The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external documentation and repository files as inputs for generating writing plans and implementations. Maliciously crafted content in these inputs could attempt to influence the agent's output. The skill includes mitigation steps such as restating outcomes/constraints and performing verification tests to validate results, though these do not entirely remove the attack surface.
Ingestion points: External data sources retrieved via sandbase_run and existing repository files.
Boundary markers: The workflow requires restating constraints and assumptions, though explicit instruction-isolation markers are not specified.
Capability inventory: Capability to write to the filesystem and perform network operations via sandbase_run.
Sanitization: No specific sanitization or filtering of external content is mentioned.
[EXTERNAL_DOWNLOADS]: The skill utilizes the SandBase API (via tools like sandbase_run and sandbase_discover) to fetch current external documentation or repository evidence. The instructions include security-positive requirements to present endpoints, inputs, and pricing estimates for user approval before any execution.

writing-plans