Menú
Dispatch independent adversarial reviews for ZFC, ZFC leveling, root-cause-first, and thermo without duplicating their standards.
Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.
| name | code-standards |
| description | Dispatch independent adversarial reviews for ZFC, ZFC leveling, root-cause-first, and thermo without duplicating their standards. |
Use this skill to dispatch code, diff, PR, or proposed-implementation review against the repo's core standards. This is a pointer/dispatch skill: reference the source skills below instead of duplicating their standards.
NO REVIEW LANE PASSES WITHOUT VERIFICATION EVIDENCE A lane returning PASS must include the specific file/line evidence that proves compliance. A lane returning FAIL must include the exact location and required fix. Rationalizations are not evidence.
This Iron Law mirrors the superpowers systematic-debugging principle: if you can't point to the exact line, the lane hasn't completed its work.
Load the relevant source skills by path and treat them as authoritative.
Path resolution: Prefer personal overrides from ~/.claude/skills/
when they exist, then fall back to the repo-local project skills listed below.
The repo-local .claude/skills/ paths are authoritative for a normal checkout
because they are versioned with this repository. The .codex/skills/ directory
contains pointer/mirror files for Codex-based agents and is not the primary
lookup path.
|- .claude/skills/zero-framework-cognition/SKILL.md (/zfc)
|- .claude/skills/zfc-leveling-roadmap/SKILL.md (/zfclevel)
|- .claude/skills/root-cause-first/SKILL.md (/root-cause-first)
|- .claude/skills/code-quality/SKILL.md (/code-quality, alias /cq) — metric-driven complexity / duplication / coupling / AI-import verification. Out of scope: syntax, naming, formatting, lint rules (those are linter territory, see "Lane ownership" below).
|- Thermo-nuclear: invoke via Agent tool with subagent_type: thermo-nuclear-code-quality-review (/thermo). This is an Agent tool subagent type, not a bash command or external process.
Each lane has a defined scope. A lane must hand off — not duplicate — work that belongs to a sibling lane.
| Concern | Owner | Why |
|---|---|---|
| Syntax, formatting, import order, lint rules | ESLint / Ruff / mypy / prettier / pre-commit hooks | Mechanical, deterministic, already enforced in CI |
| Naming conventions, file layout | /code-standards style lane + linters | Project conventions, not metric-driven |
| Logic, architecture, complexity, duplication, AI-import verification | /code-quality | Metric + pattern reasoning |
| ZFC contract, agent routing, structured fields | /zfc | Domain semantics |
| Level-up / XP / reward semantics | /zfclevel | Game-loop semantics |
| Root-cause-first methodology for fix proposals | /root-cause-first | Debugging discipline |
| Holistic adversarial review | /thermo | Independent subagent model |
If a finding would belong to another lane, the current lane must name the correct lane and stop rather than produce feedback that conflicts with it.
SUPERPOWERS INTEGRATION (from obra superpowers framework):
/zfc, /zfclevel, /root-cause-first, /code-quality, and /thermo.
Each lane has a defined scope — see "Lane ownership" below./thermo lane uses subagent_type: thermo-nuclear-code-quality-review
(NOT a bash command). Pass: diff summary, file list, working dir.Each lane follows the four-phase process from systematic-debugging:
Applied to code review:
For any PR touching production code (non-test files):
Behavioral prompt prose belongs in $PROJECT_ROOT/prompts/ markdown files, not in
Python string literals. Python may assemble deterministic structured context,
load prompt files, and substitute explicit placeholders such as current time or
location, but it must not embed new model instructions, behavioral rules, or
audit criteria inline. If a review finds inline prompt prose added or expanded
in Python, mark /code-standards FAIL unless the PR moves that prose into a
prompt file or documents why the text is purely deterministic formatting.
Return a concise report:
## Code Standards Report
Scope: <file, diff, PR, or task>
- ZFC: PASS/WARN/FAIL - <evidence>
- ZFC leveling: PASS/WARN/FAIL - <evidence or permitted N/A caveat>
- Root-cause-first: PASS/WARN/FAIL - <evidence or permitted N/A caveat>
- Code-quality: PASS/WARN/FAIL - <evidence or permitted N/A caveat (see "Short-circuit clauses" in code-quality/SKILL.md)>
- Thermo: PASS/WARN/FAIL - <evidence or structural blockers>
Blockers:
- <line-level issue and required fix>
Test Coverage:
- <test command run and output, or CI evidence reference>
- PASS/FAIL for test coverage gate
Next checks:
- <tests, evidence, or review steps needed>
Iron Law Enforcement:
From systematic-debugging rationalization tables:
| Rationalization | Reality |
|---|---|
| "Tests are obvious, no need to verify" | Test existence ≠ test correctness |
| "Code is simple, review is optional" | Simple code has simple bugs with simple root causes |
| "I'll verify later" | Later = never; verification must be in this session |
| "Close enough to pass" | A FAIL is a FAIL; reconcile by fixing, not diluting |
If you catch yourself thinking any of these, stop and complete the verification step.
Token-efficient second opinion slash command /advice. Extracts decision point + artifact (≤150 lines), then fans out in parallel: (1) Opus subagent reviewer with fallback chain codex→agy→cursor, (2) /research on the decision topic, (3) /secondo multi-model opinion. Use instead of advisor() which ships the full conversation uncached.
Lazy senior-dev mode — the seven-rung ladder that decides whether to write code at all, whether to reuse code that already exists, and whether to prefer stdlib/platform/installed deps over a new dependency. Use before writing any code, every PR diff, every fix, every feature. Source attribution included.
Use when making any "X is enabled" / "feature X works" / "cache works" claim in a running session — requires runtime probe output from the standard harness startup path, NOT a launchd/cron test that explicitly sets the activation env var.
Pre-flight checklist for self-hosted runner failures. Use when investigating low disk, missing runner, or stuck Green Gate on the jleechanorg/worldarchitect.ai fleet. Always verifies host-level container state, not just GitHub API.
When asked to verify whether a Claude Code feature works (especially slash commands, dialogs, pickers, status indicators), spawn a real interactive TUI session in cmux — never use `claude --print "/feature"` as a test, because --print is non-interactive and will always return "isn't available in this environment" regardless of whether the feature actually works.
Read ~/.hermes/agent-orchestrator.yaml BEFORE declaring an --agent <X> value unsupported. Lists plugin shorthands (wafer, minimax, agy).