The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to determine repository routing and branch naming based on external metadata found in Linear tickets (specifically the "Repo — agent routing" block). This creates a surface where untrusted data from a project management tool could influence the agent's file system operations and repository selection.
Ingestion points: Linear ticket descriptions and project metadata blocks (SKILL.md, Law 47).
Boundary markers: Absent; the instructions do not specify delimiters or warnings for the content of the ticket descriptions.
Capability inventory: Repository navigation, git branch creation, and file system modifications across different projects.
Sanitization: Absent; the agent is instructed to treat the ticket's routing values as "the only authority" for where work happens.
[SAFE]: The skill explicitly enforces industry-standard security best practices for secret management. It instructs the agent to never log, print, or commit secrets, environment files, or tokens (SKILL.md Law 15, references/data-and-trust.md).
[SAFE]: The skill incorporates a human-in-the-loop requirement for privilege escalation. The agent is instructed to confirm with the user before executing any commands with elevated privileges (sudo) when starting development services (references/data-and-trust.md).
[SAFE]: The skill outlines clear data privacy principles, such as ensuring telemetry is content-free, opt-out by default for production, and that personal data search remains local to the device (references/data-and-trust.md).

ponti-principles