The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted natural language data from the user (e.g., 'Rule intent in plain English') and uses it to draft executable Cypher queries.
Ingestion points: The 'Required inputs' section in SKILL.md specifies that 'Rule intent in plain English' is used to define findings and query logic.
Boundary markers: No specific delimiters or warnings to ignore embedded instructions are provided when interpolating user input into the query drafting process.
Capability inventory: The skill uses subimageRunCypher to execute queries against the live tenant graph and subimageCreateCustomRule to persist configurations.
Sanitization: The instructions explicitly state that server-side validation rejects write operations (CREATE, MERGE, DELETE, SET, REMOVE), providing a baseline defense against database modification attacks.
[DYNAMIC_EXECUTION]: The skill generates Cypher query strings at runtime based on user intent and executes them via the subimageRunCypher tool. While it includes guidance to use LIMIT clauses and read-only operations, this pattern of dynamic script generation and execution represents a potential risk if the AI is misled by malicious input.
[SAFE]: The skill references the vendor's official documentation at app.subimage.io, which is consistent with the author 'subimagesec'. The tools subimageCreateCustomRule, subimageRunCypher, and subimageListRules are legitimate vendor-provided MCP tools for this platform.

create-custom-rule