Agent Skills › PrimeIntellect-ai/prime-agent › compact

compact

GitHub

用于在上下文即将满时压缩对话历史，释放空间以支持长任务持续进行。通过Python REPL调用compact.run()和status()管理会话状态，确保工作不因上下文限制而中断。

packages/coding-agent/skills/compact/SKILL.md PrimeIntellect-ai/prime-agent

Trigger Scenarios

上下文使用率接近上限需要继续长时间运行的任务

Install

npx skills add PrimeIntellect-ai/prime-agent --skill compact -g -y

More Options

Non-standard path

npx skills add https://github.com/PrimeIntellect-ai/prime-agent/tree/main/packages/coding-agent/skills/compact -g -y

Use without installing

npx skills use PrimeIntellect-ai/prime-agent@compact

指定 Agent (Claude Code)

npx skills add PrimeIntellect-ai/prime-agent --skill compact -a claude-code -g -y

安装 repo 全部 skill

npx skills add PrimeIntellect-ai/prime-agent --all -g -y

预览 repo 内 skill

npx skills add PrimeIntellect-ai/prime-agent --list

SKILL.md

Frontmatter

{
    "name": "compact",
    "description": "Check context usage and compact the conversation from the Python REPL. Use when context is filling up and substantial work remains, so the session is summarized and you keep working instead of stopping early."
}

Compact

Compaction replaces older conversation history with a dense summary, freeing context so long-running work can continue. The implementation lives in the host (the same one behind the user's /compact command); this skill is the kernel-side interface to it. Call it directly from the Python REPL:

await compact.status()
await compact.run()
await compact.run("keep the failing test names and the migration checklist")

API

await compact.status() — current context usage as a dict: tokens, context_window, and percent (None right after a compaction until the next model response), plus scheduled (whether a requested compaction is already pending).
await compact.run(instructions=None) — schedule compaction. Returns {"scheduled": True}, or {"scheduled": False, "reason": ...} when there is nothing to compact yet. Optional instructions focus the summary on what matters for the remaining work.

Rules

Compaction never runs mid-cell. A scheduled compaction runs when the current turn ends; the harness then resumes you automatically with the summary plus recent messages, and you continue the task.
The Python kernel persists through compaction — variables, imports, and helpers you defined all remain available.
Compact at a natural boundary when context usage is high and substantial work remains, instead of becoming terse or returning to the user early. Check await compact.status() when unsure.
One request per turn is enough; calling run again before the turn ends only updates the instructions.