cdp-smoke-test
Skill37 starsChanged 20 days agoApache-2.0View on GitHub
Launch a local SLICC dev instance in a controllable browser (CDP) and run smoke tests against the latest build. Three tiers: Tier 1 needs no AI provider (boot, panels, terminal, accounts dialog); Tier 2 exercises the agent loop through a connected provider (chat, shell tool, browser control, scoops, sprinkles, transcript export); Tier 3 pairs a leader with a real follower runtime for tray features (teleport, delegated OAuth, federated CDP, follower rails). Use when asked to start a SLICC dev instance in a browser you control, smoke-test a build, test a leader/follower or tray feature end to end, run an autonomous debugging session against the UI, or send a prompt to a running SLICC and read its assistant reply.
- Installs packages
In this repository9 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- docs/CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- adding-slicc-featuresSkill
- demo-recordingSkill
- deploying-tray-workerSkill
- slicc-handoffSkill
- verifying-before-pushSkill
- writing-slicc-testsSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Very specific: line 85
Use when you need to test a local webapp or worker change before deploying.
- Very specific: line 148
Never use pkill or killall on Chrome, node, or wrangler.
- Distinctive: line 181
Prefer restarting the harness when nothing valuable would be lost.
- Distinctive: line 182
Never print the token into transcripts or scripts.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
CDP Smoke Test
Boot the standalone harness with the latest local build, attach over CDP,
and work through the two test tiers. Manual local-harness driving goes through
scripts/slicc-cdp (zero-dependency, Node 22+).
Setup
Production safety boundary: :5710 (bridge) and :9222 (Chrome CDP)
belong to the developer's production SLICC. Never bind, reap, or kill either
port. The smoke-test lane uses bridge :5715; its Chrome CDP port is
auto-resolved.
Default mode — hosted origin (recommended)
Chrome opens https://www.sliccy.ai with a ?bridge=ws://localhost:5715/cdp
parameter that routes it back to the local node-server. No wrangler needed.
OAuth, IMS (Adobe), and all provider relays work out of the box because they
are hosted on the production origin.
Important: in this mode the browser loads the deployed production webapp from
sliccy.ai, not your local checkout. Changes topackages/webapp/are not exercised. Use this mode for: smoke-testing the node-server bridge, provider login flows, tray and scoops features, and CDP automation. Use the local-wrangler mode below when you need to verify a webapp change end-to-end before deploying.
# 1. Record production listeners before doing anything else.
PROD_BRIDGE_PIDS=$(lsof -nP -tiTCP:5710 -sTCP:LISTEN 2>/dev/null | sort -n | paste -sd, -)
PROD_CDP_PIDS=$(lsof -nP -tiTCP:9222 -sTCP:LISTEN 2>/dev/null | sort -n | paste -sd, -)
printf 'production preflight: :5710=%s :9222=%s\n' "${PROD_BRIDGE_PIDS:-none}" "${PROD_CDP_PIDS:-none}"
# 2. Build only node-server (no webapp build required — UI loads from sliccy.ai).
npm install
npm run build -w @slicc/node-server
# 3. Launch — hosted origin, isolated bridge :5715, ephemeral profile.
# CHROME_PATH is optional; default is a labeled Chrome for Testing clone.
export SLICC_HARNESS_LOG=/tmp/slicc-dev-harness-5715.log
CHROME_PATH="/Applications/Google Chrome Canary.app" \
PORT=5715 \
nohup npm run dev:standalone:fresh > "$SLICC_HARNESS_LOG" 2>&1 &
# 4. Wait for boot and export the auto-resolved CDP port from this lane's log.
sleep 20 && grep "Chrome CDP listening" "$SLICC_HARNESS_LOG"
export SLICC_CDP_PORT=$(sed -nE 's/.*Chrome CDP listening on port ([0-9]+).*/\1/p' "$SLICC_HARNESS_LOG" | tail -1)
test -n "$SLICC_CDP_PORT"
.agents/skills/cdp-smoke-test/scripts/slicc-cdp targets
# 5. Pass criterion: production listeners are exactly the preflight PIDs.
POST_BRIDGE_PIDS=$(lsof -nP -tiTCP:5710 -sTCP:LISTEN 2>/dev/null | sort -n | paste -sd, -)
POST_CDP_PIDS=$(lsof -nP -tiTCP:9222 -sTCP:LISTEN 2>/dev/null | sort -n | paste -sd, -)
test "$POST_BRIDGE_PIDS" = "$PROD_BRIDGE_PIDS" && test "$POST_CDP_PIDS" = "$PROD_CDP_PIDS"The browser URL will be:
https://www.sliccy.ai/?bridge=ws%3A%2F%2Flocalhost%3A5715%2Fcdp&bridgeToken=<token>…bridge= connects the hosted UI back to the local node-server; bridgeToken=
gates the bridge WebSocket. The UI is always served from sliccy.ai — the
local process is only the CDP/API relay.
