Tool rejection plan

Overview

A rejection plan lets you prevent a tool from executing when certain conditions are met. You attach it to any tool call and it evaluates the recent conversation state to decide whether to reject the call.

If all conditions match (AND logic), the tool call is rejected.
To express OR at the top level, use a single group condition with operator: "OR".
If conditions is empty or omitted, the tool always executes.

Use on any tool call, e.g., Assistant.hooks.do[type=tool].tool.rejectionPlan.

Schema

conditions: Array of condition objects. Defaults to [].
- Types:
  - RegexCondition: Match message content with a regex
    - type: “regex”
    - regex: String pattern. RegExp.test-style substring matching. Escape backslashes in JSON (e.g., "\\bhello\\b"). Supports inline flags like (?i) for case-insensitive.
    - target (optional): Which message to inspect
      - role: user | assistant
      - position: Integer index in history (default -1 for the most recent). Negative counts from the end; 0 is the first message
    - negate (optional): When true, the condition matches if the regex does NOT match (default false)
  - LiquidCondition: Evaluate a Liquid template that must output exactly "true" or "false"
    - type: “liquid”
    - liquid: The template. You can access messages (recent chat messages), now, and assistant variables. Useful filters include last, where, and reverse
  - GroupCondition: Combine multiple conditions
    - type: “group”
    - operator: AND | OR
    - conditions: Nested list of conditions (can recursively nest groups)

Examples

1) Reject endCall unless the user says goodbye

1 {
2   "conditions": [
3     {
4       "type": "regex",
5       "regex": "(?i)\\b(bye|goodbye|farewell|see you later|take care)\\b",
6       "target": { "position": -1, "role": "user" },
7       "negate": true
8     }
9   ]
10 }

2) Reject transfer if the user is actually asking a question

1 {
2   "conditions": [
3     {
4       "type": "regex",
5       "regex": "\\?",
6       "target": { "position": -1, "role": "user" }
7     }
8   ]
9 }

3) Reject transfer if the user hasn’t mentioned transfer recently (Liquid)

Liquid template for readability:

1 {% assign recentMessages = messages | last: 5 %}
2 {% assign userMessages = recentMessages | where: 'role', 'user' %}
3 {% assign mentioned = false %}
4 {% for msg in userMessages %}
5   {% if msg.content contains 'transfer' or msg.content contains 'connect' or msg.content contains 'representative' %}
6     {% assign mentioned = true %}
7   {% endif %}
8 {% endfor %}
9 {% if mentioned %}false{% else %}true{% endif %}

Wired into a rejection plan:

1 {
2   "conditions": [
3     {
4       "type": "liquid",
5       "liquid": "{% assign recentMessages = messages | last: 5 %}{% assign userMessages = recentMessages | where: 'role', 'user' %}{% assign mentioned = false %}{% for msg in userMessages %}{% if msg.content contains 'transfer' or msg.content contains 'connect' or msg.content contains 'representative' %}{% assign mentioned = true %}{% endif %}{% endfor %}{% if mentioned %}false{% else %}true{% endif %}"
6     }
7   ]
8 }

4) Top-level OR using a group

1 {
2   "conditions": [
3     {
4       "type": "group",
5       "operator": "OR",
6       "conditions": [
7         { "type": "regex", "regex": "(?i)\\bcancel\\b", "target": { "role": "user" } },
8         { "type": "regex", "regex": "(?i)\\bstop\\b",   "target": { "role": "user" } }
9       ]
10     }
11   ]
12 }

Normal tool call example

Attach rejectionPlan directly on a tool in your assistant configuration (model.tools):

1 {
2   "model": {
3     "provider": "openai",
4     "model": "gpt-4o",
5     "messages": [
6       { "role": "system", "content": "Only end the call after the user says goodbye." }
7     ],
8     "tools": [
9       {
10         "type": "endCall",
11         "rejectionPlan": {
12           "conditions": [
13             {
14               "type": "regex",
15               "regex": "(?i)\\b(bye|goodbye|farewell|see you later|take care)\\b",
16               "target": { "position": -1, "role": "user" },
17               "negate": true
18             }
19           ]
20         }
21       }
22     ]
23   }
24 }

Another example: transferCall with rejection

1 {
2   "model": {
3     "provider": "openai",
4     "model": "gpt-4o",
5     "messages": [
6       { "role": "system", "content": "Transfer only if the user clearly asks to be connected." }
7     ],
8     "tools": [
9       {
10         "type": "transferCall",
11         "destinations": [
12           { "type": "number", "number": "+1234567890" }
13         ],
14         "rejectionPlan": {
15           "conditions": [
16             {
17               "type": "group",
18               "operator": "OR",
19               "conditions": [
20                 { "type": "regex", "regex": "(?i)\\bconnect\\b", "target": { "role": "user" } },
21                 { "type": "regex", "regex": "(?i)\\btransfer\\b", "target": { "role": "user" } }
22               ]
23             },
24             {
25               "type": "regex",
26               "regex": "\\?",
27               "target": { "position": -1, "role": "user" },
28               "negate": true
29             }
30           ]
31         }
32       }
33     ]
34   }
35 }

Tips

Escape backslashes in regex patterns: write \\b in JSON to mean \b in the regex engine.
position: -1 targets the most recent message. Omit role to target regardless of role.
Prefer a group with operator: "OR" for disjunctive logic at the top level.