The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses powerful shell utilities such as ripgrep, find, git, and docker to perform deep recursive scans of the local filesystem, environment variables, and full repository history.
[DATA_EXFILTRATION]: It specifically targets and reads highly sensitive file paths including SSH private keys (id_rsa, .pem), cloud provider credentials (AWS IAM, GCP Service Accounts), and environment configuration files (.env). Accessing these files exposes their contents to the AI agent's processing context.
[CREDENTIALS_UNSAFE]: The skill contains a large collection of regular expressions designed to extract hardcoded secrets for various services like OpenAI, AWS, Google Cloud, Stripe, and Meta from the codebase.
[REMOTE_CODE_EXECUTION]: Instructions reference and rely on several external scripts (e.g., scripts/secret_scanner.py, scripts/audit_report.py) that are not provided within the skill's distribution, which could lead to the execution of untrusted or unverified local code.
[COMMAND_EXECUTION]: Suggests administrative actions that typically require root or sudo privileges, such as configuring system firewalls (ufw/iptables), modifying SSH server parameters, and setting up systemd services.
[PROMPT_INJECTION]: The skill performs automated scans of untrusted data (local files and repository history) which presents a risk for indirect prompt injection if those files contain malicious instructions meant to influence the agent.
Ingestion points: Local files and Git history scanned via rg and git grep.
Boundary markers: While log redaction is mentioned, no structural delimiters are used to separate ingested file content from the agent's instructions.
Capability inventory: High-privilege shell execution, file system access, and interaction with local scripts.
Sanitization: Basic redaction using sed for environment variables is the only automated filter mentioned.

cred-omega