The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions are focused entirely on marketing workflows and content analysis. No markers for bypassing safety filters, disregarding instructions, or activating unauthorized modes were detected.\n- [CREDENTIALS_UNSAFE]: No hardcoded API keys, tokens, or sensitive credentials were found. The skill references local memory files for context management but does not access sensitive system directories like .ssh or .aws.\n- [REMOTE_CODE_EXECUTION]: The skill documentation in references/atom-extraction.md contains an explicit safety constraint prohibiting the installation of external packages or calling libraries such as whisper or pandas, ensuring the agent operates within a restricted text-only environment.\n- [COMMAND_EXECUTION]: There are no shell commands, subprocess calls, or persistent system modifications present in the skill instructions. Calculations like the Virality Heuristic and Jaccard similarity are performed as natural language reasoning steps by the agent.\n- [DATA_EXFILTRATION]: No network operations to unknown or suspicious domains were found. The skill mentions opt-in connectors for social platform metrics, but these are described as standard, external integrations rather than hidden exfiltration channels.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided content such as transcripts and reviews. While this is an ingestion surface for untrusted data, the skill is safe as it lacks the high-privilege capabilities required to turn such input into a functional attack vector.

content-amplifier