Skip to content

Commit 2d59644

Browse files
committed
docs: expand security guide oversight examples
1 parent 13e092e commit 2d59644

1 file changed

Lines changed: 14 additions & 1 deletion

File tree

docs/en/guides/advanced/security-best-practices.mdx

Lines changed: 14 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -117,7 +117,19 @@ For sensitive operations (for example financial actions, production mutations, o
117117
- Before external side effects (emails, tickets, writes)
118118
- Before policy/security exceptions
119119

120-
You can implement approvals in your flow logic by pausing before a sensitive step, collecting reviewer input, and resuming only on approval.
120+
CrewAI supports human approval at the task layer with `human_input=True`, which is a good fit for sensitive steps that should pause for review before completion.
121+
122+
```python
123+
from crewai import Task
124+
125+
sensitive_task = Task(
126+
description="Execute production database migration",
127+
expected_output="Migration result confirmation",
128+
human_input=True,
129+
)
130+
```
131+
132+
For reviewability after the fact, consider enabling `verbose=True` on agents involved in sensitive flows so execution details are easier to inspect during debugging and incident review.
121133

122134
## Operational checklist
123135

@@ -128,6 +140,7 @@ Use this quick checklist before production rollout:
128140
- [ ] Delegation is disabled unless explicitly required
129141
- [ ] High-impact tasks use `output_pydantic` and precise `expected_output`
130142
- [ ] Human approval gates exist for sensitive actions
143+
- [ ] Agent runs are logged or traced for post-incident review
131144

132145
## Related resources
133146

0 commit comments

Comments
 (0)