feat: reference the synthetic-monitoring-checks agent skill in check authoring surfaces

[markjmeier]

Problem

Authoring synthetic monitoring checks — especially k6 scripted and browser checks — has a steep learning curve. Users starting from the blank script editor have to learn SM's execution model, assertion semantics (a bare check() never fails probe_success), locator strategies, and deployment options on their own. Meanwhile, a battle-tested Agent Skill (synthetic-monitoring-checks in the public grafana/skills repo) teaches coding agents like Claude Code, Cursor, and Codex to do exactly this — choose the simplest sufficient check type, author and locally validate correct scripts, and deploy via UI, API, or Terraform — but nothing in the app tells users it exists.

Solution

References the skill at the moments users need it, framed for each surface:

• Choose-check-type page: a tool picker (Claude Code / Cursor, Codex & other agents). Selecting a tool reveals its one-time install command and copyable starter prompts — "From your site" or "From an API spec" — so a user can go from blank page to a working agent session in two copies.
• Scripted & browser docs side-panels: the full reference block (install commands, the same starter prompts, repo link) next to the script editor, where the "Need help writing scripts?" button already lands.
• Terraform config tab: a collapsed single-line section framed around adoption — "Use your coding agent to create Terraform files for your existing checks" — with a prompt that feeds the page's own export to the agent and requires a no-op terraform plan.
• Grafana Assistant: hidden page context on the choose/scripted/browser/edit routes so Assistant can recommend the skill to users who author with coding agents (Assistant cannot run the skill itself).

Measurement

New agent_skill tracking events (auto-mirrored to Faro, tagged with org_id/stack_id): section_viewed (on render in docs panels; on first reveal elsewhere, so it measures intent rather than traffic), tool_selected, install_command_copied, prompt_copied (with prompt variant), and link_clicked — all carrying a per-surface source property. Copying an install command sets a local marker; on return visits the existing Feedback widget asks "Did the skill help?" to capture qualitative signal from users who actually tried it. Effectiveness funnel: install_command_copied → scripted/browser check_created on the same stack.

The feature ships ungated to measure effectiveness immediately.

🤖 Generated with Claude Code

[markjmeier]

bugbot run

[github-actions]

Script size changes

Name	+/-	Main	This PR	Outcome
[702.js]	=	6,717.24 kB	6,717.24 kB	✅
[87.js]	+0.94%	3,635.48 kB	3,669.56 kB	✅
[469.js]	=	506.85 kB	506.85 kB	✅
[datasource/module.js]	=	119.68 kB	119.68 kB	✅
[9.js]	=	99.10 kB	99.10 kB	✅
[692.js]	=	67.57 kB	67.57 kB	✅
[398.js]	=	48.57 kB	48.57 kB	✅
[module.js]	=	35.04 kB	35.04 kB	✅
[294.js]	=	22.06 kB	22.06 kB	✅
[466.js]	=	5.15 kB	5.15 kB	✅
[131.js]	=	5.00 kB	5.00 kB	✅
[845.js]	=	3.25 kB	3.25 kB	✅
[295.js]	=	2.79 kB	2.79 kB	✅

Totals

Name	+/-	Main	This PR	Outcome
[Scripts]	+0.30%	11,267.77 kB	11,301.85 kB	✅
[Non-script Assets]	=	1,469.61 kB	1,469.61 kB	✅
[All]	+0.27%	12,737.37 kB	12,771.46 kB	✅

Generated by 🚫 dangerJS against a6d546c

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit acb459d. Configure here.}