The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and interpret project artifacts such as PRDs, READMEs, and source code, which could contain malicious instructions.
Ingestion points: The 'Scanner layer' in SKILL.md identifies primary and secondary sources including external documentation, API specifications, and code entry points.
Boundary markers: The skill uses Markdown code blocks for output but does not explicitly define delimiters to isolate or ignore instructions embedded within the ingested data.
Capability inventory: The skill possesses file system access (Read, Write, Edit) and shell execution capabilities through the Bash toolset (grep, rg, npx, node, mmdc).
Sanitization: The skill implements a 'Discovery & Inquiry Protocol' which prohibits the agent from making guesses about logic and requires it to halt and ask clarifying questions, providing a human-in-the-loop sanitization step.
[COMMAND_EXECUTION]: The skill executes a provided shell script to validate the syntax of generated diagrams.
Evidence: The scripts/lint.sh file is invoked to perform syntactic checks. The script follows security best practices, such as using mktemp for isolated workspaces, quoting variables to prevent injection, and relying on well-known utilities like grep, awk, and mmdc (Mermaid CLI).

mermaid-architect