Example: Analyzing a codebase

Documented for Akmon 2.0.0.

Scenario

Industry/context (illustrative): regulated fintech service onboarding a new engineer for read-only architecture review.

Constraints

  • Data boundary: no uncontrolled writes during discovery.
  • Approval requirement: keep first pass read-oriented.
  • Audit need: produce artifacts reviewers can verify.

Akmon workflow

cd unfamiliar-project
akmon init
akmon --plan --task "map architecture, auth path, config flow, and top risk modules"
akmon --yes --output json --task "summarize architecture and list highest-risk modules with evidence links" | tee analysis-run.json

Useful prompt patterns

high-level architecture and main data flows

trace the auth path from HTTP request to response

list the riskiest modules if they were wrong

where is configuration loaded and how does it propagate?

find external HTTP calls missing timeouts

generate a Mermaid diagram of major modules

which tests should I run locally before contributing?

Outcome

After the run, you should have:

  • analysis-run.json
  • .akmon/audit/<session-id>.jsonl
  • .akmon/evidence/<session-id>.json

Reviewer question this answers: "What did the analyst inspect, and is the summary traceable to a verifiable session?"

Anti-patterns

  • Using broad write tasks during initial architecture discovery.
  • Reporting "risk areas" without keeping the corresponding session evidence.