The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructions guide the agent to read an API token from a local .env file and use it to authenticate requests to api.extruct.ai. This behavior is consistent with standard API usage for the skill's intended purpose and does not represent an unauthorized exfiltration attempt.
[COMMAND_EXECUTION]: The reference documentation includes shell snippets utilizing curl, grep, cut, and python3 to facilitate the research workflow. These commands are used for credential extraction, API polling, and JSON parsing of research data within the project's scope.
[PROMPT_INJECTION]: The skill possesses a potential surface for indirect prompt injection as it ingests research reports and source data from the Extruct API. This is inherent to its primary function of gathering external intelligence.
Ingestion points: Authenticated retrieval of research reports and metadata from https://api.extruct.ai/v1/deep_research_tasks/.
Boundary markers: The skill workflow specifies that the agent should read from the report rather than memory, although it does not define rigid delimiters for separating the fetched content from the main system context.
Capability inventory: The agent has permissions to write structured dossier files to the claude-code-gtm/accounts/ directory and execute network commands for research task management.
Sanitization: The process relies on the structured nature of the API's JSON output and the agent's instructions to provide citations, rather than explicit technical sanitization of the research text.

account-research