What problem does it solve?
It reduces the risk that the second-brain plugin silently fails or drifts from its stated behavior by forcing you to read the real code and verify the real on-disk outputs through an adversarial, evidence-backed workflow.
Core Features & Use Cases
- Layered architectural skepticism: Doubts specific plugin subsystems (hooks, stop-hook extraction, learning, MCP server, compaction, wiki maintenance) using a taxonomy mapped to concrete files.
- History-aware question selection: Rotates focus across runs using git change recency and doubt-history rotation to avoid repeatedly asking the same weak questions.
- Runtime artifact verification: Checks that expected files and state actually exist on disk, not just that code would write them.
- Conversational branch drilling with citations: Generates multiple attack vectors per claim, escalates the highest-scoring branch first, and requires file:line grounding for pruned paths.
- Quality-critic validation gate: Sends findings to a dedicated reviewer agent to confirm issues and catch systematic over-pruning.
Quick Start
Run doubt with no arguments to automatically select the most relevant plugin layers and adversarial perspectives, then produce a doubt session report with code citations and runtime-state checks.