The Agent Skills Directory

[SAFE]: The skill instructions enforce strict factual boundaries. It explicitly forbids the agent from inventing names, titles, round sizes, or quotes, requiring the use of placeholders like '[QUOTE NEEDED FROM...]' for missing information. This prevents the generation of deceptive or hallucinatory content.- [SAFE]: File system access is narrowly scoped and follows the principle of least privilege. The skill is restricted to reading context from specific 'knowledge/' directories and writing artifacts exclusively to 'output/press-release/'. No patterns of sensitive file access (e.g., SSH keys, environment variables) or unauthorized path traversal were found.- [SAFE]: The skill does not perform any network operations, external downloads, or shell command executions. There are no signs of obfuscation, dynamic code execution, or persistence mechanisms.- [INDIRECT_PROMPT_INJECTION]: The skill has a data ingestion surface as it reads from external knowledge files and user-supplied input to generate press releases. However, the risk is neutralized by the skill's rigid adherence to specific journalistic frameworks (Inverted Pyramid) and a mandatory 11-point self-check process that validates figures against their sources before finalizing the output. As this represents standard functionality for a writing agent and includes significant guardrails, it does not escalate the severity.

press-release-writer