菜单
Summarize Mux session folders into a clean markdown report. Use when you need fast context at task start, want to review tool calls, delegated agents, artifacts, or outcomes across one or more sessions.
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
Authoritative user guide + reference map for Mux-Swarm itself (v0.12.1). Use when the user asks how Mux works, how to configure it, what a command/flag/config-key does, how modes/teams/sandbox/auth/ACP/delegation work, or how to troubleshoot Mux. Points to exact sections in the bundled DOCS.md instead of dumping it.
Structured diff and PR code review with a prioritized checklist. Use when reviewing staged changes, pull requests, or any code diff before merge.
Profile, clean, and analyze tabular data (CSV/JSON/Parquet) with pandas/polars. Use when exploring datasets, computing aggregations, joining tables, resampling time series, or producing analysis charts.
Find vulnerable and outdated dependencies across ecosystems (Python, Node, .NET, Rust, Go). Use when auditing a project for CVEs, preparing a release, or investigating a security report.
Design distinctive, polished UI that avoids generic 'AI slop'. Use when making visual design decisions: typography, color, spacing, hierarchy, motion, and component composition.
Run real SAST and secret-scanning tools locally (semgrep, bandit, gitleaks, trivy). Use when auditing a codebase for vulnerabilities, exposed secrets, or vulnerable dependencies before shipping or on demand.
| name | session-reader |
| description | Summarize Mux session folders into a clean markdown report. Use when you need fast context at task start, want to review tool calls, delegated agents, artifacts, or outcomes across one or more sessions. |
| requires_bins | ["uv"] |
Delegate this task to CodeAgent. No memory retrieval is needed — session files are the source of truth. This should be fast and require no more than one agent.
Sessions are saved by both swarm mode and chat (single agent) mode into the same directory. You can identify the mode by the number of agent files present:
*_session.json files, one per agent (e.g. codeagent_session.json, memoryagent_session.json, etc.)agent_session.json file onlyBoth are valid context sources. A chat session with only one file is not incomplete — it reflects a single-agent interaction and should be treated as full context for whatever goal was pursued.
Session directories are named by timestamp (yyyy-MM-dd_HH-mm-ss). Use this to reason about how far back to scan — sort by name descending to find the most recent sessions, or filter by date prefix to find sessions from a specific day or time window.
The summarize script is located at:
{{paths.skills}}/session-reader/scripts/summarize_session.py
uv venv {{paths.base}}/session-reader-venv
# No extra dependencies needed — script uses stdlib only
Identify session folders in {{paths.sessions}}, then pass them along with your allowed paths for artifact detection:
{{shell}} {{shell.flag}} "python {{paths.skills}}/session-reader/scripts/summarize_session.py {{paths.sessions}}/<ts1> {{paths.sessions}}/<ts2> --allowed-paths {{paths.sandbox}}"
Each session will be clearly delimited in the output with a header and footer so the agent can distinguish where one session ends and the next begins.
The script detects artifacts by matching file paths against your configured allowed paths. Pass them via:
--allowed-paths /path/one /path/two — CLI argument (recommended)QWE_ALLOWED_PATHS environment variable — colon-separated on Unix, semicolon-separated on WindowsIf no paths are provided, artifact detection is disabled and a warning is printed.
{{shell}} {{shell.flag}} "python {{paths.skills}}/session-reader/scripts/summarize_session.py {{paths.sessions}}/2026-02-22_23-11-36 {{paths.sessions}}/2026-02-23_08-15-09 {{paths.sessions}}/2026-02-24_14-30-00 --allowed-paths {{paths.sandbox}}"