flows
Pass
Audited by Gen Agent Trust Hub on Nov 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONDATA_EXFILTRATIONPRIVILEGE_ESCALATIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from local JSON flow definitions and payload files via the
-fflag (documented in reference/flow_cli_flags.md). There are no boundary markers or sanitization logic specified for the content of these files, while the skill possesses capabilities to perform network operations and execute commands based on these inputs.\n- [DYNAMIC_EXECUTION]: Theflows run <path>command (documented in SKILL.md) enables the execution of local workflow graphs. These graphs represent dynamically executed logic that is defined outside of the skill's static instructions, potentially allowing for complex and unpredictable API interactions at runtime.\n- [DATA_EXFILTRATION]: The skill documents the--show-secretsflag for thetriggersubcommand, which reveals authentication tokens in the CLI output. This capability exposes sensitive credentials to the agent's context, which could lead to unauthorized data exposure if the output is not strictly controlled.\n- [PRIVILEGE_ESCALATION]: The skill provides instructions for theflows updatecommand, which includes an--auth offoption to remove authentication from deployed flow triggers. This grants the agent the ability to downgrade security configurations for remote resources.\n- [COMMAND_EXECUTION]: The skill facilitates the execution of various Postman CLI subcommands. While these are part of a legitimate developer toolset, the agent gains a broad set of functional permissions to interact with both local and remote environments via shell commands.
Audit Metadata