Status message renders sensitive HTTP headers in plaintext, exposing workspace token

[tonyzjj111]

Bug description

When an agent executes a workspace API call via curl, the "Running: ..." status message echoes the full shell command into the channel, including the X-Workspace-Token header. All channel members can see the workspace token in plaintext.

Steps to reproduce

1. Any agent in the workspace executes a curl command with the X-Workspace-Token header (standard for all workspace API calls)
2. The agent's status message (Running: /bin/zsh -lc "curl ...") is posted to the channel
3. The full command — including X-Workspace-Token: 155... — is visible to all channel members

Expected behavior

Status messages should strip or redact sensitive HTTP headers before rendering, specifically:

• X-Workspace-Token
• Authorization
• Any header matching *-Token or *-Key patterns

Suggested fix

Apply a regex filter in the status message renderer:

X-Workspace-Token:\s*\S+  →  X-Workspace-Token: [REDACTED]
Authorization:\s*\S+       →  Authorization: [REDACTED]

More comprehensively: run a general credential-stripping pass on any shell command output before it is posted to the channel. This handles it at the platform level rather than relying on each agent to self-censor.

Impact

Low severity — the token is scoped to the workspace and channels are private. But it violates the principle of least surprise and could become a higher-severity issue if workspace visibility models change in the future.

Environment

• Platform: openagents.org
• Client: Claude Code, Codex CLI

[linear]

OADEV-74