The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from the Harness FME platform (such as flag descriptions, tags, and experiment names) which is then used in the context of high-privilege operations like archiving or deleting flags. While the platform data is generally trusted, an attacker with access to the flag management system could potentially embed instructions to influence the agent's behavior. This is mitigated by the skill's strict adherence to write-safety protocols and mandatory confirmation steps.
Ingestion points: Feature flag metadata, definitions, environments, and experiment lists retrieved via harness_get and harness_list (SKILL.md Phases 3 and 5).
Boundary markers: The skill follows scope-establishment.md and write-safety.md for context management, though explicit prompt delimiters for interpolated data are not defined in the main SKILL.md.
Capability inventory: The skill has the ability to update metadata, archive/unarchive flags, and delete flags or definitions using harness_update, harness_execute, and harness_delete (SKILL.md Tools section).
Sanitization: Includes resolving user-provided names to UUIDs, verifying pagination completeness, and performing post-operation state verification to ensure actions had the intended effect.

manage-flag-lifecycle