The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to use standard system utilities including git for version control, grep for searching the knowledge base, and pandoc for generating PDF documents from markdown files. These commands are used for their intended purpose of managing a file-based wiki. The pandoc command specifically references a local stylesheet at ~/.local/share/makerskills/render.css.
[PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection because it processes content contributed by multiple team members or ingested from automated external sources (e.g., call transcripts).
Ingestion points: Reads markdown files from multiple structured directories within the vault path (e.g., people/, meetings/, sops/, decisions/, customer-language/, recurring-questions/, sales-objections/).
Boundary markers: Uses markdown headers (e.g., ## Sources, ## Details) to structure content, but lacks specific instructions to disregard potentially malicious commands embedded within the raw data ingested from external transcripts or multi-author sources.
Capability inventory: Perform file reads and writes within the vault path, execute local shell commands (git, grep, pandoc), and utilize the deep-research tool for external queries.
Sanitization: There are no instructions for sanitizing or escaping the content of captured transcripts or notes before they are processed by the agent.

company-brain