The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the local environment to drive its orchestration logic. * Ingestion points: In Phase 4, the skill performs a codebase detection pass to identify language, frameworks, and existing feature flag usage. In Phase 7 and 8, it reads application behavior and SDK evaluation details from the runtime environment to verify treatment outcomes. * Boundary markers: No specific delimiters or instructions to ignore embedded content are defined for these data ingestion steps, although the skill relies on manual user authorization for state changes. * Capability inventory: The skill uses read-only tools like harness_list and harness_get. All write operations are delegated to specialized sub-skills such as instrument-feature-flag and update-flag-targeting. * Sanitization: The instructions explicitly require using "non-sensitive subject references" and "non-secret progress summaries" to avoid handling PII or credentials within the agent's context.

feature-flag-onboarding