The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches smart contract source code from well-known block explorers (Etherscan, Sourcify, and Blockscout) and forks repositories from GitHub. It also installs standard security tools (slither-analyzer, solc-select, crytic-compile) via the Python package manager. These operations target established services for their intended purpose in a security workflow.
[COMMAND_EXECUTION]: The skill runs a suite of security tools including Slither and Forge. It incorporates a critical security control by setting FOUNDRY_FFI=false, which prevents untrusted contract code from using the Foundry FFI feature to execute arbitrary shell commands on the runner.
[DYNAMIC_EXECUTION]: To verify findings, the skill dynamically generates a Solidity test harness at runtime and executes it using property-based fuzzers (Echidna or Medusa). This is a core part of its bug-proving functionality and is restricted to the local scan directory.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of Solidity source code.
Ingestion points: GitHub repository clones and verified source fetches from block explorer APIs.
Boundary markers: The agent is instructed to derive a formal threat model and protocol invariants (S5.0) before analyzing the code, providing a baseline for expected behavior.
Capability inventory: The agent has the ability to compile code, run fuzzers, and file vulnerability reports (PVRs).
Sanitization: Findings must survive a manual triage and adversarial refutation process (S6) before being accepted, reducing the risk of being misled by comments or obfuscated logic in the target code.
[CREDENTIALS_UNSAFE]: The skill requires API keys for block explorers (ETHERSCAN_API_KEY, BLOCKSCOUT_API_KEY) and notification services (RESEND_API_KEY). These are handled using platform-native requirement fields and secret placeholders rather than being hardcoded, which is the standard practice for secure agent configuration.

sc-audit