The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data, such as case study text and company descriptions, which is then interpolated into LLM prompts for classification and resonance analysis. This represents a vulnerability surface where maliciously crafted external content could potentially influence the agent's logic or output.
Ingestion points: Inputs defined in SKILL.md for case_study_text and company_description are processed by the LLM prompts.
Boundary markers: The skill uses delimiter tags (e.g., <case study text>, <company description>) to wrap external content, but does not provide specific instructions to the model to ignore or sanitize embedded instructions within these blocks.
Capability inventory: The workflow employs the scrape-website tool and Clay CLI routines to perform network lookups and search operations based on the processed data.
Sanitization: No evidence of input validation, escaping, or filtering of the external text is present in the provided instructions.
[DYNAMIC_EXECUTION]: The skill includes Python code snippets in clay-workflow.md designed to be implemented as workflow nodes. These scripts handle logic for filtering industry candidates and verifying site liveness using the content of scraped homepages. While the code is static and template-based, it represents runtime logic that processes external site data.

playbook-lookalikes