The Agent Skills Directory

Indirect Prompt Injection (HIGH): The skill's core function is to ingest untrusted data from an external codebase to generate organizational knowledge.
Ingestion points: Uses cat on READMEs, package manifests, and CI/CD workflows (.github/workflows/*.yml). Background agents are tasked with sampling source files and searching for specific comment strings.
Boundary markers: None. Data from the codebase is processed and interpolated into prompts or shell commands without delimiters or 'ignore embedded instructions' warnings.
Capability inventory: The skill can execute arbitrary shell commands (find, awk, sed) and write data to an external service via the neotex CLI.
Sanitization: No sanitization or validation of the scanned content is performed before it is used to generate 'learnings' or reports.
Command Execution (HIGH): The skill relies heavily on shell execution for file system discovery and data processing (find . -type f | wc -l, awk, sed). This pattern is susceptible to command injection if filenames in the scanned codebase are maliciously crafted (e.g., containing backticks or semicolons).
Data Exfiltration (MEDIUM): Collected architectural data, decision logs, and code patterns are sent to an external tool (neotex add). The neotex utility is not a recognized trusted source, and the destination of the data is not transparent.
Data Exposure (LOW): The skill explicitly targets sensitive configuration areas, including .env.example and CI/CD workflow files, which may leak internal infrastructure details or naming conventions to the external service.

neotex-init