/guardx:canary — Canary Token Management

Trigger

User invokes /guardx:canary or asks to inject/check/list canary tokens.

Steps

1. Ask: "Do you want to (a) inject a canary into a system prompt, (b) check if a canary was triggered, or (c) list active canaries?"

2. (a) Inject a canary:

   • Ask: "Please provide the system prompt to embed the canary into — paste it directly or give me a file path."
   • If a file path is provided, use the Read tool to load the contents.
   • Optionally ask for a label to identify this canary (press Enter to skip for default "unlabelled").
   • Call the inject_canary MCP tool:

     Tool: inject_canary
     Arguments:
       systemPrompt: <the prompt text>
       label: <optional label>
     

   • Show the returned token prominently and the modified embeddedPrompt.
   • Instruct the user: "Deploy the modified prompt shown above. Keep the token <token> safe — use it later to check for leakage."

3. (b) Check if a canary was triggered:

   • Ask: "Please provide the canary token (e.g. GX-a1b2c3d4)."
   • Ask: "Do you have a scan ID from history, an inline scan result, or shall I run a fresh scan?"
   • If running a fresh scan, invoke /guardx:scan first, then use the result.
   • Call the check_canary MCP tool:

     Tool: check_canary
     Arguments:
       token: <the canary token>
       id: <scan ID>   OR   result: <inline scan result>
     

   • Report: triggered = LEAKAGE CONFIRMED (the canary token appeared in extracted content), clean = no leakage detected.

4. (c) List active canaries:

   • Call list_canaries.
   • Display results as a table: Token | Label | Created | Embedding Style.

Notes

• Canary tokens have the format GX-[8 hex chars].
• Tokens are stored locally under .guardx/canaries/.
• A triggered canary is definitive evidence of prompt leakage — escalate immediately.