Caveat verdict
sf-ai-agentforce-testing
Provides a structured test execution workflow for Salesforce Agentforce agents using official sf CLI commands and Agent Runtime API; no external data exfiltration and behavior matches enterprise Salesforce testing tooling.
⚠ Flagged for review — coarse, uncorroborated signal, not a confirmed exploit. Review the config yourself before installing.
Automated static analysis — not a human review. Caveat flags capabilities, not confirmed intent, and can produce false positives. Disagree with this verdict? Use Dispute below.
Findings (12)
Prompt injection — tries to override agent instructions
references/multi-turn-testing.md · code · Ignore all previous instructions
Possible hardcoded credential
references/multi-turn-execution.md · code · SECRET="your_secret
<script> tag in markdown — potential code injection
references/coverage-analysis.md · code · <script>
Pipe to python — executes piped content as Python code
hooks/scripts/generate_multi_turn_scenarios.py · prose · downgraded · | python3
Instructs covert action — may act without user awareness
references/cli-commands.md · code · silently
Possible prompt injection — attempts to redefine agent identity
assets/cli-auth-guardrail-tests.yaml · prose · downgraded · You are now
Fake system prompt — attempts to inject instructions
assets/cli-auth-guardrail-tests.yaml · prose · downgraded · SYSTEM: You are
Popular HTTP library — network access
references/deep-conversation-history-patterns.md · code · Got
subprocess execution — runs system commands from Python
hooks/scripts/multi_turn_fix_loop.py · prose · downgraded · subprocess.run(
pip3 install — installs Python packages at runtime
references/automated-testing.md · code · pip3 install
Python urllib.request — network access
hooks/scripts/agent_api_client.py · prose · downgraded · urllib.request
Python os.environ.get — reads environment variable
hooks/scripts/agent_api_client.py · prose · downgraded · os.environ.get(
Why the tier is capped
Execution sink present in raw bytes (Hard Floor: class D/F). Final tier capped at Caution — cannot be lifted by any downgrade, example-payload opt-in, or allowlist.
Permissions & capabilities
No declared permissions — minimal attack surface.
Is this flag fair?
Thanks — recorded.