Archi Evidence Map

Archi Evidence Map

Use this skill before archi-diff, archi-full, archi-goal, or archi-advice when the right evidence source is unclear.

Read references/architecture-toolbox.md when tool selection, external provenance, language-specific architecture tests, or skill-library patterns matter.

When the user asks what public skills or review workflows are fused into archi, answer from the "Public Skills Carried Or Fused Into Archi" section of that reference and the vendored provenance reference.

Workflow

1. Classify scope:
   • diff review;
   • full-project baseline;
   • goal-driven refactor;
   • tool readiness;
   • review process design.
2. Inventory available evidence without installing anything:
   • current diff and touched files;
   • project docs, module maps, ADRs, and plan-tree notes;
   • language/build/test files;
   • existing CI/static-analysis configuration;
   • generated .architec/ and .hippocampus/ artifacts;
   • dependency or architecture-test configs such as .dependency-cruiser.*, codeql.yml, semgrep.yml, ArchUnit tests, ADRs, module maps, C4 diagrams, service maps, or SLO/runbook evidence.
3. Select the lightest evidence set:
   • direct source review for small diffs or missing tools;
   • project-native tests and type/lint checks for behavior and contract risk;
   • Architec/Hippo for semantic architecture summaries when available;
   • dependency-cruiser or similar dependency graph tools for JS/TS import boundaries and circular dependencies;
   • ArchUnit or similar architecture tests for Java/JVM projects;
   • Semgrep or CodeQL for security, data-flow, and policy evidence;
   • ADRs, decision logs, and git history for decision drift;
   • module maps, domain docs, code owners, and public APIs for boundary review;
   • C4 diagrams, traces, runbooks, and SLOs for distributed-system review;
   • vendored code-review-and-quality for broad code-review coverage;
   • vendored improve-codebase-architecture for deep-module, interface, seam, adapter, leverage, and locality analysis;
   • vendored requesting-code-review or receiving-code-review when the task needs independent review handoff or review-feedback triage;
   • hosted or AI review tools only as advisory external signals.
4. Run or request tools only when the user asked for fresh evidence, the snapshot is stale, or the missing evidence is central to the question.
5. Read tool output as evidence, not authority. Verify actionable findings against source files.
6. Route:
   • active patch risk -> archi-diff;
   • full baseline -> archi-full;
   • specific refactor/objective -> archi-goal;
   • roadmap or sequencing -> archi-advice.
   • dependency direction, cycles, or graph shape -> archi-dependency-topology;
   • module, layer, domain, or DDD leakage -> archi-module-boundaries;
   • executable architecture gates -> archi-fitness-functions;
   • ADRs, assumptions, or decision consistency -> archi-decision-drift;
   • migration, blast radius, or rollback -> archi-change-impact;
   • service/dataflow/reliability tradeoffs -> archi-distributed-systems.

Output

Evidence Plan
- scope:
- direct files:
- existing artifacts:
- optional tools:
- skipped tools and why:

Review Route
- next skill:
- expected output:

Risks
- missing evidence:
- false-positive risk:
- verification needed:

Keep the plan short. Do not block architecture review just because Architec or another optional tool is missing.