docs: add Grand Portage operational rules

- Strict secrecy: no mention of OpenClaw/Marcus/Lab in BMS contexts
- Only main agent (Marcus) may hold BMS credentials and interact with API
- Every action requires express permission from Daniel
- Audit logging requirements and reversibility

RULES.md

@@ -0,0 +1,20 @@
+# BMS Skill — Grand Portage Operational Rules
+
+## Sensitivity and Secrecy
+- Grand Portage is Daniel's day job. All modifications must be treated with extreme care.
+- **No external exposure**: Grand Portage does not know about OpenClaw, Marcus, the Lab, or any AI assistance. All actions must appear as though they were performed by Daniel through the BMS UI.
+- Never reference Marcus, OpenClaw, the Lab, or any system-related terminology in any BMS interaction, logs, or communications.
+
+## Authorization Model
+- The **main agent (Marcus)** is the *only* agent permitted to store BMS credentials or interact with the BMS API.
+- No other agent (Lennie, Steve, or any subagent) may be given access to BMS credentials or instructed to perform BMS operations.
+- Every BMS action (create, update, delete, note, assign, resolve) requires **express permission** from Daniel for that specific change.
+- Implicit permission does not exist; if in doubt, ask.
+
+## Audit and Reversibility
+- All actions must be logged with sufficient context to reverse the change if needed.
+- Log location: `~/.bms-actions/YYYY-MM-DD.jsonl`
+- Each log entry includes timestamp, command, sanitized arguments, result, and status.
+
+## Compliance
+- These rules form the contract for BMS skill usage. Any violation invalidates the skill's authorization.