Install the SentryDock agent skill
By Blake Folgado · Updated 2026-10-03
The SentryDock skill teaches an agent when to search the index, how to preserve evidence and how to manage an authorised monitor. It is a Markdown instruction file with supporting resources. It complements an MCP connection or CLI; installing the file does not sign in or activate a paid plan.
The public HumanLeap repository contains separate skills for each product. Select sentrydock so your agent receives the right instructions. Read SKILL.md before enabling it in a workspace that handles private research.
The workflow
- Run the skill installation command in the project where the agent works and select your supported client when prompted.
- Open the installed SKILL.md and review the news, monitoring and delivery workflow.
- Connect the remote MCP server using your client guide, or configure the CLI with a key from Settings → Agents.
- Start a new session and ask the agent to read the skill, call account.overview and run one source-linked news search.
pnpm dlx skills add Humanleap/agent-skills --skill sentrydockopenclaw skills install skills-sh:Humanleap/agent-skills/sentrydockWhich clients load the file?
Claude Code reads .claude/skills; Codex reads .agents/skills; Cursor supports .agents/skills and .cursor/skills; Gemini CLI and Kimi Code have their own skill folders and support .agents/skills. Hermes uses ~/.hermes/skills; OpenClaw has a native skills installer. Use the installer selection or the exact folder documented by your client.
Claude chat and Cowork use account skills uploaded through Customize → Skills. A local Claude Code file does not automatically appear there. Put the sentrydock folder, including its root SKILL.md and any resources, in a ZIP and use Upload a skill when that option is available.
What the skill tells the agent to do
- Inspect account access and current tool schemas first.
- Search with a topic and time window; preserve original URLs and publication times.
- Prepare a source-linked note and state gaps in indexed coverage.
- Inspect existing monitors, review sources and cadence, then make only the authorised change.
- Read alerts and verify delivery separately; use signed callbacks for a persistent runtime.
Test that the skill is useful
“Read the SentryDock skill, show my account overview, and search the last 48 hours for copper strikes in Chile. Return the source URLs and dates. Do not change monitoring.”
The useful check is an actual tool result plus a source-linked answer. A skill name in the client list does not show that account authentication or news access worked.
Updates and sources
Public SKILL.md · Repository · Codex skills documentation · Claude account skills.
Review upstream changes before refreshing an installed skill. Keep keys in your client credential store or runtime environment, outside the skill file.