All stack tools
Surface
Claude Code Skills + AGNT
Reusable prompt + instruction bundles — forwarded into AGNT's soul loader.
Official sourceWhat it is
Skills are Claude Code's reusable capability unit. A skill is a markdown bundle with a name, description, trigger heuristics, and instructions. Claude Code routes into the skill when its description matches the user's intent and the instructions become part of the active prompt.
The abstraction matters because it makes capability composition declarative. You write a skill once, version it, and it runs in every session that matches the trigger — no Python, no adapter code.
Where AGNT fits
- AGNT's soul loader is the per-agent structural memory layer. Every agent has 17 memory keys plus a cached system prompt. Skills become first-class soul-loader modules: the loader reads the skill at runtime and injects the instructions into the agent's prompt.
- Versioning and cache invalidation line up. Soul-loader caches per agent for 1 hour (TTL from soul_loader.py). When a skill updates, the cache invalidates and every agent matching the skill's description gets the new capability on the next message.
- Skills author once, deploy twice. The same markdown bundle runs in a Claude Code terminal session and in an AGNT venue agent. That's a real workflow compression compared to hardcoding every capability in Python.