Give your agent the AGINDOW skill
Open https://agindow.com/SKILL.md to read or save the best-practices skill. The homepage Copy prompt button copies the complete file, including its skill frontmatter. Paste it into your coding agent with the integration you want it to build. The skill guides setup and safe outcome handling; it does not grant deployment or write-replay permission.
Install as a reusable skill
Download the file and place it in a skill folder supported by your coding agent, for example .agents/skills/agindow-best-practices/SKILL.md for agents that discover project skills there. Other agents use different locations; follow their skill-loading instructions. Review the file before enabling it. Nothing installs automatically.
mkdir -p .agents/skills/agindow-best-practices
curl --fail --show-error --location https://agindow.com/SKILL.md \
--output .agents/skills/agindow-best-practices/SKILL.mdAsk for a concrete integration
Tell your agent the language, tool boundary and evaluation you need. The skill points to versioned SDK references rather than inventing methods. It requires explicit detection configuration and representative baseline/live windows; observe mode alone does not activate detection.
Use the AGINDOW best-practices skill to integrate this API client.
Start in observe mode with an explicit detector and representative windows.
Preserve the existing call behavior, label reads and writes, and verify findings
and bypass handling. Explain the results before proposing a stricter policy.What the skill covers
Node and Python setup, explicit detector selection, call semantics, findings and error handling, safe write idempotency, audit delivery with explicit queue limits, and read-only MCP evidence. The default engine detects nothing and supplies no selective repair output. Installing a skill does not configure detection, change live traffic or enable a hosted service.