The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs repository maintenance using standard developer tools, including git add, git commit, git push, and uv run prek. These commands are used to lint modified files and persist updates to the remote origin.
[INDIRECT_PROMPT_INJECTION]: The skill processes metadata and content from markdown files within the ./research/ directory to determine staleness and update tracking sections. This represents a data ingestion surface where external content (if modified by an adversary) could theoretically influence the agent's logic during the inventorying phase or when constructing prompts for the research-curator sub-agents.
Ingestion points: Markdown files in ./research/**/*.md (YAML frontmatter and Freshness Tracking sections).
Boundary markers: None explicitly defined for the parsing logic beyond markdown section identification.
Capability inventory: File writes to ./research/, network access via MCP tools and GitHub CLI, and shell command execution for git workflows.
Sanitization: Not explicitly described; the skill relies on the LLM to parse and filter the inventoried data.
[EXTERNAL_DOWNLOADS]: Fetches fresh research data from primary sources and GitHub repository metadata using the gh CLI and search tools (mcp__exa). These operations are aligned with the skill's primary function of keeping research entries current.

refresh-research