The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs and executes shell commands (curl and jq) using variables like {org_name} and {twitter_handle} derived from external API data. This pattern is vulnerable to command injection if the data providers return malicious strings designed to break out of the shell command context.
[DATA_EXFILTRATION]: User-provided lead data, including email addresses and names, is shared with multiple external enrichment services such as Apollo, Hunter, Tomba, and Sixtyfour.
[EXTERNAL_DOWNLOADS]: The skill's operation depends on downloading data from various third-party enrichment APIs and the public GitHub API.
[PROMPT_INJECTION]: An indirect prompt injection surface is present because the agent is tasked with parsing and classifying information (company descriptions and keywords) from external sources that may contain adversarial content.
Ingestion points: Data fields from Apollo, Hunter, Brand.dev, and Sixtyfour (e.g., description, keywords, industries) are processed in Phase 1 and 4.
Boundary markers: None; the skill does not use delimiters or instructions to ignore embedded commands in the fetched data.
Capability inventory: Execution of API calls via orth run and shell commands via curl and jq.
Sanitization: None; the skill lacks validation or escaping for the data interpolated into shell commands or processed by the LLM for classification.

gtm-enrichment-smart