The Agent Skills Directory

[PROMPT_INJECTION]: The skill's instructions are focused on guiding the user through reflection exercises and do not contain any patterns intended to bypass safety filters, extract system prompts, or override agent behavior.
[DATA_EXFILTRATION]: The skill manages data by writing reflection entries to a local directory (the MELdigitalgarden vault) via a filesystem tool or providing them as markdown artifacts. There is no evidence of unauthorized sensitive file access or network operations targeting external domains.
[REMOTE_CODE_EXECUTION]: The skill does not download external scripts, install third-party packages, or execute remote code. All operations are limited to text processing and local file management.
[COMMAND_EXECUTION]: The skill does not perform shell command execution or subprocess spawning. It relies on standard filesystem interactions to save the generated markdown content.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input (reflection answers) to generate its output. While this creates a surface for potential formatting issues within the markdown vault, the risk is negligible as it is intended for a personal journaling context with no capability for downstream code execution or privilege escalation.
Ingestion points: User responses to the four reflection modes (SKILL.md).
Boundary markers: Obsidian-compatible YAML frontmatter and Markdown headers are used to structure the output.
Capability inventory: The skill can write to the filesystem if a filesystem MCP is connected; otherwise, it returns content as an artifact.
Sanitization: The skill explicitly aims to preserve the user's voice and does not perform aggressive sanitization of input, which is consistent with its purpose as a journaling tool.

journal-reflection