The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to analyze and audit smart contract source code, which constitutes untrusted external data.
Ingestion points: The agent is instructed to read and analyze contract files, typically located in a contracts/ directory, using grep patterns and manual code review as seen in SKILL.md.
Boundary markers: The skill includes a 'Mandatory confirmation gate' that requires the agent to ask for target confirmation and explicit user authorization before performing active tasks.
Capability inventory: The agent uses shell commands like grep and cat in SKILL.md, and is instructed to generate and execute test suites using the forge framework as described in references/details.md.
Sanitization: There are no explicit mechanisms described for sanitizing or escaping the content of the contracts before the agent processes them, allowing potential for instructions embedded in code comments or strings to influence agent behavior.
[COMMAND_EXECUTION]: The skill provides numerous pre-defined grep commands and search patterns for the agent to execute in the terminal to identify security flaws in local source code.
[EXTERNAL_DOWNLOADS]: The documentation references a community GitHub repository (elementalsouls/Claude-BugHunter) as its source and suggests using external security tools such as Foundry, Slither, Echidna, and Mythril.
[DYNAMIC_EXECUTION]: The instructions in references/details.md include the generation of a Solidity Proof-of-Concept (PoC) based on a template and its execution using the forge test utility, which involves runtime code generation and execution.

web3-audit