跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
开发与测试

ring:reviewing-code

Reviewing code by dispatching the default reviewer subagents in parallel (plus conditional specialists for lib-observability, lib-systemplane, or lib-streaming when the diff triggers them), then aggregating findings by severity into a report. Use as Gate 8 of ring:running-dev-cycle at epic cadence over the cumulative diff, or before merging. Report-only. Skip for a single-command Go pre-merge verdict (use ring:verifying-code).

LerianStudio · 已有 AI 审核记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add LerianStudio/ring --skill ring:reviewing-code

技能说明


name: ring:reviewing-code description: "Reviewing code by dispatching the default reviewer subagents in parallel (plus conditional specialists for lib-observability, lib-systemplane, or lib-streaming when the diff triggers them), then aggregating findings by severity into a report. Use as Gate 8 of ring:running-dev-cycle at epic cadence over the cumulative diff, or before merging. Report-only. Skip for a single-command Go pre-merge verdict (use ring:verifying-code)."

Code Review (Gate 8)

When to use

  • Gate 8 of development cycle
  • After completing major feature implementation
  • Before merge to main branch
  • After completing complex bug work

Skip when

  • Task is purely conversational or informational with no code changes
  • Changes are limited to documentation or comments with zero logic modifications
  • Code has not been modified since the last completed review cycle

Sequence

Runs after: ring:implementing-tasks Runs before: ring:validating-acceptance-criteria

Related

Complementary: ring:running-dev-cycle, ring:implementing-tasks

Dispatch the 9 default reviewer subagents in parallel, plus any triggered conditional specialists. Dispatch count is dynamic: 9 + triggered specialists, max 12. Do not say or imply all 12 always dispatch.

Announce at start: "Using ring:reviewing-code to dispatch 9 default reviewers plus triggered conditional specialists."

Report-only boundary: This skill does not remediate findings, dispatch implementation work, write comments into source files, generate external artifacts, invoke secondary review tools, or re-run reviewers automatically. It only dispatches the selected reviewers once and reports their findings in the current session.

Default Reviewers (Hard Gate)

#AgentFocus
1ring:code-reviewerArchitecture, design patterns, code quality
2ring:logic-reviewerDomain correctness, business rules, edge cases
3ring:security-reviewerVulnerabilities, authentication, OWASP risks
4ring:test-reviewerTest quality, coverage, edge cases, anti-patterns
5ring:nil-reviewerNil/null pointer safety for Go and TypeScript
6ring:dead-code-reviewerOrphaned code detection, reachability analysis
7ring:perf-reviewerPerformance hotspots, allocations, goroutine leaks, N+1
8ring:tenancy-reviewerMulti-tenant patterns, tenantId propagation, DB isolation
9ring:commons-reviewerlib-commons package usage and reinvented-wheel opportunities

Base hard gate: all 9 default reviewers must PASS.

Conditional Specialist Reviewers

Run these only when the diff matches their trigger. If triggered, include the specialist in aggregation and require PASS for overall PASS.

AgentTrigger
ring:obs-reviewerDiff touches tracing, metrics, logging, runtime recovery/panic safety, redaction, observability constants, or goroutines with recover/SafeGo implications.
ring:systemplane-reviewerDiff touches runtime config, hot-reload knobs, admin config surface, tenant-scoped settings, or systemplane imports/config.
ring:streaming-reviewerDiff touches business events, outbox, event producers, broker publishing, CloudEvents, or event manifests/catalogs.

Role Clarification

WhoResponsibility
This SkillSelect triggered specialists, dispatch reviewers once, aggregate findings, report all severities in-session
Reviewer AgentsAnalyze code, report issues with severity

Step 1: Gather Context (Auto-Detect if Not Provided)

Auto-detect: unit_id (generate if missing), base_sha (git merge-base HEAD main), head_sha (git rev-parse HEAD), implementation_files (git diff --name-only), implementation_summary (git log --oneline).

Display context banner before dispatching.

Step 2: Select Reviewers and Initialize Review State

Start with the 9 default reviewers. Inspect changed files and diff content to decide which conditional specialists are triggered. Track unit_id, base/head SHA, selected reviewer list, reviewer verdicts, and aggregated issues by severity: Critical, High, Medium, Low.

Step 3: Dispatch Selected Reviewers in Parallel

STOP-CHECK BEFORE DISPATCH

Before emitting any Task call, count the reviewers you intend to launch in this turn.

  • Count MUST equal 9 + triggered_specialists.
  • Count MUST be at least 9 and at most 12.
  • If count < 9 or count does not include every triggered specialist -> STOP. Reconcile against the default reviewer table and trigger table above.

MUST NOT trickle-dispatch

All selected reviewers leave in the SAME TURN, before reading any reviewer output.

Forbidden sequences:

  • Dispatch reviewer 1 -> read result -> dispatch reviewer 2
  • Dispatch a subset -> wait -> dispatch the rest
  • Dispatch conditional specialists after partial reviewer output
  • Loop sequentially over the reviewer list

If you find yourself about to dispatch a reviewer in a turn AFTER any reviewer has already returned a result -> STOP. You violated parallel dispatch. Report the violation to the user and mark the gate INCOMPLETE rather than completing the trickle.

Self-verify after dispatch

After the dispatch turn, verify all selected Task calls were emitted in that single turn. If fewer than selected reviewers went out, the gate did NOT execute correctly. Mark the run INCOMPLETE and surface the dispatch failure.

Parallel dispatch — atomic batch

Emit all selected Task calls in a SINGLE TURN, as one atomic batch.

If your runtime exposes a multi_tool_use.parallel wrapper, use it to dispatch the complete selected pool in one wrapped invocation. The STOP-CHECK, anti-trickle, and self-verify guards remain binding regardless of runtime.

Read reviewers/dispatch-prompts.md for the prompt templates. Inject:

  • Epic-level scope header (when scope=epic)
  • base_sha / head_sha from cumulative_dif
阅读完整来源 ↗