菜单
Diagnostic and debugging workflow for analyzing letitloop task failures, three-strike impossibility artifacts, WAL journal recovery, and state machine transition errors.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/sdageltc/letitloop --skill debugging-failures命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
Comprehensive Autonomous Macro-Task Control Loop & Verification Engine. Plans DAG dependency contracts, executes tasks via pluggable worker adapters (Claude Code, Antigravity CLI, Omniroute, Ollama, direct LLMs), enforces zero-trust deterministic acceptance checks, arbitrates multi-lens quality reviews, and reconciles cryptographic evidence ledgers. Fully provider-agnostic (0-subscription local Ollama, routing gateways, or commercial APIs).
Comprehensive Autonomous Macro-Task Control Loop & Verification Engine. Plans DAG dependency contracts, executes tasks via pluggable worker adapters (Claude Code, Antigravity CLI, Omniroute, Ollama, direct LLMs), enforces zero-trust deterministic acceptance checks, arbitrates multi-lens quality reviews, and reconciles cryptographic evidence ledgers. Fully provider-agnostic (0-subscription local Ollama, routing gateways, or commercial APIs).
Comprehensive code review agent skill for letitloop pull requests and diffs. Enforces deterministic contract contracts, typed exceptions, zero blind except Exception blocks, Bandit SAST security checks, and triple-OS test verification.
| name | debugging-failures |
| description | Diagnostic and debugging workflow for analyzing letitloop task failures, three-strike impossibility artifacts, WAL journal recovery, and state machine transition errors. |
| metadata | {"author":"sdageltc (@sdageltc)","version":"1.0.0"} |
| compatibility | Cross-platform (CLI diagnostic suite) |
When a contract attempt fails or hits a three-strike impossibility escalation in letitloop, use this diagnostic playbook to locate root causes and recover state.
Diagnose Task State:
lil doctor <task_id>
Inspects attempt counts, verification exit codes, error logs, and whether the task is terminal.
View Full Task State:
lil status <task_id>
Dumps the JSON state including history of tried approaches and output hashes.
Inspect Crash Recovery & Checkpoints:
lil checkpoint-recover <goal_id>
Scans the Write-Ahead Log (WAL) journal and reconstructs valid in-memory state.
| Symptom | Root Cause | Fix / Resolution |
|---|---|---|
| Three-Strike Impossibility | Worker repeated non-divergent approaches 3 times without passing verification. | Check scratch/orchestrator_runs/tasks/<task_id>/impossibility_report.json. Adjust verification command or prompt instructions. |
| Out-of-Bounds Scope Violation | Worker wrote files outside workspace_scope.allow. | Check scratch/orchestrator_runs/tasks/<task_id>/scope_violation.log. Add needed directories to contract allow list. |
Budget Exceeded (BudgetExceededError) | Token burn or cost ceiling exceeded BudgetGuard limit. | Increase budget in contract metadata or split the task into smaller subtasks via lil replan. |
| Orphan Lock Contention | Previous crash left an active .lock file. | Check process start tokens via lock.py; lock auto-expires if owner PID is dead. |