← 返回 Skills 市场
Self Improving Intent Security Agent
作者
Nishant Patil
· GitHub ↗
· v1.0.7
· MIT-0
193
总下载
0
收藏
1
当前安装
8
版本数
在 OpenClaw 中安装
/install self-improving-intent-security-agent
功能描述
Documentation-first skill and workflow toolkit for intent-based security. Provides templates, examples, and local helper scripts for capturing intent, review...
安全使用建议
This package appears to be a documentation-first toolkit and is internally consistent. Before using it: (1) inspect the included shell scripts (scripts/setup.sh, validate-intent.sh, scaffold-run.sh, report.sh) to confirm they only operate on local files and do not invoke network endpoints or run privileged commands in your environment; (2) run scripts in a controlled workspace (e.g., a disposable repo or container) the first time; (3) only provide NPM/C LAWHUB tokens to publishing workflows if you intend to publish the package—publishing credentials are optional and not required for the tool's local use; (4) if you plan to integrate these templates into an autonomous agent, consider restricting autonomous invocation or adding human approval gates for high-risk tasks. Overall the skill is coherent with its stated purpose, but standard caution (review scripts and package.json) is advised.
功能分析
Type: OpenClaw Skill
Name: self-improving-intent-security-agent
Version: 1.0.7
The skill bundle is a documentation-centric toolkit designed to implement 'intent-based security' and auditing workflows for AI agents. It provides a structured framework for capturing user intent, validating actions against constraints, and logging anomalies or learnings into a local `.agent/` directory. Analysis of the included bash scripts (`setup.sh`, `report.sh`, `scaffold-run.sh`, and `validate-intent.sh`) confirms they perform only local file operations and string matching (grep) without any network activity, data exfiltration, or obfuscated execution. The instructions in `SKILL.md` and `CLAUDE.md` are strictly defensive, guiding the agent to follow safety protocols rather than attempting to bypass them.
能力评估
Purpose & Capability
Name/description match the contents: extensive markdown docs, templates, examples, and small helper shell scripts for scaffolding, validating, and reporting. The requested artifacts (intent templates, violation logs, learning records) align with an intent-security documentation/tooling package.
Instruction Scope
SKILL.md limits runtime activity to local file operations (.agent/ tree) and explicitly instructs users to review scripts before running. The documented workflows focus on creating and validating local markdown files, logging, and reporting. There are no instructions to read unrelated system secrets or to exfiltrate data.
Install Mechanism
No install spec is provided (instruction-only skill), which reduces risk. Packaging as an npm project and references to publishing are only documentation for optional publishing; nothing in the runtime install path pulls arbitrary remote binaries or extracts archives.
Credentials
No required environment variables or credentials are declared. The SKILL.md and CLAUDE.md mention optional config env vars for tuning (paths, thresholds) but explicitly state credentials are not required and data remains local, which is proportionate for this skill's functionality.
Persistence & Privilege
Skill flags are default (always: false, user-invocable: true, model invocation allowed). The skill does not request permanent/always-on presence or attempt to modify other skills or system-wide configs; publishing automation references are optional and unrelated to runtime privilege.
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 self-improving-intent-security-agent - 安装完成后，直接呼叫该 Skill 的名称或使用
/self-improving-intent-security-agent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
v1.0.7
- Clarified that this skill is documentation-first and does not include a production runtime enforcement engine.
- Updated description and docs to emphasize templates, examples, and shell scripts for intent validation and audit workflows.
- Added prominent warnings that automation, enforcement, and anomaly detection must be implemented by the host agent or user.
- README and documentation restructured for clearer guidance on using templates and the provided workflow.
- Shell scripts and templates operated locally only; review scripts before use.
v1.0.6
- Added scaffold-run.sh script to automate creation of example/demo conversation workspaces.
- Introduced conversation-driven workflow and documentation, including new guidance in SKILL.md and quick-start guide.
- Added a full customer-feedback-demo example, including conversation transcript, violation, rollback, and learning logs.
- Expanded documentation and usage examples for real-world, audit-friendly agent runs.
- Updated setup steps and quick-start instructions to make it easier to get started and use demos.
v1.0.5
Add comprehensive Clawhub publishing & security learnings to CLAUDE.md: document security scan triggers, before-publishing checklist, scan response patterns, and trust signals for future maintainers
v1.0.4
Fix hooks directory inconsistency: remove non-existent hooks/ from package.json files array, update documentation to clarify hooks are optional and user-created (not included with this skill)
v1.0.3
Fix security scan issues: clarify all environment variables are optional (not required), add security notes documenting no credentials needed and no data transmitted externally, replace [your-org] placeholder with nishantapatil3 in documentation
v1.0.2
Fix install cmd
v1.0.1
Fix metadata inconsistency in SKILL.md
v1.0.0
Initial release of self-improving-intent-security-agent
- Validates every action against user intent before execution, blocking and logging violations.
- Provides automatic rollback of actions if intent is violated, supporting checkpoint creation and manual intervention.
- Monitors for behavioral anomalies, logging and responding to deviations, misuse, or excessive resource consumption.
- Captures structured learning after task completion to continually improve future performance and strategies.
- Maintains full audit trail of intents, violations, anomalies, learnings, and rollbacks for transparency and review.
- Supports human approval for high-risk operations and includes a method for evolving strategies based on A/B testing and observed results.
元数据
常见问题
Self Improving Intent Security Agent 是什么？
Documentation-first skill and workflow toolkit for intent-based security. Provides templates, examples, and local helper scripts for capturing intent, review... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 193 次。
如何安装 Self Improving Intent Security Agent？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install self-improving-intent-security-agent」即可一键安装，无需额外配置。
Self Improving Intent Security Agent 是免费的吗？
是的，Self Improving Intent Security Agent 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Self Improving Intent Security Agent 支持哪些平台？
Self Improving Intent Security Agent 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 Self Improving Intent Security Agent？
由 Nishant Patil（@nishantapatil3）开发并维护，当前版本 v1.0.7。
推荐 Skills