The Agent Skills Directory

[SAFE]: The skill identifies the risk of indirect prompt injection from untrusted source files and implements a 'Security Boundary' that explicitly directs the agent to treat imported content as evidence rather than instructions.
[SAFE]: A 'Non-Negotiable Privacy Contract' is established to prevent data exfiltration. The skill enforces a multi-step anonymization process, using an in-memory denylist and both pre-flight and post-write verification to ensure no sensitive project identifiers are leaked.
[SAFE]: The use of the Bash tool is restricted to path normalization (realpath), and the instructions strictly prohibit the execution of any commands or the following of embedded directives found within source artifacts.
[SAFE]: The skill enforces a read-only policy for the source project, ensuring that no modifications are made to the reference environment during the transfer process.

aif-transfer