👷 add dev-server CLI + manual-testing skill

[BenoitZugmeyer]

Motivation

Manual testing of SDK features is currently hard to reproduce and share with reviewers.
When writing PR test instructions, there's no structured way to verify what events were
actually sent to intake. This PR introduces a daemonized dev-server CLI that makes
manual testing steps reproducible and self-documenting. It also ships a /manual-testing
Claude Code skill that automates the full workflow: starts the server, creates a sandbox
page, drives it with playwright-cli, inspects intake, and outputs ready-to-paste PR
test instructions.

Changes

• Replace the single scripts/dev-server.ts with a daemonized CLI (yarn dev-server <command>)
• start: spawns the server as a detached background process, writes PID/URL to .dev-server/state
• stop: kills the daemon and clears state
• status: shows whether the server is running, with URL, PID, and file paths
• logs: streams the server log file, with --follow/-f support and ANSI stripping when piped
• intake: reads captured intake requests from .dev-server/intake-requests.json, with selectors
  (rum-views, rum-errors, rum-actions, logs-events, replay-records, etc.), human-readable
  output by default (TTY) and NDJSON when piped (--json to override)
• Intake proxy logs requests to a write stream and prints jq query examples in --help
• Add .claude/skills/manual-testing.md — a Claude Code skill (/manual-testing) that
  orchestrates the full end-to-end test flow using the dev-server CLI and playwright-cli,
  then emits self-contained bash test instructions ready to paste into the PR

Test instructions

You can generate these instructions automatically by running /manual-testing in Claude Code.

yarn dev-server start         # starts the server, prints URL and log path
yarn dev-server status        # confirms it's running
yarn dev-server logs -f       # follow logs in another terminal

# open the sandbox in a browser, interact with it, then:
yarn dev-server intake rum-views | jq .view.loading_time
yarn dev-server intake rum-errors | jq .error.message
yarn dev-server intake        # human-readable output in terminal

yarn dev-server stop
yarn dev-server status        # confirms it stopped

Checklist

• Tested locally
• Tested on staging
• Added unit tests for this change.
• Added e2e/integration tests for this change.
• Updated documentation and/or relevant AGENTS.md file

[cit-pr-commenter-54b7da]

Bundles Sizes Evolution

📦 Bundle Name	Base Size	Local Size	𝚫	𝚫%	Status
Rum	N/A	175.02 KiB	N/A	N/A	N/A
Rum Profiler	6.16 KiB	6.16 KiB	0 B	0.00%	✅
Rum Recorder	27.46 KiB	27.46 KiB	0 B	0.00%	✅
Logs	56.80 KiB	56.80 KiB	0 B	0.00%	✅
Rum Slim	130.66 KiB	130.66 KiB	0 B	0.00%	✅
Worker	23.63 KiB	23.63 KiB	0 B	0.00%	✅

🚀 CPU Performance

Action Name	Base CPU Time (ms)	Local CPU Time (ms)	𝚫%
RUM - add global context	0.004	0.008	+100.00%
RUM - add action	0.0131	0.0204	+55.73%
RUM - add error	0.0127	0.0234	+84.25%
RUM - add timing	0.003	0.0052	+73.33%
RUM - start view	0.0121	0.0208	+71.90%
RUM - start/stop session replay recording	0.0006	0.0008	+33.33%
Logs - log message	0.0145	0.0178	+22.76%

🧠 Memory Performance

Action Name	Base Memory Consumption	Local Memory Consumption	𝚫
RUM - add global context	27.21 KiB	26.87 KiB	-347 B
RUM - add action	51.76 KiB	52.10 KiB	+350 B
RUM - add timing	26.28 KiB	26.13 KiB	-152 B
RUM - add error	56.24 KiB	55.26 KiB	-1006 B
RUM - start/stop session replay recording	25.86 KiB	25.61 KiB	-256 B
RUM - start view	461.46 KiB	461.16 KiB	-310 B
Logs - log message	43.98 KiB	44.76 KiB	+803 B

🔗 RealWorld

[datadog-prod-us1-6]

✅ Tests

🎉 All green!

❄️ No new flaky tests detected
🧪 All tests passed

🎯 Code Coverage (details)
• Patch Coverage: 100.00%
• Overall Coverage: 77.23%

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: f166f10 | Docs | Datadog PR Page | Was this helpful? React with 👍/👎 or give us feedback!}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 2157b7d2ed

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Verify daemon identity before sending kill signal

stop sends process.kill(state.pid) after only checking that the PID exists; isRunning() in state.ts is just process.kill(pid, 0). If the dev-server process exited and that PID was reused, yarn dev-server stop can kill an unrelated process on the host instead of the daemon.

Useful? React with 👍 / 👎.

[BenoitZugmeyer]

note: I'm not too happy with this, but I'm not sure where this intakeProxyMiddleware should stay, whether in the test/e2e or the scripts folder. I chose to keep it there for now, we might have more shared code in the future, we can always revisit.