The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user data from multiple sources to generate technical design documents. An attacker could embed malicious instructions within a BRAINSTORM or DEFINE Markdown file to influence the architecture or the agent's behavior.
Ingestion points: The skill reads BRAINSTORM_*.md, DEFINE_*.md, and UX_REVIEW_*.md files provided by the user.
Boundary markers: The skill establishes a hierarchy of sources (Define > Brainstorm) but lacks explicit delimiters or instructions to ignore embedded prompts within the content of these files.
Capability inventory: The skill has the ability to write files to the local filesystem (sdd/DESIGN_*.md) and execute a local Python validation script (scripts/validate_design.py).
Sanitization: There is no specific sanitization logic to filter out prompt injection attempts from the input Markdown files beyond the instruction to "Converter apenas o que afeta decisões técnicas."
[COMMAND_EXECUTION]: The skill explicitly instructs the agent to execute a local Python script for structural validation.
Evidence: python scripts/validate_design.py <DESIGN.md> is called in step 11 of the procedure. The script itself is provided within the skill and uses only standard libraries (re, sys, pathlib, collections), which is a legitimate and safe use of command execution for linting purposes.
[SAFE]: The skill promotes several security best practices in the design it generates, including least privilege, avoiding hardcoded secrets in code, and validating PII/sensitive data handling.

sdd-design