The Agent Skills Directory

[SAFE]: The skill implements strong internal governance, including a "Language contract" for multi-lingual consistency and an "Anti-Confirmation Bias" rule that explicitly forbids reading the 'user_context.md' file. This reduces the risk of the agent being influenced by potentially malicious or misleading user-provided narratives.- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from 'structured_data.md' and various 'core' reports. While this represents a potential attack surface for indirect prompt injection, the risk is mitigated by strict data-mapping rules (e.g., "SAV reading iron rule") that force the agent to read from specific structured tables rather than interpreting raw text, and by the lack of high-privilege capabilities beyond writing markdown files.- [COMMAND_EXECUTION]: The skill utilizes a 'write_to_file' mechanism to output its analysis in phases. This is the intended functional behavior for a report-generation tool and is not associated with arbitrary or dangerous shell command execution.

vedic-career