The Agent Skills Directory

[DATA_EXFILTRATION]: Automated synchronization of user documents to a remote repository.
Evidence: git -C "$V" add -A && git -C "$V" commit -m "ai-reindex: rebuild wikilink indexes" && git -C "$V" push in SKILL.md.
Impact: The skill performs a automated git push operation that transmits the contents of the user's Obsidian vault (located at ~/Documents/obsidian/obsidian) to a remote server. By using git add -A, the skill stages and pushes every modification within the repository, including files and notes not related to the index regeneration. This exposes the entire vault's content to the configured Git origin.
[COMMAND_EXECUTION]: Execution of system shell commands to manage files and version control.
Evidence: Use of test -d, ls -1, and git commands in SKILL.md.
Impact: The skill relies on direct execution of system commands to interact with the file system and perform network synchronization operations.
[PROMPT_INJECTION]: Indirect prompt injection surface via unsanitized file system metadata.
Ingestion points: The skill reads directory structures using ls -1 and file contents via the Read tool within the engineering/ vault directory (as seen in SKILL.md and references/index-conventions.md).
Boundary markers: No specific delimiters or instructions to ignore embedded content are provided when formatting the resulting index markdown.
Capability inventory: The skill can overwrite files (Write) and perform network operations (git push).
Sanitization: There is no evidence of sanitization or escaping of filenames, project names, or feature slugs before they are interpolated into the generated index.md files. This allows attacker-controlled file names to potentially inject malicious markdown or instructions into the vault, which are then persisted and potentially synced externally.

ai-reindex