Alternate mode — local wrangler dev server
Use when you need to test a local webapp or worker change before
deploying. Set WORKER_BASE_URL to a loopback URL; the port is derived from
the URL automatically. OAuth and IMS (Adobe) will not work from localhost
origins — use the hosted-origin mode for any provider login testing.
For explicit non-loopback staging/alternate origins (e.g.
WORKER_BASE_URL=https://slicc-staging.example.workers.dev), the script uses
that URL directly without starting wrangler.
export SLICC_HARNESS_LOG=/tmp/slicc-dev-harness-5715.log
CHROME_PATH="/Applications/Google Chrome Canary.app" \
PORT=5715 WORKER_BASE_URL=http://localhost:8787 \
nohup npm run dev:standalone:fresh > "$SLICC_HARNESS_LOG" 2>&1 &This also starts wrangler on :8787 (or reuses an existing one). A full
build (npm run build -w @ai-ecoverse/cherry -w @slicc/webapp -w
@slicc/node-server) is required beforehand so wrangler has dist/ui/ to
serve.
Attach the console watcher before testing — a clean log at the end is part of the pass criteria. Reset the log first (it appends, so a stale error from a previous run would fail a clean run) and verify the watcher actually attached before starting checks:
rm -f /tmp/slicc-console.log
nohup .agents/skills/cdp-smoke-test/scripts/slicc-cdp watch /tmp/slicc-console.log >/dev/null 2>&1 &
sleep 2 && grep -q 'watcher attached' /tmp/slicc-console.log || echo 'WATCHER FAILED'Tier 1 — no AI provider required
Validates infrastructure: build, harness, bridge, UI shell, kernel. Checks and tier-specific pitfalls: tier1-infrastructure.md.
Tier 2 — AI provider required (chat interaction)
Validates the agent loop: provider connect (with user credential handoff),
streaming, tool use, scoops, sprinkles, and transcript export (cone + scoop
bundle via session export). Checks and tier-specific pitfalls:
tier2-agent-loop.md.
Tier 3 — leader + follower (tray features)
Needed for anything crossing the tray: teleport, delegated OAuth, federated CDP, follower rails. Two SLICC runtimes, not two browser windows — and the follower must both advertise the capability under test and run your build. Follower selection, the origin rule, the port map, and the terminal-driving pitfalls: tier3-multi-harness.md.
Report
Summarize pass/fail per tier, console-watcher findings, harness-log anomalies, and the total session cost from the header counter. For Tier 3 also state which build each runtime ran and which capabilities each follower advertised — a result from a follower running released code, or one that never advertised the capability, is not evidence.
Pitfalls
- Never use
pkillorkillallon Chrome, node, or wrangler. Neverkilla PID resolved from:5710or:9222. If any chosen harness port is occupied, select another isolated port instead of clearing it. - The standalone guard fails fast, exits non-zero, and reports the PID and
command holding the selected bridge port.
SLICC_FRESH_REAP=1opts into reaping only a stale harness on a non-production bridge port that you have verified you own. Forced reaping of:5710is refused; choose another port or stop your own:5710process manually. Chrome CDP is never reaped. Closed the leader tab? The harness survives. Diagnostic-only, do not automate: the harness log redacts
bridgeTokendeliberately — the token is a capability. As a manual last resort during an interactive debugging session, it can be recovered from the ephemeral profile's history (sqlite3 <profile>/Default/History "select url from urls where url like '%bridgeToken%'") and the tab reopened via CDPPage.navigate(see below). Do not use/json/new?URL— Chrome parses everything after the first&as a separate query parameter on the outer/json/newrequest, silently dropping&bridgeToken=...from the loaded URL;parseBridgeLaunchParamsrequires the token and returnsnullwithout it, sosetLocalApiBaseUrlis never set and every/api/fetch-proxycall lands on wrangler (404). UsePage.navigateinstead:node --input-type=module << 'EOF' const ws = new WebSocket('ws://localhost:<CDP_PORT>/devtools/page/<PAGE_ID>'); ws.addEventListener('open', () => { ws.send(JSON.stringify({ id: 1, method: 'Page.navigate', params: { url: '<FULL_SLICC_URL_WITH_BRIDGE_TOKEN>' } })); }); ws.addEventListener('message', (e) => { console.log(e.data); ws.close(); }); EOFPrefer restarting the harness when nothing valuable would be lost. Never print the token into transcripts or scripts.
- Additional instance alongside: choose another unused bridge port, for
example
PORT=5716 WRANGLER_PORT=8787 npm run dev:standalone:fresh. Profiles and Chrome CDP ports auto-isolate; an occupied bridge fails fast.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

