved', 'approval.rejected')",
|
|
)
|
|
.fetch_one(&pool)
|
|
.await
|
|
.unwrap();
|
|
assert_eq!(audited as usize, RUNS);
|
|
|
|
// No approval left pending.
|
|
assert!(approvals::list_pending(&pool, ws.id)
|
|
.await
|
|
.unwrap()
|
|
.is_empty());
|
|
}
|