The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides instructions for building scouts that ingest data from various external sources (Slack, GitHub, CRM, etc.). It includes a dedicated 'Safety' section in the documentation that explicitly warns against prompt injection, instructing agents to treat ingested content as data rather than instructions and to sanitize output.
Ingestion points: Scouts are designed to ingest data from PostHog events, data warehouse tables, and external APIs/repositories.
Boundary markers: The instructions recommend using scratchpad memory and explicit summaries rather than verbatim quotes to maintain boundaries.
Capability inventory: The skill describes the use of standard PostHog MCP tools (execute-sql, read-data-schema) and scout-specific reporting tools (emit_report, edit_report).
Sanitization: There are explicit instructions for PII sanitization and the use of paraphrasing instead of raw data quotation.
[PRIVILEGE_ESCALATION]: The skill documents the llm_skill:write scope, which allows scouts to modify their own bodies or those of sibling scouts. The documentation provides clear guidance on the risks of this permission and advises limiting its use to scenarios where 'tending skills' is the primary job.
[REMOTE_CODE_EXECUTION]: The documentation discusses 'External-tool' scouts that may run CLI tools (like git or gh) or execute static analysis within the sandbox environment. These capabilities are described as part of the intended platform functionality for specialized scouts, with associated warnings about environment verification and network access controls.

authoring-scouts