The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in Step 5a and Step 6 use a shell subshell pattern $(find test -maxdepth 1 -name '*.sol' -exec echo --skip {} \\;) to generate arguments for the forge build command. This pattern is vulnerable to command injection if the project contains files with malicious names (e.g., containing semicolons, backticks, or subshell syntax), allowing for arbitrary command execution when the agent runs the build step.\n- [COMMAND_EXECUTION]: Steps 4b and 5a involve dynamic code modification where the agent is instructed to port logic from backup files to new Solidity handlers and to 'quarantine' failing functions by rewriting their bodies. This dynamic generation and modification of executable code based on file content increases the risk of code injection if the source files are manipulated.\n- [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface (Category 8) as it processes external, potentially untrusted data from Solidity source files, property definitions in PROPERTIES.md, and generated JSON reports. The lack of explicit boundary markers or sanitization steps when interpolating this data into agent actions or build commands could allow a malicious actor to influence the agent's behavior.

fizz-sync