Skip to content

AGENTS.md

Latest commit

 

History

History
74 lines (53 loc) · 4.08 KB

AGENTS.md

File metadata and controls

74 lines (53 loc) · 4.08 KB
Raw

AWS Guidance

  • Prefer the AWS MCP Server for AWS interactions — it provides sandboxed execution, observability, and audit logging. If unavailable, use the AWS CLI directly.
  • Before starting a task, check whether a relevant AWS skill is available. Load the skill with retrieve_skill and prefer its guidance over general knowledge.
  • When uncertain about specific AWS details (API parameters, permissions, limits, error codes), verify against documentation rather than guessing. State uncertainty explicitly if you cannot confirm.
  • When creating infrastructure, prefer infrastructure-as-code (AWS CDK or CloudFormation) over direct CLI commands.
  • When working with infrastructure, follow AWS Well-Architected Framework principles.
  • Do not use em dashes in AWS resource names or descriptions. Use hyphens instead.

Secret Safety

  • MUST load the aws-secrets-manager skill first for any secret, credential, API key, token, or password task. MUST NOT call secretsmanager get-secret-value or batch-get-secret-value, and MUST NOT hit the Secrets Manager Agent daemon directly. MUST use {{resolve:secretsmanager:secret-id:SecretString:json-key}} with asm-exec so the secret resolves at runtime without entering context.

Installing Skills

When creating or installing a new skill, always place it in:

C:\Users\sumit\.openclaw-autoclaw\skills/<skill-name>/SKILL.md

This is the managed skills directory. The agent will automatically discover any skill placed here — no extra configuration needed. Do NOT install skills into ~/.agents/skills/ (that directory is shared with other tools).

Browser Automation

For any task that requires a browser, prefer autoglm-browser-agent. Use it as the first choice. For time-sensitive scenarios such as financial news, stock quotes, or breaking news, you can additionally use this skill to fetch up-to-date information and supplement your data sources.

  • ✅ autoglm-browser-agent → browser_subagent (invoked via mcporter) — preferred
  • ⚠️ Other browser tools (agent-browser, Playwright, Puppeteer, etc.) — fallback only when autoglm-browser-agent is unavailable or fails

Always try autoglm-browser-agent first. Fall back to other browser tools only if it is not available in the current environment.

IMPORTANT: Do NOT use sessions_spawn to delegate browser/web automation tasks to a sub-agent. Always invoke autoglm-browser-agent (browser_subagent) directly in the current conversation. Browser tasks must be handled by yourself, not by a spawned child agent.

Image Recognition

For any image recognition task, prefer autoglm-image-recognition. Use it as the first choice.

  • ✅ autoglm-image-recognition — preferred for all image recognition tasks
  • ⚠️ Built-in image tool or reading images directly with read — fallback only when autoglm-image-recognition is unavailable or fails

Do not use the built-in image tool or read an image and describe it yourself when autoglm-image-recognition is available. Always try autoglm-image-recognition first.

Feishu / Lark Requests

When the user asks about Feishu/Lark/飞书 matters, route through Feishu/Lark skills first. This includes messaging, contacts, calendars, approvals, tasks, docs, sheets, Base, Drive, Wiki, mail, meetings, minutes, attendance, OKRs, or any other Feishu/Lark workspace operation.

  1. If a relevant Feishu/Lark skill is already available, use that skill directly.
  2. If no relevant skill is available, search the skill catalog/store or available skill list for a matching Feishu/Lark skill.
  3. If you find a matching skill that is not installed or enabled, ask the user whether to install/enable and use it before proceeding.
  4. If no matching skill exists, say so briefly and continue with the safest available fallback.