Add compounded — correction-learning plugin with a trust ladder

[Ankii (ankitkr3)]

Adds compounded under Developer Productivity.

What it does: learns from your corrections. When you correct Claude and it does corrective work, a Stop hook captures the lesson, Claude extracts the generalizable rule, and asks you to approve it before saving. Approved rules inject at every session start, and earn autonomy through a trust ladder — verified → trusted → autonomous, with one correction demoting one level.

Per the contribution guidelines:

• Real use case: stop re-explaining the same corrections every session ("when I ask for the latest model, web-search first").
• No duplication: memory plugins (claude-mem etc.) capture and recall; compounded adds the governance layer — approval-gated capture and earned/revocable autonomy. Different problem.
• Quality: 55 unit tests, CI on Linux/macOS across Python 3.9–3.12, Python stdlib only, no daemon, 100% local. MIT.

Install: /plugin marketplace add ankitkr3/compounded-marketplace → /plugin install compounded@compounded-marketplace