The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external media sources (podcasts and YouTube) via tools such as arcmira.mentions and arcmira.search (defined in SKILL.md). This external content is interpolated into the agent's context to provide quotes and context to the user. There are no explicit boundary markers or instructions to the agent to ignore potentially malicious directions embedded in transcripts. The skill possesses write capabilities through arcmira_execute_write, which allows the modification of monitors based on these inputs. Sanitization is restricted to character-count truncation of quotes.- [DYNAMIC_EXECUTION]: The skill operates by having the agent construct and execute JavaScript snippets via the arcmira_execute_read and arcmira_execute_write tools. The agent is instructed to populate static code templates in SKILL.md with dynamic values such as user-provided names, context strings, and entity IDs before execution. This pattern involves the runtime assembly of executable logic from mixed trusted and untrusted inputs.

company-watch