The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell command templates for git, gh, and rg (ripgrep) that incorporate user-influenced parameters such as file paths, commit hashes, and search patterns. It also constructs SQL queries for Databricks using potentially untrusted strings in LIKE and ILIKE clauses. Without proper sanitization of these inputs, the skill is vulnerable to command injection and SQL injection.\n- [PROMPT_INJECTION]: The skill processes large volumes of untrusted text from external platforms, creating a surface for indirect prompt injection. Ingestion points: Slack conversations, GitHub pull request comments, Linear tickets, and Notion documents (found in references/sources/). Boundary markers: The investigator-prompt.md template provides instructional boundaries such as "quote, don't paraphrase," but these are not secure delimiters. Capability inventory: The skill can execute shell commands, perform SQL reads, and spawn subagents via spawn_agent. Sanitization: There is no evidence of explicit sanitization or filtering of content retrieved from external sources before it is processed by the synthesizer subagent.

why