Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.
Quelldateien prüfen
Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Ein direkter Befehl überspringt den Prüf-Prompt. Prüfen Sie die Quelle, bevor Sie ihn ausführen.
Quality — Go testing, linting, coverage, race detection, LSP gates
Each harness specialist agent (moai-harness-hook-ci-specialist,
moai-harness-workflow-specialist, moai-harness-quality-specialist) loads this single
skill for shared context. Supplements general agents (expert-devops, manager-spec,
manager-quality) with project-specific patterns.
27 hook events spanning Session, Tool, Agent, State, Permissions, Interaction, Team,
and Worktree categories. Verify: ls .claude/hooks/moai/handle-*.sh | wc -l == 27. Full
event list in .claude/rules/moai/core/agent-hooks.md.
SPEC Workflow Pipeline
/moai plan "description" → manager-spec → SPEC document with EARS requirements
/moai run SPEC-XXX → manager-develop → Implementation (TDD or DDD)
/moai sync SPEC-XXX → manager-docs → Documentation + PR creation
Quality Targets
go vet ./... and golangci-lint run must be zero errors at all phases.
go test ./... all-pass at all phases; -race must be zero races at run.
Coverage: per-package >= 85% at sync (go test -cover ./...); critical packages
(internal/cli/, internal/template/) >= 90%. LSP gates: run = zero errors / zero type
errors / zero lint errors; sync adds warnings cap (max 10).
Implementation Guide
Section 1 — Hook & CI Patterns
Hook wrapper pattern (thin shell wrapper, Go does the work): read stdin JSON via
INPUT=$(cat), pipe to moai hook {event} <<< "$INPUT". Lives at
.
.claude/hooks/moai/handle-{event}.sh
settings.json invocation rules: always quote $CLAUDE_PROJECT_DIR (e.g.
"$CLAUDE_PROJECT_DIR/.claude/hooks/moai/handle-X.sh"); set timeout (default 5s, max
10s for post-processing); use full path, do not rely on PATH.
Agent-scoped hooks declared in agent YAML frontmatter under hooks: map (per event:
PreToolUse, PostToolUse, SubagentStop, etc.) with matcher regex and command
shell call to handle-agent-hook.sh {action}. Default timeout 5s (10s for post-processing).
GitHub Actions workflows in .github/workflows/: ci.yml (push/PR to main — matrix
ubuntu/macos/windows × Go 1.24, lint + test + build); release.yml (tag v* —
GoReleaser 5 platforms); release-drafter.yml (PR merge to main — auto-label + draft
changelog); auto-merge.yml (Dependabot CI pass — squash); codeql.yml (push/PR — Go
security analysis); spec-lint.yml (PR — SPEC frontmatter validation);
spec-status-auto-sync.yml (schedule — status drift detection); docs-i18n-check.yml
(PR touching docs-site — 4-locale sync verification).
Release process — always use ./scripts/release.sh vX.Y.Z "description" (or
--hotfix); never manual tag push. Chain: script → tag → release.yml → GoReleaser →
5-platform binaries → GitHub Release → Release Drafter updates changelog.
Hook handler testing: Go unit tests in internal/hook/ read JSON from stdin and
exercise the handler logic directly.
Section 2 — Workflow Patterns
SPEC structure under .moai/specs/<SPEC-ID>/: spec.md (EARS + AC), plan.md (M1..Mn
milestones), acceptance.md (binary AC + REQ↔AC traceability, Tier M+), scenarios.md
(Tier L), risks.md (Tier L), progress.md (auto-generated during run). Tier S = 2
artifacts (spec + plan, AC inline); Tier M = 3 (adds acceptance.md); Tier L = 5 (adds
design + research).
EARS requirement patterns: Ubiquitous (The system shall [action]); Event-Driven
(When [event], the system shall [action]); Unwanted (If [bad condition], the system shall [action]); State-Driven (While [state], the system shall [action]); Optional
(Where [feature] enabled, the system shall [action]).
Plan-in-main doctrine: SPEC plan PRs merge to main (not feature branches). Run phase
uses worktree isolation. Sync PRs merge to main with full review history.
Milestones: M1 Foundation → M2 Core → M3 Integration → M4 Edge cases → M5 Polish.
Waves: 30+ task SPECs split into wave-PRs; track in progress.md.
AC format: AC-{SHORT}-{NN}: {verifiable condition} with Verification: command
and Priority: P0|P1|P2.
Critical test isolation rules: (1) always t.TempDir() — auto-cleanup under /tmp;
(2) filepath.Join trap: Join("/a/b", "/var/folders/x") = /a/b/var/folders/x (WRONG)
— use filepath.Abs(); (3) OTEL_* env vars MUST NOT use t.Setenv in parallel tests
(global state race); (4) after any fix, run FULL suite (go test ./...); (5) flaky debug
disables cache via go test -count=1 ./....
Test execution: go test ./... (full), -race (race detection), -cover (coverage),
-run TestX ./pkg/ (specific), -count=1 (disable cache for flaky debug), -v (verbose).
Table-driven test pattern: standard Go convention — tests := []struct{name, input, want string; wantErr bool}{...} + for _, tt := range tests { t.Run(tt.name, ...) }.
Cover happy path + error path + edge cases per row.
LSP quality gate thresholds by phase:
Phase
LSP Errors
Type Errors
Lint Errors
Warnings
plan
Baseline captured
Baseline
Baseline
Baseline
run
Zero required
Zero required
Zero required
—
sync
Zero required
Zero required
Zero required
Max 10
Pre-commit quality gate: go vet ./... && golangci-lint run && go test ./....
Coverage report: go test -coverprofile=coverage.out ./... then
go tool cover -html=coverage.out.
Common quality issues: filepath.Join with absolute path → use filepath.Abs();
OTEL data race in parallel tests → fake/no-op exporter; test writes outside temp dir →
always t.TempDir(); flaky CI → go test -race -count=1.
