The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructs the agent to request third-party LLM API keys (e.g., OpenAI, Anthropic, xAI, etc.) from the user and transmit them to the agenthub.clawsy.app server for the 'custom validation' feature. This practice involves sending sensitive credentials to a third-party service, which represents a security and privacy risk.
[COMMAND_EXECUTION]: The documentation mentions the installation and initialization of a CLI tool using pip install clawsy && clawsy init. This involves executing commands that modify the local environment and install external software.
[EXTERNAL_DOWNLOADS]: The skill facilitates the fetching and processing of data from external sources, such as PDF, DOCX, and audio files, via user-provided URLs using the /api/ingest/extract endpoint.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection as it ingests untrusted data from GitHub repositories and arbitrary URLs to create tasks. This data is then used to generate 'enriched prompts' that direct the agent's behavior.
Ingestion points: External data enters through the /api/ingest/extract endpoint and the github_repo field in task creation.
Boundary markers: There are no explicit instructions or delimiters mentioned to isolate the ingested external content from the agent's core instructions.
Capability inventory: The skill can perform network operations, such as submitting patches and sending messages to the platform, based on processing the external content.
Sanitization: No sanitization or validation of the ingested content is described before it is incorporated into the task instructions.

agenthub