backend-audit
- Group: Code quality
- When it activates: Auditing backend and architecture with proof (tools per stack, evidence per finding, report), technical debt, legacy code, circular dependencies and safe refactoring with characterization tests
- Requires:
code-quality - Installed: Always: comes with every project.
Diagnoses the backend like a doctor: tests first, diagnosis second. Nothing gets reported that
cannot be proven. This skill FINDS; the fixes live where they already were:
code-quality (recipes by topic), ddd-hexagonal (structure) and deploy-ops (infrastructure).
Each finding type in the catalog points to its recipe: diagnosis and treatment in harmony.
Minimum reading per task
Section titled “Minimum reading per task”| Task | Read only |
|---|---|
Audit the backend or an area (/audit) |
references/protocolo.md → then whatever it points to |
| Which tools to use in THIS stack | references/herramientas-por-stack.md §the project’s stack |
| What to look for, what evidence each item requires and which recipe fixes it | references/catalogo-hallazgos.md |
| Write the report | references/informe.md (report and finding template) |
Refactor without breaking anything (/refactor) |
references/refactor-seguro.md |
| Architecture rules that check themselves | references/reglas-arquitectura.md |
| Where risk concentrates (code that changes a lot and is complex) | scripts/hotspots.mjs |
Hard rules
Section titled “Hard rules”- No proof, no finding. Every finding carries at least one verifiable piece of evidence: output
from a tool,
file:linewith the snippet, a failing test that reproduces it, or a measurement. “It seems”, “it could” or “it smells like” without evidence go to a separate list of suspicions. - Tools before reading. First run what measures (static analysis, dependency audit, layer rules, coverage, hotspots); then read the flagged code. Reading 200 files “to see what I find” is not an audit.
- Verify each finding before reporting it: open the code, confirm it is not a false positive from the tool and note the confidence level (confirmed or probable).
- The real version rules (
code-quality/references/stack-versions.md): what is fine in Laravel 13 may not exist in 10. Do not report as a defect what the project’s version does not allow. - The project’s conventions are not findings. If
senzu/conventions.mdsets something, following it is not technical debt, even if it is not your preference. - Read-only. The audit does not change code (it only creates the report and, if needed, reproduction
tests on a branch). Fixes go to the plan and are done with
/refactoror/next. - Prioritize by real risk: severity × likelihood × surface (hotspots). A security flaw in a public endpoint comes before 40 long functions in a script nobody touches.
- Tools missing from the project: propose them, but do not install them without asking (they are new dependencies). If they are not installed, state it in the report as a limit of the audit.
Output
Section titled “Output”senzu/auditoria/<YYYY-MM-DD>-<area>.md: executive summary, risk map, findings with evidence, unconfirmed suspicions, limits of the audit and proposed plan.- Cards in
senzu/plan/PLAN.md(skillproject-planner) for whatever the user approves fixing. - Entry in
senzu/devlog/with the summary and the tools that were run.
Relationship with other skills
Section titled “Relationship with other skills”- Fix a finding → the recipe the catalog points to (
code-quality,ddd-hexagonal,deploy-ops). - In-depth security →
code-quality/references/security-owasp.md. - The project’s own style →
/adopt(and its architecture rules,references/reglas-arquitectura.md).