The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/dashboard.py uses subprocess.run() to invoke the companion script scripts/h1_reference.py. This is used to orchestrate the internal toolkit and provide a local web UI for querying HackerOne disclosures.
[EXTERNAL_DOWNLOADS]: Reference files such as references/tooling-install.md contain numerous commands instructing the operator to download and install security tools from third-party sources (e.g., go install, pip install, git clone). These target well-known open-source projects from organizations like ProjectDiscovery and OWASP.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection by processing untrusted data from local file scans and external API responses.
Ingestion points: Untrusted data enters the agent context via scripts/secret_scan.py (which reads arbitrary local files) and scripts/h1_reference.py (which fetches data from the HackerOne GraphQL API).
Boundary markers: The scanning script uses structured JSONL output to delimit findings.
Capability inventory: The skill possesses file system read capabilities, network request capabilities (urllib.request), and local command execution for helper scripts.
Sanitization: Match evidence in scripts/secret_scan.py is truncated to 80 characters to mitigate potential injection or buffer issues.
[SAFE]: Automated scanners flagged socketscan.io (a cryptocurrency bridge explorer) as phishing and references/specialized-osint.md as containing malware. These are evaluated as reputation-based flags common for resources in the crypto and threat intelligence sectors. Given the skill's stated purpose as an "offensive OSINT" arsenal, the inclusion of these links to research tools and IOC feeds is expected and consistent with legitimate security research.

offensive-osint