← 返回 Skills 市场
Ocas Praxis
作者
Indigo Karasu
· GitHub ↗
· v2.0.0
· MIT-0
206
总下载
0
收藏
0
当前安装
2
版本数
在 OpenClaw 中安装
/install ocas-praxis
功能描述
Bounded behavioral refinement loop. Records outcomes, extracts micro-lessons from repeated patterns, consolidates them into capped active behavior shifts, ap...
安全使用建议
This skill appears to do what it says: manage a bounded loop of events→lessons→shifts and write auditable journals. Before installing, confirm you are comfortable granting the agent read/write access to ~/openclaw/data/ocas-praxis and ~/openclaw/journals (or set OCAS_ROOT to an isolated directory). Ask the author to update the manifest to declare OCAS_ROOT or required config paths so the required filesystem access is explicit. Review the retention settings (config.json) to avoid persisting sensitive inputs, and ensure any automated handoff (Corvus → intake) comes from trusted sources. If you need stronger assurance, run the skill in an isolated environment or inspect runtime-produced files periodically.
功能分析
Type: OpenClaw Skill
Name: ocas-praxis
Version: 2.0.0
The ocas-praxis skill implements a structured behavioral refinement loop for an AI agent, focusing on recording task outcomes, extracting lessons from patterns, and maintaining a capped set of active behavior shifts. The skill includes explicit safeguards in SKILL.md against identity rewriting, silent safety boundary changes, and unlimited rule accumulation, ensuring all shifts are traceable and auditable via debriefs. No indicators of malicious intent, data exfiltration, or unauthorized execution were found; the self-modification capabilities are bounded and aligned with the stated purpose of performance optimization.
能力评估
Purpose & Capability
The skill's stated purpose (recording events, extracting lessons, proposing/activating behavior shifts, and producing debriefs) matches the commands, data model, and storage layout in the documentation. It legitimately needs read/write access to a dedicated data directory. However, the SKILL.md references OCAS_ROOT and specific data/journal paths while the skill metadata declares no required config paths or env vars — a minor mismatch in the manifest vs runtime docs.
Instruction Scope
Runtime instructions are narrowly scoped: they read intake JSON files from a dedicated intake directory, record events/lessons/shifts into local JSONL files, move processed files to intake/processed, and write per-run journals. There are no network endpoints or instructions to read unrelated system files. This behavior fits the stated purpose.
Install Mechanism
This is an instruction-only skill with no install spec and no code files — lowest-risk install footprint. Nothing is downloaded or written by an installer.
Credentials
The skill declares no required environment variables, but the docs say OCAS_ROOT can override ~/openclaw. The runtime uses filesystem paths under the user's home; that is reasonable for the functionality, but the manifest should explicitly declare OCAS_ROOT (or the required config path) so users know filesystem access will be needed. Also consider that journals and events may contain sensitive context or hashes.
Persistence & Privilege
The skill stores data under its own ~/openclaw/data/ocas-praxis and ~/openclaw/journals/ocas-praxis locations and enforces caps and retention configuration. It does not request always:true, and it does not purport to modify other skills or system-wide configuration. Persistence is scoped to its own directories.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install ocas-praxis - 安装完成后，直接呼叫该 Skill 的名称或使用
/ocas-praxis触发 - 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
v2.0.0
ocas-praxis 1.1.1
- Added journal output capability: introduced `references/journal.md` and `praxis.journal` command for writing a run summary at the end of each run.
- Clarified scope and inter-skill interfaces, specifying cooperation with Corvus and Dispatch, and improved documentation of responsibility boundaries.
- Updated storage layout and config documentation for new journal directory and OCAS_ROOT override.
- Added specific OKRs and clarified output/retention practices for better audit and quality tracking.
v1.1.0
Summary: Introduces a structured, bounded behavioral refinement loop for agent behavior improvement.
- Adds event recording, micro-lesson extraction, and consolidation into capped active behavior shifts.
- Enforces a strict limit (default 12) on active behavior shifts with merging or replacement mechanisms at cap.
- Provides commands for recording, extracting, activating, expiring, and listing behavior shifts.
- Generates plain-language debriefs and concise runtime briefs containing only active shifts.
- Ensures every behavior shift is traceable to specific recorded events for auditability.
- Incorporates hard constraints to prevent unlimited growth, silent personality changes, or rule duplication.
元数据
常见问题
Ocas Praxis 是什么？
Bounded behavioral refinement loop. Records outcomes, extracts micro-lessons from repeated patterns, consolidates them into capped active behavior shifts, ap... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 206 次。
如何安装 Ocas Praxis？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install ocas-praxis」即可一键安装，无需额外配置。
Ocas Praxis 是免费的吗？
是的，Ocas Praxis 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Ocas Praxis 支持哪些平台？
Ocas Praxis 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 Ocas Praxis？
由 Indigo Karasu（@indigokarasu）开发并维护，当前版本 v2.0.0。
推荐 Skills