The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/coverage_audit.py employs subprocess.run() to call the local git binary. This is used specifically to perform a git show command, allowing the skill to compare current file content against previous versions in the repository. This mechanism is a core part of the skill's data-loss prevention audit.
[INDIRECT_PROMPT_INJECTION]: As the skill involves the agent reading and triaging user-provided notes, it is exposed to potential indirect prompt injections within those notes. The skill addresses this vulnerability surface by requiring the agent to 'Read before you classify' and by providing automated scripts to verify that note content is preserved and correctly structured, reducing the likelihood of malicious instructions causing silent data loss or misclassification.
Ingestion points: User-provided markdown notes read during triage (referenced in SKILL.md and processed by scripts in scripts/).
Boundary markers: The workflow instructions emphasize manual reading and understanding of notes before classification.
Capability inventory: File system read/write, local git command execution via subprocess.run.
Sanitization: None explicitly mentioned for prompt content, but scripts perform structural and content coverage audits.

obsidian-vault-triage