Agent Skills › EliaAlberti/cpr-compress-preserve-resume › resume

resume

GitHub

用于从项目会话日志中加载上下文，支持按最近次数或主题搜索历史会话。在会话开始时使用以恢复状态、识别阻塞点并规划下一步行动，提升开发连续性。

skills/resume/SKILL.md EliaAlberti/cpr-compress-preserve-resume

Trigger Scenarios

需要回顾之前会话内容时新会话开始需同步上下文时

Install

npx skills add EliaAlberti/cpr-compress-preserve-resume --skill resume -g -y

More Options

Use without installing

npx skills use EliaAlberti/cpr-compress-preserve-resume@resume

指定 Agent (Claude Code)

npx skills add EliaAlberti/cpr-compress-preserve-resume --skill resume -a claude-code -g -y

安装 repo 全部 skill

npx skills add EliaAlberti/cpr-compress-preserve-resume --all -g -y

预览 repo 内 skill

npx skills add EliaAlberti/cpr-compress-preserve-resume --list

SKILL.md

Frontmatter

{
    "name": "resume",
    "description": "Load context from recent CPR session logs, and search past sessions by topic. Use at the start of a session or when you need to know what happened before.",
    "allowed-tools": "Read, Bash(python3 *)",
    "argument-hint": "[N] [topic]",
    "disable-model-invocation": true
}

/cpr:resume

Bring this session up to speed from the project's session logs. The data below was collected by a script, so there is nothing to open or list: read it, then write the report.

Usage: /cpr:resume for the last 3 sessions, /cpr:resume 5 for the last 5, /cpr:resume auth to also search every log for "auth", /cpr:resume 10 migration for both. CLAUDE.md is already in your context; do not re-read it.

Session data

!python3 "${CLAUDE_PLUGIN_ROOT}/scripts/resume_context.py" $ARGUMENTS

Report

Write the report from the data above, in this shape. Keep each bullet to one line, and keep exact values from the logs: paths, identifiers, flags.

══════════════════════════════════════════════
 RESUMING: {project name}
══════════════════════════════════════════════

CONTEXT:
- {what was last worked on}
- {important project state from the logs and CLAUDE.md}

BLOCKERS:
- {from the logs, or "None"}

══════════════════════════════════════════════
 MOST RECENT SESSION: {date time}
 Topic: {topic}
══════════════════════════════════════════════

**Keywords:** {confidence keywords}
**Outcome:** {outcome}

**Key points:**
- {decision, learning or fix}
- {pending task}

══════════════════════════════════════════════
 PREVIOUS SESSIONS ({count})
══════════════════════════════════════════════

- {date}: {topic}, {outcome in a few words}

══════════════════════════════════════════════
 RELATED SESSIONS (topic: "{topic}")      only when a topic was given
══════════════════════════════════════════════

- {date}: {topic}, {why it matched}

══════════════════════════════════════════════
 READY TO:
══════════════════════════════════════════════

- {next step from CLAUDE.md or the latest log}
- {pending task}
══════════════════════════════════════════════

If the data says there are no logs yet, say so in two lines and point to /cpr:compress. If it lists autosaved transcripts newer than the last curated log, mention them once: they hold the full conversation but no summary, and /cpr:resume <topic> searches them.

Read a raw log only when the user asks a question the summaries cannot answer; the summaries exist so that the raw archives never have to be loaded.