Summary

In process_pending_secret_requests() (crates/zeph-core/src/agent/mod.rs), req.secret_key is embedded directly in the confirmation prompt with no length bound:

let prompt = format!(
    "Sub-agent requests secret '{}'. Allow?{}",
    req.secret_key,  // no length cap
    ...
);

secret_key is pre-validated to [a-zA-Z0-9_-] (ASCII-only, see manager.rs comment), so there is no panic risk on UTF-8 char boundaries. However, there is no upper length limit enforced, so a malicious or buggy sub-agent could supply an arbitrarily long key name and produce an oversized prompt string.

Severity

Low. The key is ASCII-validated so no panic. The oversized prompt is only shown to the operator for confirmation. No data loss or security bypass possible.

Steps to Reproduce

1. Spawn a sub-agent that sends a SecretRequest with a secret_key of e.g. 10,000 ASCII characters.
2. Observe that the confirmation prompt displayed to the operator is extremely long.

Expected Behavior

secret_key should be bounded (e.g., to 100 chars) before being embedded in the prompt, consistent with how reason is now bounded.

Related

• Fixed in sec(orchestration): truncate SecretRequest.reason before displaying to user #1456: reason field is now truncated to 200 chars via truncate_to_chars.