The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection by design as it reads and executes natural language instructions retrieved from external files.\n
Ingestion points: The agent is instructed to scan slides/*/index.tsx for @slide-comment markers.\n
Boundary markers: While instructions are stored in a base64url-encoded JSON payload, there are no directives to the agent to treat the decoded 'note' content as untrusted data or to ignore embedded commands.\n
Capability inventory: The agent has access to the Edit tool for file modifications and can execute shell commands like pnpm and node.\n
Sanitization: No validation or sanitization is performed on the decoded comment text before it is processed by the agent.\n- [COMMAND_EXECUTION]: The skill involves the execution of local shell commands for verification and utility.\n
The agent is directed to run pnpm tsc and pnpm lint to verify the integrity of the code after applying changes.\n
A JavaScript helper function is provided with the recommendation to execute it via node -e for decoding payloads.

apply-comments