agent-governance

Name: agent-governance
Author: github

9.4kinstalls4.6kforks

How to install agent-governance?

github/awesome-copilot/agent-governance

$npx -y skills add github/awesome-copilot --skill agent-governance

Installs into the current project.

Summary

Patterns and techniques for adding governance, safety, and trust controls to AI agent systems. Use this skill when: - Building AI agents that call external tools (APIs, databases, file systems) - Implementing policy-based access controls for agent tool usage - Adding semantic int

Files · 1

View on GitHub

SKILL.md

1	# Agent Governance Patterns
2
3	Patterns for adding safety, trust, and policy enforcement to AI agent systems.
4
5	## Overview
6
7	Governance patterns ensure AI agents operate within defined boundaries — controlling which tools they can call, what content they can process, how much they can do, and maintaining accountability through audit trails.
8
9	```
10	User Request → Intent Classification → Policy Check → Tool Execution → Audit Log
11	↓ ↓ ↓
12	Threat Detection Allow/Deny Trust Update
13	```
14
15	## When to Use
16
17	- Agents with tool access: Any agent that calls external tools (APIs, databases, shell commands)
18	- Multi-agent systems: Agents delegating to other agents need trust boundaries
19	- Production deployments: Compliance, audit, and safety requirements
20	- Sensitive operations: Financial transactions, data access, infrastructure management
21
22	---
23
24	## Pattern 1: Governance Policy
25
26	Define what an agent is allowed to do as a composable, serializable policy object.
27
28	```python
29	from dataclasses import dataclass, field
30	from enum import Enum
31	from typing import Optional
32	import re
33
34	class PolicyAction(Enum):
35	ALLOW = "allow"
36	DENY = "deny"
37	REVIEW = "review" # flag for human review
38
39	@dataclass
40	class GovernancePolicy:
41	"""Declarative policy controlling agent behavior."""
42	name: str
43	allowed_tools: list[str] = field(default_factory=list) # allowlist
44	blocked_tools: list[str] = field(default_factory=list) # blocklist
45	blocked_patterns: list[str] = field(default_factory=list) # content filters
46	max_calls_per_request: int = 100 # rate limit
47	require_human_approval: list[str] = field(default_factory=list) # tools needing approval
48
49	def check_tool(self, tool_name: str) -> PolicyAction:
50	"""Check if a tool is allowed by this policy."""
51	if tool_name in self.blocked_tools:
52	return PolicyAction.DENY
53	if tool_name in self.require_human_approval:
54	return PolicyAction.REVIEW
55	if self.allowed_tools and tool_name not in self.allowed_tools:
56	return PolicyAction.DENY
57	return PolicyAction.ALLOW
58
59	def check_content(self, content: str) -> Optional[str]:
60	"""Check content against blocked patterns. Returns matched pattern or None."""
61	for pattern in self.blocked_patterns:
62	if re.search(pattern, content, re.IGNORECASE):
63	return pattern
64	return None
65	```
66
67	### Policy Composition
68
69	Combine multiple policies (e.g., org-wide + team + agent-specific):
70
71	```python
72	def compose_policies(*policies: GovernancePolicy) -> GovernancePolicy:
73	"""Merge policies with most-restrictive-wins semantics."""
74	combined = GovernancePolicy(name="composed")
75
76	for policy in policies:
77	combined.blocked_tools.extend(policy.blocked_tools)
78	combined.blocked_patterns.extend(policy.blocked_patterns)
79	combined.require_human_approval.extend(policy.require_human_approval)
80	combined.max_calls_per_request = min(
81	combined.max_calls_per_request,
82	policy.max_calls_per_request
83	)
84	if policy.allowed_tools:
85	if combined.allowed_tools:
86	combined.allowed_tools = [
87	t for t in combined.allowed_tools if t in policy.allowed_tools
88	]
89	else:
90	combined.allowed_tools = list(policy.allowed_tools)
91
92	return combined
93
94
95	# Usage: layer policies from broad to specific
96	org_policy = GovernancePolicy(
97	name="org-wide",
98	blocked_tools=["shell_exec", "delete_database"],
99	blocked_patterns=[r"(?i)(api[_-]?key\|secret\|password)\s*[:=]"],
100	max_calls_per_request=50
101	)
102	team_policy = GovernancePolicy(
103	name="data-team",
104	allowed_tools=["query_db", "read_file", "write_report"],
105	require_human_approval=["write_report"]
106	)
107	agent_policy = compose_policies(org_policy, team_policy)
108	```
109
110	### Policy as YAML
111
112	Store policies as configuration, not code:
113
114	```yaml
115	# governance-policy.yaml
116	name: production-agent
117	allowed_tools:
118	- search_documents
119	- query_database
120	- send_email
121	blocked_tools:
122	- shell_exec
123	- delete_record
124	blocked_patterns:
125	- "(?i)(api[_-]?key\|secret\|password)\\s*[:=]"
126	- "(?i)(drop\|truncate\|delete from)\\s+\

