How other developers wire Munin into their session corpus. What changed after the first month.
One memory surface, two agents, different directives. How friction reports turn into per-agent behaviour change without cross-contamination. Worked example from a 12-month Rust repo.
Turn a three-page strategy memo into a deterministic kernel. Goals, KPIs, initiatives, constraints, assumptions. Munin joins it with your evidence and emits bounded nudges.
Read guide →
When should a candidate promote to an assertion, and an assertion to a rule? Worked example from a Rust repo with 12 months of session history.
Read guide →
One .munin/ at the root, per-package rule scopes, and shared evidence that flows across workspaces. Walkthrough from a 42-package TypeScript monorepo.
Read guide →
Replay/eval against a held-out corpus before switching agents mid-task. What to monitor, when to fall back, how to read the proof packet.
Read guide →
Three times is a pattern. Turn a repeated correction into a pre-edit hook that blocks the mistake without blocking the flow.
Read guide →
Freshness dropped, confidence tanked, the agent keeps suggesting the old thing. Walk through the decay path and what to do about it.
Read guide →
Open a PR on the Munin repo with a Markdown guide. Keep it under 800 words and include the before/after evidence.