← Back to Skills Marketplace
72
Downloads
0
Stars
1
Active Installs
1
Versions
Install in OpenClaw
/install nm-conserve-clear-context
Description
Automatic context management with graceful handoff to a continuation subagent at 80% usage
Usage Guidance
This skill is coherent with its stated goal, but review before use: 1) The SKILL.md references environment variables (CONSERVE_* and CLAUDE_*) although the registry lists none — set and audit those intentionally. 2) The skill writes .claude/session-state.md (or a custom path); ensure your workflows do not cause secrets or credentials to be written to that file or a world-readable location. 3) Beware of 'dangerous' / 'unattended' modes: if CLAUDE_DANGEROUS_MODE or CLAUDE_UNATTENDED is set, continuation agents may proceed without user confirmation — test on non-sensitive projects and prefer interactive mode. 4) Confirm how your platform's Task/TaskUpdate/TaskCreate tools work and what permissions continuation agents receive. 5) If you accept the skill, consider restricting the session state path to a safe location (e.g., project-local, not /tmp or home) and audit generated files after handoffs. If you want greater assurance, ask the maintainer to update registry metadata to declare the env vars the skill actually reads and to add explicit guidance to avoid persisting secrets.
Capability Analysis
Type: OpenClaw Skill
Name: nm-conserve-clear-context
Version: 1.0.0
The skill implements a context-management mechanism that saves session state to `.claude/session-state.md` and delegates tasks to subagents. It contains explicit instructions in `SKILL.md` and `modules/session-state.md` directing the agent to bypass user confirmation ("DO NOT pause for user confirmation") and propagate "dangerous" or "unattended" execution modes across agent boundaries. While these features are aligned with the stated goal of maintaining workflow continuity, the programmatic suppression of human-in-the-loop controls and the use of environment variables like `CLAUDE_DANGEROUS_MODE` to trigger automated execution constitute high-risk behavior.
Capability Assessment
Purpose & Capability
The name/description (automatic context management and handoff) align with the SKILL.md: it saves a session-state file, spawns a continuation subagent, and instructs how to continue. There are no unexpected credential requests or unrelated binaries.
Instruction Scope
Runtime instructions tell the agent to read and write .claude/session-state.md (or a configured path), read existing session state first, consult git/PRs, and spawn continuation agents using the Task tool. The docs also allow the agent to "make reasonable assumptions" and to continue work automatically when execution_mode.auto_continue is true. That grants broad discretion and can lead to unattended actions; the instructions are somewhat open-ended and encourage non-interactive continuation when 'dangerous'/'unattended' mode is detected.
Install Mechanism
Instruction-only skill with no install spec and no code files. Nothing will be downloaded or written to disk by an installer; the only file writes are the session-state files the agent is instructed to create at runtime.
Credentials
The skill's registry metadata declares no required env vars, but SKILL.md references several environment variables (CONSERVE_EMERGENCY_THRESHOLD, CONSERVE_SESSION_STATE_PATH, CLAUDE_DANGEROUS_MODE, CLAUDE_UNATTENDED). That mismatch means required configuration isn't surfaced by the registry. Also, execution depends on CLAUDE_DANGEROUS_MODE / CLAUDE_UNATTENDED which, when set, cause continuation agents to auto-run without confirmation—this is powerful and potentially dangerous. Writing session-state files can also accidentally persist sensitive data if not managed carefully; the doc asks not to include secrets but cannot enforce it.
Persistence & Privilege
The skill does not request permanent installation or 'always: true'. However, it instructs creating and reading a project-scoped state file and delegating work to continuation agents. If continuation agents honor auto_continue/dangerous flags, they may perform unobstructed actions across handoffs — consider that risk when enabling unattended/dangerous modes.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install nm-conserve-clear-context - After installation, invoke the skill by name or use
/nm-conserve-clear-context - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
- Introduces automatic context management that saves session state and delegates continuation to a subagent when usage reaches 80%.
- Provides detailed auto-clear workflow, including session state checkpointing and task list reconciliation to prevent duplication.
- Supports configuration of threshold and session state path through environment variables.
- Detects and propagates execution mode (interactive, unattended, dangerous), ensuring the continuation agent behaves accordingly.
- Optimizes integration with long-running or batch workflows, minimizing user intervention and context loss.
Metadata
Frequently Asked Questions
What is Nm Conserve Clear Context?
Automatic context management with graceful handoff to a continuation subagent at 80% usage. It is an AI Agent Skill for Claude Code / OpenClaw, with 72 downloads so far.
How do I install Nm Conserve Clear Context?
Run "/install nm-conserve-clear-context" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Nm Conserve Clear Context free?
Yes, Nm Conserve Clear Context is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Nm Conserve Clear Context support?
Nm Conserve Clear Context is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Nm Conserve Clear Context?
It is built and maintained by athola (@athola); the current version is v1.0.0.
More Skills