The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted user data in the form of existing Semgrep rule YAML content. This introduces a surface for indirect prompt injection where malicious instructions could be embedded in the rule's metadata (e.g., message or description fields) to attempt to influence the agent's porting process.
Ingestion points: Untrusted YAML content enters the context via user input as specified in SKILL.md and processed during the Phase 1: Applicability Analysis described in workflow.md.
Boundary markers: Absent; the instructions do not currently require the agent to wrap input rule content in specific delimiters or use safety instructions when parsing the YAML.
Capability inventory: The skill uses Bash (to run the semgrep binary), and Write/Edit (to produce new rule and test files).
Sanitization: No explicit sanitization or structure validation of the input YAML is documented before the analysis phase.
[COMMAND_EXECUTION]: The skill instructions and workflow utilize the Bash tool to execute legitimate semgrep CLI commands. These include AST dumping (--dump-ast), rule validation (--validate), and regression testing (--test). These are essential, well-documented operations performed on locally created or existing files and do not involve piped remote execution.
[EXTERNAL_DOWNLOADS]: The documentation includes references to the official Semgrep documentation (semgrep.dev) and security research guidelines from Trail of Bits. These are trusted, well-known service providers and represent safe documentation links rather than the downloading of untrusted scripts or executable payloads.

semgrep-rule-variant-creator