Agentlie — Development skill for Claude Code
Compare fix claims in Claude Code and Codex logs with recorded edits and heuristic text/AST evidence.
How to install Agentlie
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open SuperMarioYL/agentlie and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Agentlie does
Compare fix claims in Claude Code and Codex logs with recorded edits and heuristic text/AST evidence.
Alternatives in Development
- Investigate Performance — Investigate a Nimbalyst performance problem (freeze, lag, idle CPU, slow op, memory growth) with measured evid 1.6k ★
- Haft — Engineering decisions engine that know when they're stale 1.4k ★
- Cloud Run Debug — Diagnose a failing Cloud Run service — Antigravity (agy/Gemini) digests the error logs cheaply, Claude infers 284 ★
README
[English](README.en.md) | **简体中文**
**从 Claude Code 或 Codex 日志提取修改声明,与记录中的文件变化对照并展示证据。**
`v0.10.0` · `Python 3.10+` · [Apache-2.0](LICENSE)
[Website](https://agentlie.lei6393.com) · [Demo record](docs/demo-results.json)
为什么使用
会话结束时的自然语言总结可能与工具日志不一致。agentlie 把 fix、add、remove、rename、update 声明关联到目标路径及当轮修改,用规则给出 PASS、VAGUE 或 LIE 标签和证据,帮助人工复查。标签不是对 Agent 主观意图的判断。
架构
parser.py 和 codex.py 还原会话及文件前后态,extractor.py 提取声明,verifier.py 比较路径、文本和可用 tree-sitter AST 节点变化,report.py 生成表格或 JSON。默认离线模式不调用模型。
验证逻辑见 [verifier.py](src/agentlie/verifier.py),格式入口见 [cli.py](src/agentlie/cli.py)。日志中的 originalFile 优先,缺失时使用回放重建态;报告会标记 source,不能把重建态当成额外采集到的现场文件。
安装
需要 Python 3.10+。安装会获取依赖;下面显式 --offline,只读取随仓库提供的日志。
git clone https://github.com/SuperMarioYL/agentlie.git
cd agentlie
python3 -m venv .venv
source .venv/bin/activate
python -m pip install -e .
快速开始
实际检查一条人为构造的会话,得到 7 个声明:3 PASS、2 VAGUE、2 LIE。没有执行 Agent、重放 shell 动作或证明代码功能已正确。
python -m agentlie.cli check tests/fixtures/l
Related Skills
Hearmemory
Shared memory for multi-agent coding: subagents and Codex/Claude Code/Cursor hand-offs share one project memor
ReproRun
🔬 A skill that automatically reproduces the numerical claims of academic papers — read paper → find code → bu
Videobug
Fix a screen-recorded bug without touching the terminal. Finds the recording, converts it locally, and fixes f
Fidelity Repair
Claude/Hermes agent skill: repair a rewritten text to match the original's meaning with minimal edits (cuts in
Edit Doctor
Claude Code plugin that explains failed Edits before Claude's next step: the exact whitespace, invisible-chara
Jev Guard
Probability-scored guardrails for Claude Code: deny rule-breaking edits and unasked-for deploys, route your do
Related Agents
GPT Astra High
External worker: gpt-6-astra at high effort, running in the codex CLI. Give it a self-contained brief with goa
Cha Adjudicator
Settles review claims by executing them and returns true / false-with-counter-evidence / unmeasurable. Use whe
Scenario Review
Compare a scenario's recorded assumptions against its calculated results and flag inconsistencies. Use before