The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's operation depends on executing a suite of CLI tools (e.g., nmap, dig, curl, subfinder, gau, ffuf). All entry points, including recon_pipeline.sh, findings_gen.py, and build_xlsx.py, implement strict regular expression validation on the domain argument. This sanitization ensures that only valid hostnames are processed, preventing shell metacharacters from being injected into the command pipeline.
[EXTERNAL_DOWNLOADS]: The documentation and metadata for the host enumeration workflow recommend installing ffuf from its official GitHub repository. This reference is to a well-known, legitimate security tool and occurs in the context of environment setup rather than hidden or automated malicious execution.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external OSINT providers (CertSpotter, crt.sh, HudsonRock) and live target HTTP responses to generate its report.
Ingestion points: Subdomain lists, breach metadata, and HTTP/JS content are fetched using curl and gau in recon_pipeline.sh and the Node.js workflow.
Boundary markers: The agent prompts in host_enum.workflow.js include specific instructions to process 'Authorized red-team recon' and return data in a compact schema, which helps maintain operational focus.
Capability inventory: The skill writes evidence artifacts and generated findings to the ~/Research/engagements/ directory and performs network discovery.
Sanitization: The pipeline uses tools like jq, awk, and Python regular expressions to extract specific patterns (e.g., CIDRs, version strings, secret patterns) rather than blindly executing or trusting the external content, which limits the attack surface for injection.

osint-autopilot