The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external content (articles, papers, notes) to generate summaries and cross-references.
Ingestion points: External files in the raw/ directory are processed via scripts/ingest_source.py.
Boundary markers: Instructions in SKILL.md and wiki-ingestor.md mandate that the agent discuss takeaways and proposed edits with the user before writing to the vault.
Capability inventory: The sub-agents possess Bash for running management scripts and Write capabilities for the Markdown knowledge base.
Sanitization: The scripts perform basic text extraction but rely on manual user verification of the LLM's synthesis to prevent malicious instruction following.
[COMMAND_EXECUTION]: The skill uses local shell commands to execute its management tools.
Evidence: Automated workflows for vault initialization, indexing, and ingestion involve running the provided Python scripts via the Bash tool.

llm-wiki