Agent Rules and Memories
Two parallel note systems in Settings → Agent Rules and Settings → Memories. Humans edit them in the UI; Cursor Agent can create, read, list, and delete them with built-in fences.
What they’re for
| Kind | Intent |
|---|---|
| Agent Rules | Durable instructions for how the Agent should work in this extension (style, constraints, preferred workflows). |
| Memories | Short factual notes (targets, creds references, hunt context) the Agent should be able to recall when relevant. |
They share the same UX (create / edit / delete / search / sort / pagination). Conceptually: rules steer behavior; memories store facts.
Persistence
Both lists live in host user preferences, not in the Burp project’s chat store.
That means:
- Rules and Memories carry across Burp projects on the same host/user prefs.
- Chat transcripts stay per project (sessions.md).
- Clearing or deleting a chat does not remove Rules or Memories.
This is not encrypted-at-rest isolation — treat secrets like any other Burp preference data.
Human UI
Under Cursor Agent → Settings:
- Open Agent Rules or Memories.
- + Create rule / + Create memory.
- Edit subject + body → Save.
- Edit / delete (✕) on existing rows; delete asks for confirmation.
- Search titles and bodies; sort by Updated / Created / Title A–Z.
- List length follows Settings → General → List page size.
Changes apply when you Save Settings (same as other settings).
How the Agent sees them
Each prompt includes a title index only (Extension rules (titles only) / Extension memories (titles only)), plus short fence usage hints.
Full bodies are not injected every turn. When the Agent needs content, it emits a read or list fence; the extension resolves it, shows a sticky card in the chat, and can follow up so the Agent gets the text.
That keeps prompts lean as your note lists grow.
Built-in fences
Documented in Settings → Tools. Canonical names:
Rules
| Fence | Role |
|---|---|
burpai-rule/create | Create or update (upsert) a rule |
burpai-rule/read | Fetch one rule (title: or query:) |
burpai-rule/list | List matching titles (optional query:) |
burpai-rule/delete | Remove by exact title: or id: |
Memories
| Fence | Role |
|---|---|
burpai-memory/create | Create or update (upsert) a memory |
burpai-memory/read | Fetch one memory (title: or query:) |
burpai-memory/list | List matching titles (optional query:) |
burpai-memory/delete | Remove by exact title: or id: |
There is no separate /update fence — create (and accepted edit / upsert aliases) writes or replaces.
Chat rename (related)
| Fence | Role |
|---|---|
burpai-rename | Rename the current chat — always requires Allow, even under Allow all |
Permissions
| Action | Gate |
|---|---|
| create / upsert / delete Rules or Memories | Follows the chat permission policy: Ask → confirm (Change rules / memories?); Allow all → allowed; Deny and Allow safe (read-only) → denied |
| read / list | No write gate |
burpai-rename | Always Ask |
Manual edits in Settings do not go through the Agent permission banner.
See permissions.md.
Tips
- Prefer fetch (
/read,/list) over asking the Agent to paste every note into the chat. - Keep Memories short and factual; put standing instructions in Rules.
- For Burp proxy/Repeater work, still prefer Cursor’s Burp MCP over inventing parallel tools (mcp.md).
Related
- Sessions · Permissions · MCP