Stats
Actions
Tags
From nja
Use when auditing uncommitted changes for architecture compliance — before committing, before handing work back to the user, after generating or implementing a module, or whenever asked to verify/audit/review a diff against the nestjs-neo4jsonapi + nextjs-jsonapi architecture rules.
How this skill is triggered — by the user, by Claude, or both
Slash command
/nja:nja-verifyThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Audit a working tree's uncommitted changes against the architecture rules. This is a read-only review pass: it finds violations and cites them with evidence. It does NOT fix them.
Audit a working tree's uncommitted changes against the architecture rules. This is a read-only review pass: it finds violations and cites them with evidence. It does NOT fix them.
The nja-architecture skill is the ONLY source of truth for rules. Do not cite rules from memory, from CLAUDE.md, or from prior conversation — cite them by skill-doc path + section. If a rule is not in the skill, it is not a rule for this audit.
Audit the changes introduced in the working tree that have not been committed yet. Compute the scope-introduced file list FIRST, before reading any rule, via:
git status --short
Audit ONLY these files. Anything outside this list is pre-existing — flag it separately, do not block on it.
Invoke the nja-architecture skill via the Skill tool. Read the full routing table AND every layer-reference doc relevant to the files in scope. Match each scope file against the routing table; read the reference(s) it points to before judging that file.
Compute the scope file list first (the git status --short above). Pre-existing code outside the list is noted, never blocking.
Do not rationalize. If a scope file matches a pattern listed in references/anti-patterns.md, it is a violation — even if sibling files in the same directory do the same thing. Pre-existing wrongness is not precedent.
Evidence per claim. Every flagged violation cites file_path:line_number + verbatim code + the exact rule it breaks (skill-doc path + line/section). Every "compliant" claim cites the skill-doc's canonical example by file:line it follows. A claim without evidence does not go in the report.
For each violation:
references/anti-patterns.md or a layer doc explicitly forbids it) / DEFENCE-IN-DEPTH / COSMETICfile:line + verbatim codefile:line of the canonical skill example it mirrorsEnd with a summary table:
| Metric | Count |
|---|---|
| Scope-introduced files audited | … |
| BLOCKING violations | … |
| DEFENCE-IN-DEPTH violations | … |
| COSMETIC violations | … |
| Pre-existing violations noted (not blocking) | … |
Do not fix anything during this pass. Audit only. The user decides what to fix and in what order. Proposing the fix code in the report is required; applying it is not — that is a separate, user-authorized step.
| Failure mode | Why it's wrong |
|---|---|
| "The sibling files do this too, so it's fine." | Pre-existing wrongness is not precedent. Each scope file is judged against the skill, not its neighbours. |
| "I know this rule, I'll cite it from memory / CLAUDE.md." | The nja-architecture skill is the only authority. Cite by skill-doc path + section, or don't claim it. |
| "This is close enough to compliant." | Either it matches a canonical example (cite it) or it's a violation. There is no "close enough." |
| "I'll just fix this one while I'm here." | This pass is hands-off. Report the fix; let the user decide. Fixing mid-audit hides what was wrong. |
| "Let me audit the whole repo to be thorough." | Scope is uncommitted changes only. Pre-existing issues are noted separately, never blocking. |
| "No line number handy, the file is enough." | Evidence requires file:line + verbatim code. A claim without it is not reportable. |
npx claudepluginhub carlonicora/nja --plugin njaCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.