The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill features a 'freeform' mode that generates Solidity smart contract code and behavioral tests based on user-provided briefs. This generated code is then compiled and executed on the host system using Foundry's 'forge' tool. This dynamic generation and compilation of code from untrusted user input represents a significant security surface area.
[COMMAND_EXECUTION]: In Step 11, the skill executes a bash command to call a Python script, passing arguments (such as hook name, category, and rules) that are derived directly from the user's free-text brief. If the agent does not properly sanitize or escape these values when constructing the command line, it could lead to arbitrary command execution on the host environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input (the 'brief') which directly influences the generation of executable smart contract logic.
Ingestion points: The user's brief variable '${var}' in SKILL.md.
Boundary markers: None are present to distinguish user input from system instructions or to warn the agent about malicious intent in the brief.
Capability inventory: Includes 'onchain_writes', 'writes_external_host', and 'sends_notifications', as well as shell execution via the 'hook-deploy.sh' script.
Sanitization: The skill relies on agent-led 'reasoning' and a basic regex-based audit script to identify dangerous code patterns.
[DYNAMIC_EXECUTION]: The 'hook-deploy.sh' script includes a 'scan_dangerous' function that uses 'grep' to detect risky Solidity keywords like 'selfdestruct' or 'delegatecall'. This static audit gate is fragile and can be bypassed through common techniques such as multi-line comments, string concatenation, or specific formatting that evades simple regular expressions, potentially providing a false sense of security.

deploy-uni-hook