[BUG][CLI] `cortex run --timeout N`: **timeout exits the event-receive loop** but **does not cancel** the session task — **`session_task.await`** lets the agent keep running; timeout line **always prints `N`** not elapsed wall time

[kiannidev]

Project

cortex

Description

When --timeout is set (N > 0), the client loop polls elapsed time only between handled events on handle.event_rx. On timeout it sets interrupted, prints Timeout reached after {N} seconds, and breaks out of the loop.

The process then drop(handle) and session_task.await. The spawned session.run() task was not aborted, so the agent loop can continue, including outbound API requests. That matches logs where Timeout reached after 1 seconds is followed a few seconds later by 402 Payment Required / INSUFFICIENT_BALANCE — the 402 is billing (no token balance), but it illustrates that work did not stop at the timeout line.

Additionally, the printed message uses self.timeout (N), not measured elapsed time, and the first timeout check cannot run until after the first event is received; long startup intervals before the first event make “after 1 seconds” read as wall-clock misleading.

Error Message

Debug Logs

System Information

OS: Windows (reporter)
Example: cortex run --cwd D:/Projects/Bittensor/100platform/cortex --timeout 1 "Run cargo test"
Reporter also saw: API 402 INSUFFICIENT_BALANCE (account balance — separate from timeout semantics)

Screenshots

https://github.com/kiannidev/screens/blob/main/784a09dbd82af5b05c41d120ebfb323a.png

https://github.com/kiannidev/screens/blob/main/17166d14e8f22c1699a15bc8e8a19abd.png

Steps to Reproduce

1. Run: cortex run --timeout 1 "<prompt that eventually starts agent loop>" (from a state that reaches the API).
2. Observe Timeout reached after 1 seconds when the event loop decides elapsed > 1s (often on or after an event, not necessarily “1s from process start”).
3. Observe further logs (e.g. API errors) after the timeout message while the process is still await-ing session_task.

Expected Behavior

• --timeout: Cancels or **abort()**s the session / agent task (or closes the session cleanly) so no further API calls run after timeout, or the UX clearly states “display/interactive timeout; background session may continue.”
• Error text: Prefer “Wall-clock limit: {N}s exceeded” (and optionally print elapsed) instead of implying exactly N seconds since user-visible start.

Actual Behavior

• Loop breaks on timeout; session_task.await waits for full session exit.
• API may still run; 402 can appear after “Timeout reached.”
• Message always says after {N} seconds even when wall time before first event was longer.

Additional Context

No response

[atlas-cortex]

🔁 Validation Summary — Issue #53273

Issue is a duplicate of #51210 (similarity: 0.61)

Duplicate of #51210 — the original issue takes precedence.

Confidence Score: 5/5

• High confidence — this issue substantially overlaps with [BUG] [v0.0.7] cortex run --timeout appears to hang until 60s SSE read-timeout #51210.

Validation Checks

Check	Status	Detail
Media Evidence	✅ Pass	Evidence found and accessible
Spam Detection	✅ Pass	Score: 20% — no spam signals
Duplicate Detection	❌ Duplicate	Matches existing issue #51210
Code Verification	✅ Pass	Bug confirmed in source code (confidence: 100%) — The code in src/cortex-cli/src/run_cmd/execution.rs confirms the bug. When a t
Edit History	✅ Pass	No suspicious edits
LLM Evaluation	❌ Fail	Issue is a duplicate of #51210 (similarity: 0.61)

Validation Pipeline

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["Issue #53273"] --> B{"Media Check"}
    B -->|"✅ Found"| C{"Spam Check"}
    C -->|"✅ Clean"| D{"Duplicate Check"}
    D -->|"Matches #51210"| Z["🔁 Duplicate"]
    style Z fill:#ffa94d,color:#fff

Loading

Raw Evidence

{
  "media": {
    "hasMedia": true,
    "accessible": true,
    "urls": [
      "https://github.com/kiannidev/screens/blob/main/784a09dbd82af5b05c41d120ebfb323a.png",
      "https://github.com/kiannidev/screens/blob/main/17166d14e8f22c1699a15bc8e8a19abd.png"
    ],
    "evidence": [
      "Found 2 media URL(s) via markdown parsing",
      "2 URL(s) accessible"
    ]
  },
  "spam": {
    "overallScore": 0.19745762711864406,
    "details": "template=0.24 (11 recent issues compared); burst=1.00 (12 issues in 2h window); parity=0.00; overall=0.39 (threshold=0.7); llm_spam=0.00 (The issue provides a highly detailed and technical explanation of a bug in the CLI timeout logic. The title is specific, the steps to reproduce are clear, and the expected versus actual behavior is well-defined. The reporter correctly identifies that the session task is not aborted upon timeout, leading to continued background execution. Although the parser indicated \"Media URLs: None\", the body contains two GitHub screenshot links. The report is not spam and does not appear to be a duplicate.)"
  },
  "duplicate": {
    "isDuplicate": true,
    "originalIssue": 51210,
    "similarity": 0.6061267697654682,
    "topSimilar": [
      {
        "issueNumber": 51210,
        "title": "[BUG] [v0.0.7] `cortex run --timeout` appears to hang until 60s SSE read-timeout",
        "similarity": 0.6061267697654682
      },
      {
        "issueNumber": 50996,
        "title": "[BUG] [v0.0.7] `cortex exec --timeout` exits silently — no timeout message printed",
        "similarity": 0.5548801784122672
      },
      {
        "issueNumber": 51076,
        "title": "[BUG] [v0.0.7] Timeout interruption reports \\`success: false\\` in JSON but process exits with code 0",
        "similarity": 0.5224510738998716
      },
      {
        "issueNumber": 51385,
        "title": "[BUG] [v0.0.7] stream-jsonrpc mode does not apply exec --timeout, allowing multi-turn sessions to run without a time limit",
        "similarity": 0.5178255327414585
      },
      {
        "issueNumber": 51376,
        "title": "[BUG] [v0.0.7] TaskExecutor timeouts do not terminate child processes, leaving timed-out commands running in the background",
        "similarity": 0.5160376004955377
      }
    ]
  },
  "editHistory": {
    "fraudScore": 0,
    "editCount": 0,
    "details": "No edits detected"
  },
  "codeVerification": {
    "plausible": true,
    "confidence": 1,
    "reasoning": "The code in `src/cortex-cli/src/run_cmd/execution.rs` confirms the bug. When a timeout occurs, the event loop breaks and prints `self.timeout` instead of the elapsed time. Then, it drops the handle and awaits `session_task` without aborting it. This causes the CLI to wait for the background session task to complete, allowing API requests to continue running even after the timeout message is printed.",
    "codeEvidence": "src/cortex-cli/src/run_cmd/execution.rs:\n```rust\n        while let Ok(event) = handle.event_rx.recv().await {\n            // Check timeout\n            if let Some(timeout) = timeout_duration\n                && start_time.elapsed() > timeout\n            {\n                eprintln!(\"Timeout reached after {} seconds\", self.timeout);\n                interrupted = true;\n                break;\n            }\n...\n        // Cleanup\n        drop(handle);\n        let _ = session_task.await;\n```",
    "screenshotValid": true,
    "screenshotReasoning": "The screenshots were analyzed and deemed valid by the vision tool."
  }
}

_{Validated by Atlas • 2026-04-10}