The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from wiki markdown files to perform its analysis. Specifically, the 'L15 — Near-duplicate pages' check in references/lint-checks.md uses an 'LLM-as-judge' mechanism where the model evaluates content from the 'title', 'Summary', and 'Key Facts' sections of wiki pages.
Ingestion points: The agent reads content from wiki/*.md, .obsidian-wiki.config, and wiki/.manifest.json.
Boundary markers: The skill instructions do not specify explicit delimiters or warnings to ignore embedded instructions when the LLM judge evaluates the ingested content.
Capability inventory: The skill is limited to reading files and appending a single entry to wiki/log.md. It has no network access or tool-use capabilities defined that would allow for data exfiltration or system modification.
Sanitization: The skill identifies the use of yaml.safe_load for parsing frontmatter in the L14 check, which is a key security measure against unsafe deserialization.
[SAFE]: The skill demonstrates safe coding practices by explicitly requiring yaml.safe_load (or equivalent) for parsing metadata and configuration files. This prevents potential remote code execution (RCE) vulnerabilities that can occur when using unsafe YAML loading methods on untrusted input.

wiki-lint