update-flag-targeting
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes feature flag definitions and segment metadata retrieved from external sources using
harness_getandharness_listoperations. - Ingestion points: Feature flag definitions, segment data, and experiment lists are ingested from the Harness environment at runtime (specified in
SKILL.mdtools and instructions). - Boundary markers: The instructions mandate generating a diff and a plain-language explanation of proposed changes, and require the agent to stop for explicit human confirmation before any write operation (referencing
write-safety.md). - Capability inventory: The skill possesses capabilities to modify environment state via
harness_update,harness_create, andharness_execute(kill/restore) tools (specified inSKILL.md). - Sanitization: The skill relies on user review of the generated diffs and plan descriptions rather than automated input sanitization of the retrieved flag definitions.
Audit Metadata