The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill executes npx agentcash, which downloads and runs a package from the NPM registry at runtime. The package is not from a recognized trusted vendor, creating a risk of executing unverified or potentially compromised code.- [REMOTE_CODE_EXECUTION]: By using npx, the skill triggers the remote download and immediate execution of scripts from a third-party registry, which is a significant vector for remote code execution.- [DATA_EXFILTRATION]: User PII, including names, emails, and social media handles, is extracted from local markdown files and transmitted to external endpoints (stableenrich.dev) via the agentcash CLI tool. This represents a privacy risk as sensitive data is shared with non-whitelisted third-party services.- [COMMAND_EXECUTION]: The skill constructs and executes shell commands incorporating data parsed from user files (e.g., LinkedIn URLs, names) into command arguments. Without explicit sanitization, this creates a potential surface for command injection if user-controlled files contain malicious shell metacharacters.- [PROMPT_INJECTION]: The skill has a surface for indirect prompt injection because it parses untrusted data from user-managed markdown files to determine its workflow and command arguments.
Ingestion points: Files located in directories such as wiki/people/, people/, contacts/, or network/ (SKILL.md).
Boundary markers: None identified. The instructions do not specify using delimiters or 'ignore' instructions when processing external content.
Capability inventory: Subprocess execution via npx, network operations via agentcash fetch, and file system write access for updating markdown files.
Sanitization: Absent. Extracted frontmatter values are interpolated directly into shell command strings and prompts.

enrich-knowledge-graph