← 返回 Skills 市场
waldekmastykarz

atifact

作者 Waldek Mastykarz · GitHub ↗ · v0.7.0 · MIT-0
cross-platform ✓ 安全检测通过
107
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install atifact
功能描述
This skill should be used when the user asks to "extract agent trajectory", "convert HAR to trajectory", "get trajectory from session", "parse agent session"...
使用说明 (SKILL.md)

Extract Agent Trajectories with atifact

Convert agent session recordings (HAR files, Claude Code JSONL logs, Copilot CLI JSONL logs) into structured ATIF v1.6 trajectory JSON using the atifact CLI.

Prerequisites

Verify the CLI is available:

command -v atifact

If not installed, install globally:

npm install -g atifact

Supported input formats

Format File type Description
har .har HAR files with OpenAI (Chat Completions, Responses API) or Anthropic (Messages API) requests
claude-code-jsonl .jsonl Claude Code CLI session logs
copilot-cli-jsonl .jsonl Copilot CLI session logs

Format is auto-detected from file contents (not extension). Use --format / -f to force a specific format when auto-detection fails.

Usage

Basic conversion

The --output / -o option takes a prefix, not a filename. Output files are derived from the prefix:

  • Main trajectory: \x3Cprefix>.trajectory.json
  • Subagent trajectories: \x3Cprefix>.trajectory.\x3Cname>.json

Default prefix is the input file path:

atifact session.har
# Writes: session.har.trajectory.json

Specify output prefix

atifact session.har -o out
# Writes: out.trajectory.json
# If subagents exist: out.trajectory.\x3Cname>.json

Force input format

Use when auto-detection picks the wrong format for .jsonl files:

atifact session.jsonl -f claude-code-jsonl
atifact session.jsonl -f copilot-cli-jsonl

Pipe JSON to stdout

Use --json with --quiet to suppress diagnostics and get clean JSON on stdout. Output is a JSON array of all trajectories (main first, then subagents). No files are written.

atifact session.har --json --quiet

Combine with other tools:

atifact session.har --json --quiet | jq '.[0].steps | length'

CLI options

Option Alias Description
\x3Cinput-file> Path to the input file (required)
--output -o Output path prefix. Main: \x3Cprefix>.trajectory.json, subagents: \x3Cprefix>.trajectory.\x3Cname>.json (default: input file path)
--format -f Force input format: har, claude-code-jsonl, copilot-cli-jsonl
--json Write JSON array of all trajectories to stdout (no files written). First element is main, rest are subagents.
--quiet -q Suppress progress messages (stderr only)

Exit codes

Code Meaning
0 Success
1 Runtime error (parse failure, I/O error)
2 Invalid usage (bad arguments, missing file)

Workflow

  1. Identify the input file and its format (HAR or JSONL).
  2. For .jsonl files, determine the source (Claude Code or Copilot CLI) to use the correct --format if auto-detection fails.
  3. Run atifact with the input file. Use -o to set the output prefix (e.g., atifact /path/to/session.har -o /path/to/session). The main trajectory is written to \x3Cprefix>.trajectory.json.
  4. Report the output file path(s) and key metrics (total steps, total cost) from the generated trajectory.

Notes

  • HAR files may contain multiple API formats (OpenAI + Anthropic); all are parsed.
  • Multi-turn HAR conversations are deduplicated (each request carries full history).
  • Utility calls (e.g., gpt-4o-mini title generation) are excluded from the trajectory.
  • Tool results from request N are attached as observations to the agent step from request N-1.
  • Copilot CLI logs with subagent task tool calls produce separate trajectory files per subagent. The main trajectory references them via subagent_trajectory_ref with trajectory_path pointing to the sibling file.
  • All timestamps are preserved from source data as-is (ISO 8601).
安全使用建议
This skill appears safe for its stated purpose if you intentionally choose the session log to convert and trust the external `atifact` npm CLI. Be cautious with HAR and JSONL logs because they can contain private prompts, tool outputs, request metadata, or other sensitive session details, and the generated trajectory files may preserve that information.
功能分析
Type: OpenClaw Skill Name: atifact Version: 0.7.0 The skill bundle provides instructions for using the 'atifact' CLI tool to convert agent session logs (HAR, JSONL) into ATIF trajectory JSON. It includes standard installation steps via npm and detailed usage documentation for the CLI. No malicious patterns, data exfiltration, or prompt injection attempts were identified in SKILL.md or _meta.json.
能力评估
Purpose & Capability
The stated purpose and instructions align: converting HAR, Claude Code JSONL, and Copilot CLI JSONL logs into ATIF trajectory JSON. These inputs may contain sensitive conversation or API-session content, so users should choose files carefully.
Instruction Scope
The workflow is user-directed and scoped to a specified input file and output prefix; there are no instructions to run automatically, recurse through directories, or ignore user intent.
Install Mechanism
The skill instructs users to install the external `atifact` CLI globally with npm if missing. That setup is purpose-aligned, but the package code is not included in the reviewed artifacts.
Credentials
Running a local CLI over selected HAR or JSONL session files is proportionate to the conversion purpose, but those files and generated outputs may include private prompts, tool traces, request metadata, or timestamps.
Persistence & Privilege
The skill writes trajectory JSON files such as `<prefix>.trajectory.json` and possible subagent files. This is disclosed and expected, with no evidence of background persistence or privilege escalation.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install atifact
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /atifact 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v0.7.0
- Adds support for converting HAR files, Claude Code JSONL logs, and Copilot CLI JSONL logs to ATIF v1.6 trajectory JSON using the atifact CLI. - Introduces auto-detection of input file format, with options to force CLAUDE or Copilot format if needed. - Allows specifying an output file prefix; handles writing main and subagent trajectories with clear file naming. - Provides options to output a combined JSON array to stdout and suppress diagnostics for streamlined scripting. - Documents complete CLI usage, options, workflow, and important notes about deduplication, tool calls, and timestamp preservation.
元数据
Slug atifact
版本 0.7.0
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 1
常见问题

atifact 是什么?

This skill should be used when the user asks to "extract agent trajectory", "convert HAR to trajectory", "get trajectory from session", "parse agent session"... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 107 次。

如何安装 atifact?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install atifact」即可一键安装,无需额外配置。

atifact 是免费的吗?

是的,atifact 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

atifact 支持哪些平台?

atifact 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 atifact?

由 Waldek Mastykarz(@waldekmastykarz)开发并维护,当前版本 v0.7.0。

💬 留言讨论