hellblazer/review

review

You MUST call nx_answer for critique/audit/review work. Direct search against RDRs or code skips the decision-evolution traversal and extract → compare pipeline that reviews actually need.

The call

mcp__plugin_nx_nexus__nx_answer(
    question=<caller's phrasing>,
    dimensions={"verb": "review"},
    context=<changed_paths + depth, as JSON string if needed>,
)

One tool call. nx_answer handles match → run → record. Operator chains inside review plans (extract → compare) bundle into a single claude -p subprocess. Plan-miss falls through to an inline claude -p planner; on total miss the tool returns an error string the caller can show to the user.

Required bindings

• changed_paths — list of paths being reviewed (e.g. the git diff file list).

Optional bindings

• depth — traversal depth for decision-evolution walk (default_bindings → 1).

Specialisation hook

For security-reviews, additionally pin dimensions={domain: "security"}; the loader will prefer the strategy:security variant when it exists, falling back to strategy:default otherwise.

Typical intent shapes

• "review this change set for X"
• "did the auth middleware refactor drift from RDR-053?"
• "critique the new taxonomy assignment logic"

When direct search is fine

If you just need to find the RDR or chunk that defines something — e.g. "show me where RDR-053 states the auth middleware contract" — mcp__plugin_nx_nexus__search against rdr__<owner>__voyage-context-3__v1 (or the bare rdr prefix, which expands to all matching collections) is the right tool. This skill earns its cost when the review has to align multiple RDRs against a change set and extract drift claims.

Anti-patterns (do not do any of these)

• Calling search directly when the review requires aligning multiple RDRs against a change. You see current chunks, not the decision-evolution traversal. If the critique needs drift analysis across decision history, you need nx_answer.
• Calling plan_match directly instead of nx_answer. You lose the record step, the inline-planner fallback, and use_count telemetry.
• Running review without changed_paths. The review template has no sensible default — missing changed_paths raises PlanRunBindingError(missing=["changed_paths"]). Surface the error to the user rather than passing empty list.
• Using review for bug triage. That's /nx:debug; the two verbs differ on what drives the walk (review starts from a diff, debug starts from a failing path).

See /nx:plan-first and docs/plan-authoring-guide.md.