The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill provides explicit regex patterns to identify sensitive information including passwords, API keys, AWS access keys, and private keys. It instructs the agent to scan incoming files for these patterns and move them to a designated credentials_folder. While framed as a 'Credential Guard,' this behavior facilitates the automated discovery and centralization of secrets by the AI agent.
[PERSISTENCE]: The skill uses the CronCreate tool to establish a recurring schedule for autonomous execution (4 times per day). This allows the skill's logic to run in the background without direct user initiation.
[COMMAND_EXECUTION]: The skill relies on the Bash tool to execute local shell scripts (scripts/lint-check.sh and scripts/update-index.sh) and perform filesystem operations like moving files. The integrity and behavior of these scripts are critical to the skill's safety.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface for indirect prompt injection:
Ingestion points: Reads untrusted files from a user-defined inbox_dir (SKILL.md).
Boundary markers: The skill lacks instructions for using delimiters or 'ignore' instructions when synthesizing content from vault files.
Capability inventory: The agent has the ability to write to the filesystem, execute shell scripts via Bash, and modify its own schedule via CronCreate.
Sanitization: There is no mention of sanitizing or escaping content retrieved from the vault before it is processed by the LLM for synthesis or linting.

obsidian-brain