The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted markdown data from the Obsidian vault (bodies and frontmatter) to perform contradiction detection, synthesis gap analysis, and PII auditing. The lack of explicit boundary markers or content sanitization could allow malicious instructions within a vault page to influence the agent's behavior during the audit process.
Ingestion points: Vault markdown files, AGENTS.md, index.md, and trust metadata files.
Boundary markers: No specific delimiters or "ignore" instructions are used when interpolating page content into checks.
Capability inventory: Local file reading and writing, git operations (add, commit), ripgrep searches, and vault indexing via QMD.
Sanitization: No explicit sanitization of vault content is documented before processing.
[COMMAND_EXECUTION]: The skill relies on shell commands for vault interaction, including ripgrep (rg), git for snapshots, and custom CLI tools (obsidian-wiki, qmd). While standard for development utilities, these tools provide the agent with broad execution capabilities within the vault's environment.
[DATA_EXFILTRATION]: The skill is designed to search for and identify sensitive patterns such as passwords, API keys, tokens, and SSNs. While intended for classification and tagging (e.g., visibility/pii), these patterns are processed and reported within the agent context, which could lead to sensitive information being exposed in logs or summaries.

wiki-lint