The Agent Skills Directory

[DATA_EXFILTRATION]: The skill performs exhaustive scanning of highly sensitive file paths across 20+ categories to identify exposed credentials. Targeted paths include infrastructure keys (~/.aws/credentials, ~~/.kube/config), developer tokens (~~/.ssh/, ~/.git-credentials, ~/.npmrc), and crypto wallets (Solana id.json). While the data is recorded in a local report rather than sent to a remote server, the broad access to these paths represents a significant security surface area.
[COMMAND_EXECUTION]: The skill uses numerous shell commands to inventory the system and check for malicious activity. This includes file discovery via find and grep, inspection of application databases via strings, persistence auditing via launchctl and crontab, and process monitoring via ps and lsof. It also provides commands for evidence preservation and cleanup of compromised software.
[PROMPT_INJECTION]: The skill faces an indirect prompt injection surface as it ingests and processes data from local files that could be controlled by an attacker if the machine is already compromised. \n
Ingestion points: Data is read from .env files, shell history, and lockfiles (SKILL.md).\n
Boundary markers: The instructions do not specify the use of delimiters when interpolating external file content into the generated audit report.\n
Capability inventory: The skill has permissions to read files, execute shell commands, and write reports to the temporary directory.\n
Sanitization: The skill mitigates risks by requiring the redaction of discovered tokens (e.g., using sed to mask the middle of strings) before presentation.

dev-security-audit