The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous shell commands for target reconnaissance and includes Python and Playwright scripts for automated credential validation.\n
Evidence: msftrecon -d client.example in SKILL.md.\n
Evidence: The authentication script disables SSL verification (ssl.CERT_NONE) and the browser-based flow ignores certificate errors (--ignore-certificate-errors), making authentication traffic vulnerable to Man-in-the-Middle attacks.\n- [REMOTE_CODE_EXECUTION]: The skill requires downloading and installing third-party tools from external, unverified repositories.\n
Evidence: pip install --break-system-packages msftrecon o365spray and go install -v github.com/projectdiscovery/... in SKILL.md.\n- [PRIVILEGE_ESCALATION]: The skill instructs the agent to bypass standard Python environment protections during tool installation.\n
Evidence: Use of the --break-system-packages flag with pip install in SKILL.md.\n- [INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection by processing external data such as leaked credentials and target URLs.\n
Ingestion points: email, password, and target_sp_url parameters in the scripts within SKILL.md.\n
Boundary markers: None present.\n
Capability inventory: Network communication via urllib and Playwright, and local file writing via json.dump in SKILL.md.\n
Sanitization: No sanitization or validation of the ingested external data is implemented.

m365-entra-attack