The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill demonstrates functions that ingest untrusted data from external callers, such as usernames and post content.
Ingestion points: Functions like register in src/user_service/main.mo and createPost in src/content_service/main.mo.
Boundary markers: The code examples do not use explicit delimiters, but the documentation provides clear instructions on implementing validation.
Capability inventory: The skill is capable of state modification and inter-canister calls.
Sanitization: The skill's 'Production Readiness' sections explicitly warn about the lack of input validation in the examples and provide detailed remediation guidance.
[COMMAND_EXECUTION]: The 'Canister Factory' examples demonstrate the dynamic installation of WASM code using the management canister's install_code method. The documentation identifies this as a high-risk capability and provides security best practices, such as verifying WASM hashes and restricting access to authorized principals.

multi-canister