Run an evidence-grounded software architecture audit workflow that builds a repo brief, selects single-auditor or specialist-panel mode, inspects boundary, layering, dependency, composition, cohesion, and testability risks, writes required finding blocks, and sequences incremental refactors. Use when asked for an architecture audit, architecture review, repo-structure review, software architecture report, audit_report.md, structural issue findings, or specialist-panel synthesis across multi-module systems.
100
100%
Does it follow best practices?
Impact
100%
1.85xAverage score across 3 eval scenarios
Passed
No known issues
Run an evidence-grounded software architecture audit workflow that turns the highest-leverage findings into an ordered, incremental improvement sequence.
This skill reviews a repository, subsystem, or specific path for architecture problems grounded in direct code evidence. It builds a factual repo brief, selects single-auditor or specialist-panel mode, inspects boundary integrity, dependency direction, cohesion, composition roots, testability, architecture smells, report structure, and refactor sequencing.
The workflow starts with a factual repo brief, chooses single-auditor or specialist-panel mode based on scope, gathers direct evidence, and produces a structured report with category-specific findings and a concrete improvement sequence.
Use this skill for prompts such as:
tessl install sharaf/software-architecture-audit| Layer | Where | Purpose |
|---|---|---|
| Entry point | SKILL.md | Full audit workflow, guardrails, decision logic, and output requirements |
| Tile manifest | tile.json | Registry metadata and skill entrypoint |
The final audit report must use these headings, in order:
## Repo Brief
## Highest-Leverage Structural Issues
## Boundary and Layering Problems
## Dependency and Composition Problems
## Testability and Change Friction
## Improvement Sequence
## Open Evidence GapsEach full finding includes severity, evidence, why it matters, recommended improvement, and likely refactor surface. The improvement sequence is ordered by leverage and dependency, not abstract architectural purity.
Tested across three generated scenarios:
| Scenario | What it checks |
|---|---|
scenario-0 | Required report section structure and finding format |
scenario-1 | Single-auditor mode selection and repo brief completeness |
scenario-2 | Specialist-panel synthesis, deduplication, and forbidden recommendations |
| Model | Baseline avg | With-context avg | Uplift |
|---|---|---|---|
claude-haiku-4-5 | 50% | 99% | 1.98x |
claude-sonnet-4-6 | 52% | 99% | 1.90x |
claude-opus-4-6 | 50% | 99% | 1.98x |
Per-scenario with-context scores:
| Model | scenario-0 | scenario-1 | scenario-2 |
|---|---|---|---|
claude-haiku-4-5 | 100% | 100% | 96% |
claude-sonnet-4-6 | 100% | 98% | 100% |
claude-opus-4-6 | 100% | 100% | 98% |
Activation after the routing fix: 3/3 generated scenarios fired tessl__software-architecture-audit.
Migrated from the legacy flat skill skills/software-architecture-audit.md in this repository.
Research lineage:
software-architectureanti-patterns-code-smells, clean-layered-architecture, dependency-management-inversion, structural-patterns, testing-architecture, system-evolution-migration2026-04-01