The Agent Skills Directory

[SAFE]: The skill follows security best practices by implementing a 'Research Firewall' that isolates research subagents from project-sensitive files and context, mitigating the risk of data leakage during external data processing. This is defined in SKILL.md and references/run.md.
[COMMAND_EXECUTION]: The skill executes local Python scripts using uv run to handle deterministic tasks such as citation validation, staleness calculation, and HTML generation. These scripts are part of the skill's own distribution or the associated framework, facilitating controlled logic execution.
[PROMPT_INJECTION]: The skill processes untrusted content from external reports and web search results. Evidence chain: (1) Ingestion points: references/process.md (imports) and references/run.md (search results); (2) Boundary markers: 'Research Firewall' instructions in SKILL.md mandate subagents receive only their brief and no project context; (3) Capability inventory: includes file writing, tool calls, and script execution; (4) Sanitization: recon_kit.py performs URL validation and HTML escaping.
[SAFE]: The recon_kit.py script includes robust sanitization logic for generating HTML briefings. It validates URL schemes using urlparse to ensure only http and https protocols are used and employs html.escape to prevent potential cross-site scripting (XSS) in generated reports.

bmad-deep-recon