CLI feedback thread upload: 019e419e-8f64-7bd0-a026-cad7c2438682

Summary

Codex CLI/agent repeatedly made outbound web tool calls (web.search_query) with an empty query (q: "") even though I did not request web browsing for the task.

What happened

During a normal coding session, tool logs showed many empty-query web searches in a row.
These calls were not needed for the solution, but they were still executed and then flagged by the API.

This is not isolated: I've seen the same behavior across many prior sessions as well.

Expected behavior

• No web calls unless:

1. the user explicitly asks for browsing, or
2. the assistant clearly needs web verification for time-sensitive facts.

• Never issue empty-query web searches (q: "").

Actual behavior

• web.search_query was invoked with q: "" repeatedly in sequence.
• Calls appeared in telemetry/logs and were flagged by the API.
• Pattern has recurred across multiple sessions over time.

Impact

• Creates false-positive “internet usage”/policy flags.
• Reduces trust in tool-call intent and auditability.
• Adds repeated reviewer/operator overhead.
• Causes avoidable noise in compliance and incident triage.

Repro (intermittent but recurring)

1. Start a long coding session with multiple tool calls.
2. Continue after context compaction/resume.
3. Do not ask for web search.
4. Observe occasional bursts of web.search_query with empty query in tool activity.

Thread reference

Uploaded thread: 019e419e-8f64-7bd0-a026-cad7c2438682