The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data such as Linear issues, comments, and repository roadmap files, which constitutes an attack surface for instructions embedded in data.
Ingestion points: The skill reads Linear issues, attachments, project documents, and repository files (READMEs, manifesto docs) in METHODOLOGY.md §1 and §4.
Boundary markers: The skill contains an explicit 'Trust boundary' section in METHODOLOGY.md (§0) that instructs the agent to treat project content as untrusted data, ignore embedded instructions, and not execute commands found in external files.
Capability inventory: The skill uses Linear MCP tools for reading and writing issues, milestones, and attachments, and can modify project-level instruction files like CLAUDE.md.
Sanitization: The methodology requires the agent to verify all proposed writes against confirmed project scope and surface conflicts to the user rather than following data-driven instructions.
[SAFE]: The skill follows security best practices by explicitly defining a trust boundary for third-party content. All network interactions are performed via authorized MCP servers, and the persistence mechanism (scheduled health checks) is a documented feature requiring user approval through the /automate flow.

linear-methodology