The Agent Skills Directory

[DYNAMIC_EXECUTION]: The file reference/repository-recon.md contains a Python script snippet that monkey-patches the socket.getaddrinfo method at runtime. This allows the skill to redirect network requests for specific virtual hosts to a hardcoded target IP, bypassing standard DNS resolution for repository dumping.
[INDIRECT_PROMPT_INJECTION]: The skill's core functionality involves cloning and scanning untrusted remote repositories from platforms like GitHub and GitLab. Malicious repositories could contain embedded instructions or deceptive patterns intended to manipulate the agent during the analysis phase.
Ingestion points: git clone operations and subsequent file parsing in Phase 3 and Phase 4 described in reference/repository-recon.md.
Boundary markers: None identified in the instructions for delimiting untrusted code from analysis commands.
Capability inventory: Subprocess execution for CLI tools (gh, git, trufflehog), file system writes to {OUTPUT_DIR}/, and network access via multiple tools.
Sanitization: No specific sanitization logic is provided for the content of the cloned files before analysis.
[DATA_EXFILTRATION]: The skill is purpose-built to locate and extract sensitive credentials, including AWS access keys, private RSA/SSH keys, database connection strings, and API tokens. It provides extensive search dorks for identifying these secrets across organization repositories and developer history.
[COMMAND_EXECUTION]: The skill relies on external CLI tools including gh, git, trufflehog, gitleaks, gitrob, and grep to interact with remote services and the local filesystem. These tools are used to enumerate organizations, search code, and analyze git history.

osint