菜单
Recursive sandboxing for 100+ iterations before human review
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Automatically extract reusable patterns from Claude Code sessions and save them as learned skills for future use.
Instinct-based learning system that observes sessions via hooks, creates atomic instincts with confidence scoring, and evolves them into skills/commands/agents.
Research-before-coding workflow. Search for existing tools, libraries, and patterns before writing custom code. Invokes the researcher agent.
MECE+ methodology for breaking problems into independently-solvable atomic tasks
Simulate-Deploy-Augment loop replacing traditional PDCA
Patterns and utilities for building React components in house-maint-ai
| name | Sandbox Runner |
| description | Recursive sandboxing for 100+ iterations before human review |
Run solutions through recursive sandboxing, iterating up to 100 times before presenting to human.
"Let agents fail 100 times in simulation so humans see only the winning solution."
[Solution] → Sandbox → Test → Pass? → [Present to Human]
↑ │
└── No ──┘ (iterate, max 100x)
sandbox:
max_iterations: 100
early_exit_threshold: 0.95 # Confidence to stop early
checkpoint_every: 10 # Save state every N iterations
tests:
- unit: "npm run test"
- lint: "npm run lint"
- build: "npm run build"
success_criteria:
all_tests_pass: true
no_lint_errors: true
build_success: true
iteration:
number: 42
status: running | passed | failed
failures: ["test A", "test B"]
fixes_attempted: [...]
confidence: 0.72
Stop before 100 iterations if:
sandbox_result:
iterations_run: 47
final_status: passed | failed
solution:
code_changes: [...]
confidence: 0.89
iteration_log:
- iter: 1, status: failed, fixes: [...]
- iter: 2, status: failed, fixes: [...]
- iter: 47, status: passed
ready_for_hitl: true