The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs a version check by fetching a plain-text version string from the author's GitHub repository (https://raw.githubusercontent.com/sanbir/solana-auditor-skills/main/solana-auditor/VERSION). This is a routine operation to ensure the user is running the latest security logic and follows the trust-scope rule for vendor resources.
[COMMAND_EXECUTION]: Local shell commands (find, cat, curl, mktemp) are used to discover source files, manage temporary audit bundles, and check for updates. These operations are restricted to the skill's operational scope and the temporary directory it creates.
[PROMPT_INJECTION]: The skill creates an indirect prompt injection surface by ingesting untrusted .rs files from the project being audited. While malicious code comments could attempt to influence the auditing agents, the orchestrator uses structural delimiters (headers and code fences) and specialized judging logic to mitigate these risks. This is a standard characteristic of automated code auditing tools.
[DATA_EXFILTRATION]: The skill reads local project files to perform its audit. Access is restricted to the program directory and specific security reference files provided with the skill. No sensitive system files, such as SSH keys or environment secrets, are targeted, and no unauthorized exfiltration was detected.

solana-auditor