People who liked this also viewed

6 picks

Type

agent-governance

9.4kinstalls4.6kforks

View on GitHub

How to install agent-governance?

github/awesome-copilot/agent-governance

$npx -y skills add github/awesome-copilot --skill agent-governance

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Agent Governance Patterns
2
3	Patterns for adding safety, trust, and policy enforcement to AI agent systems.
4
5	## Overview
6
7	Governance patterns ensure AI agents operate within defined boundaries — controlling which tools they can call, what content they can process, how much they can do, and maintaining accountability through audit trails.
8
9	```
10	User Request → Intent Classification → Policy Check → Tool Execution → Audit Log
11	↓ ↓ ↓
12	Threat Detection Allow/Deny Trust Update
13	```
14
15	## When to Use
16
17	- Agents with tool access: Any agent that calls external tools (APIs, databases, shell commands)
18	- Multi-agent systems: Agents delegating to other agents need trust boundaries
19	- Production deployments: Compliance, audit, and safety requirements
20	- Sensitive operations: Financial transactions, data access, infrastructure management
21
22	---
23
24	## Pattern 1: Governance Policy
25
26	Define what an agent is allowed to do as a composable, serializable policy object.
27
28	```python
29	from dataclasses import dataclass, field
30	from enum import Enum
31	from typing import Optional
32	import re
33
34	class PolicyAction(Enum):
35	ALLOW = "allow"
36	DENY = "deny"
37	REVIEW = "review" # flag for human review
38
39	@dataclass
40	class GovernancePolicy:
41	"""Declarative policy controlling agent behavior."""
42	name: str
43	allowed_tools: list[str] = field(default_factory=list) # allowlist
44	blocked_tools: list[str] = field(default_factory=list) # blocklist
45	blocked_patterns: list[str] = field(default_factory=list) # content filters
46	max_calls_per_request: int = 100 # rate limit
47	require_human_approval: list[str] = field(default_factory=list) # tools needing approval
48
49	def check_tool(self, tool_name: str) -> PolicyAction:
50	"""Check if a tool is allowed by this policy."""
51	if tool_name in self.blocked_tools:
52	return PolicyAction.DENY
53	if tool_name in self.require_human_approval:
54	return PolicyAction.REVIEW
55	if self.allowed_tools and tool_name not in self.allowed_tools:
56	return PolicyAction.DENY
57	return PolicyAction.ALLOW
58
59	def check_content(self, content: str) -> Optional[str]:
60	"""Check content against blocked patterns. Returns matched pattern or None."""
61	for pattern in self.blocked_patterns:
62	if re.search(pattern, content, re.IGNORECASE):
63	return pattern
64	return None
65	```
66
67	### Policy Composition
68
69	Combine multiple policies (e.g., org-wide + team + agent-specific):
70
71	```python
72	def compose_policies(*policies: GovernancePolicy) -> GovernancePolicy:
73	"""Merge policies with most-restrictive-wins semantics."""
74	combined = GovernancePolicy(name="composed")
75
76	for policy in policies:
77	combined.blocked_tools.extend(policy.blocked_tools)
78	combined.blocked_patterns.extend(policy.blocked_patterns)
79	combined.require_human_approval.extend(policy.require_human_approval)
80	combined.max_calls_per_request = min(
81	combined.max_calls_per_request,
82	policy.max_calls_per_request
83	)
84	if policy.allowed_tools:
85	if combined.allowed_tools:
86	combined.allowed_tools = [
87	t for t in combined.allowed_tools if t in policy.allowed_tools
88	]
89	else:
90	combined.allowed_tools = list(policy.allowed_tools)
91
92	return combined
93
94
95	# Usage: layer policies from broad to specific
96	org_policy = GovernancePolicy(
97	name="org-wide",
98	blocked_tools=["shell_exec", "delete_database"],
99	blocked_patterns=[r"(?i)(api[_-]?key\|secret\|password)\s*[:=]"],
100	max_calls_per_request=50
101	)
102	team_policy = GovernancePolicy(
103	name="data-team",
104	allowed_tools=["query_db", "read_file", "write_report"],
105	require_human_approval=["write_report"]
106	)
107	agent_policy = compose_policies(org_policy, team_policy)
108	```
109
110	### Policy as YAML
111
112	Store policies as configuration, not code:
113
114	```yaml
115	# governance-policy.yaml
116	name: production-agent
117	allowed_tools:
118	- search_documents
119	- query_database
120	- send_email
121	blocked_tools:
122	- shell_exec
123	- delete_record
124	blocked_patterns:
125	- "(?i)(api[_-]?key\|secret\|password)\\s*[:=]"
126	- "(?i)(drop\|truncate\|delete from)\\s+\

People who liked this also viewed

6 picks

Type