[bug] Out-of-bounds IndexError in extract_description when parsing crafted component YAML

[karnakarreddi]

Environment

How do you deploy Kubeflow Pipelines (KFP)?
Not specified (affects all environments where KFP SDK loads component YAML from untrusted sources)

KFP version:
Not specified (vulnerable in versions using ComponentSpec.from_yaml_documents())

KFP SDK version:
Not specified (any version containing structures.py implementation of extract_description)

---

Steps to reproduce

1. Install KFP SDK:
   pip install kfp

2. Run the script:

from kfp.components import structures

malicious_yaml = """line0
line1
Description: some text
continued text"""

try:
spec = structures.ComponentSpec.from_yaml_documents(malicious_yaml)
except IndexError as e:
print("Crash triggered:", e)

3. Or load via:
   load_component_from_text()

4. Observe:
   IndexError: list index out of range

---

Expected result

• No crash or unhandled exception
• Invalid YAML should fail gracefully with a validation error
• Parser should not crash on malformed input

---

Materials and reference

Issue in extract_description() inside ComponentSpec.from_yaml_documents()

Problems:

• No bounds check for comments[index]
• Loop can increment index beyond list length
• Assumes fixed line position (index_of_heading = 2)

Vulnerable code:

while comments[index][:len(multi_line_description_prefix)] == multi_line_description_prefix:

Fix:

while index < len(comments) and comments[index][:len(multi_line_description_prefix)] == multi_line_description_prefix:

Also recommended:

• Validate index_of_heading before access
• Avoid hardcoded line assumptions in YAML parsing logic

---

Impact

Type: Denial of Service (DoS)
CWE: CWE-125 (Out-of-bounds access / unvalidated index)
Severity: Medium

Impact:
Any system using KFP SDK to parse untrusted YAML may crash, including:

• CI/CD pipelines
• Shared component repositories
• Multi-tenant pipeline services

---

Labels
/area sdk
/area components
/area backend
/area security