manage-feature-flags

Pass

Audited by Gen Agent Trust Hub on Mar 26, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill correctly implements tools for listing, getting, creating, toggling, and deleting Harness feature flags via the harness-mcp-v2 server. The tool parameters and instruction steps are consistent with typical feature flag management workflows.
  • [PROMPT_INJECTION]: The skill possesses a vulnerability to indirect prompt injection because it retrieves and processes metadata (such as flag descriptions and names) from the external Harness platform without sanitization or strict boundary markers.
  • Ingestion points: External data enters the agent context via the results of the harness_list and harness_get MCP tools (specifically in file SKILL.md).
  • Boundary markers: The skill does not define any delimiters or provide instructions to the agent to ignore or isolate instructions potentially embedded in the retrieved feature flag data.
  • Capability inventory: The skill provides the agent with the ability to perform impactful operations, including resource deletion (harness_delete) and state changes (harness_execute toggle), which could be misused if the agent obeys instructions found within the feature flag metadata.
  • Sanitization: There is no evidence of sanitization or validation of the data retrieved from the Harness API before it is processed by the agent.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 26, 2026, 10:23 PM
Security Audit — agent-trust-hub — manage-feature-flags