Muninn is a CLI for finding avoidable cost and friction in coding agent sessions. It turns local session metadata into actionable signals about tooling, instructions, code navigation, verification, and delivery rework.
Better prompts and agent guidance only address one axis of agent performance. Other failure modes come from tool interfaces, repository structure, runtime feedback, test setup, and delivery workflows. Muninn helps distinguish these causes so the improvement targets the system that produced the friction.
Muninn currently ingests Codex sessions. Its normalized analysis is designed to support additional providers without coupling reports to provider-specific formats.
make installThis runs the test suite and installs muninn into $(go env GOPATH)/bin.
muninn analyze --repo .
muninn analyze --repo . --since 24h
muninn analyze --repo . --since 72h --compare-previous
muninn checkpoint before-change --repo .
muninn analyze --repo . --compare before-changeThe default report leads with current findings. Narrow it when investigating a specific kind of friction:
muninn analyze --repo . --focus tooling
muninn analyze --repo . --focus structure
muninn analyze --repo . --focus quality
muninn analyze --repo . --details--compare-previous compares the current rolling lookback with the immediately
preceding window of the same size. The cohorts do not overlap, and the
comparison leads with completed-task duration and outer tool roundtrips.
Muninn does not report prompts, messages, raw commands, raw tool output, absolute paths, secrets, or provider session identifiers.
- Documentation index
- Analysis and reports
- Repository configuration
- Improvement workflow
- Privacy and data model
Run muninn help for the complete command surface.
go test ./...
go build .