The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local system commands and scripts to maintain the agent ecosystem. Specifically, scripts/main.cjs uses node:child_process.spawnSync to run registry generation scripts, and the SKILL.md workflow includes instructions for running npm run gen:all-registries to ensure consistency across the agent tool and skill indexes.
[EXTERNAL_DOWNLOADS]: The skill performs mandatory research by fetching data from external sources like arXiv using WebSearch and WebFetch. To mitigate the risk of processing untrusted content, it implements a comprehensive Security Review Gate that scans for excessive size, binary data, unauthorized tool patterns, prompt injections, and potential data exfiltration before the content is integrated into the agent definitions.
[PROMPT_INJECTION]: While the skill includes automated detection for malicious prompt injection in external data, it also features a mandated protocol injection rule. This functionality is designed to programmatically inject standardized search and memory instructions into target agents to ensure they adhere to the framework operational standards.
[DYNAMIC_EXECUTION]: The utility script scripts/main.cjs contains logic in updateRoutingTableKeywords to programmatically modify an existing JavaScript file (.claude/lib/routing/routing-table-intent-keywords.cjs). It identifies insertion points within the code and injects new entries into the exported routing table to facilitate dynamic capability updates.

agent-updater