[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M75W5TPQ2MSD5G8NECEQZWS4":3,"skill-same:01M75W5TPQ2MSD5G8NECEQZWS4":46},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":26,"commitSha":27,"skillDirectory":28,"forksCount":14,"repositoryHtmlUrl":29,"repositoryHomepageUrl":6,"ownerHtmlUrl":30,"ownerBlogUrl":6,"ownerDescription":31,"updatedAt":32,"install":33,"resources":45},"01M75W5TPQ2MSD5G8NECEQZWS4","expect-test",null,"AI-powered browser testing using expect-cli. Auto-detects dev server, reads git diff, generates and executes browser tests via Playwright. Use when you need to verify UI changes in a real browser, test user flows, or validate fixes visually. Trigger on \"test this in the browser\", \"verify the UI\", \"run expect\", \"browser test\", or after completing UI fixes.","stevengonsalvez\u002Fainb-toolkit","stevengonsalvez","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F9320602?v=4","01M75W59QRYZYANWY3XG8CJB1P","Steven Gonsalvez","01M75W59RD5YW1RT67BDV0EAZF",1,"MIT","https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Fpackages\u002Fqe-agent-pack\u002F.apm\u002Fskills\u002Fexpect-test\u002FSKILL.md",[],[],"2026-11-12 13:55:43",false,"---\nname: expect-test\ndescription: AI-powered browser testing using expect-cli. Auto-detects dev server, reads git diff, generates and executes browser tests via Playwright. Use when you need to verify UI changes in a real browser, test user flows, or validate fixes visually. Trigger on \"test this in the browser\", \"verify the UI\", \"run expect\", \"browser test\", or after completing UI fixes.\nargument-hint: \"[-m message] [--headed] [-y]\"\nuser-invocable: true\n---\n\n# Expect Test — AI Browser Testing\n\nRun AI-generated browser tests against your dev server using [expect-cli](https:\u002F\u002Fgithub.com\u002Fmillionco\u002Fexpect). No test files to write — the AI reads your git diff, generates a test plan, and executes it in a real Playwright-driven browser.\n\n## When to Use\n\n- After making UI changes — verify they work visually\n- After fixing bugs — confirm the fix in a real browser\n- To test user flows (login, navigation, forms)\n- As part of `\u002Fvalidate` to add browser verification\n- When you need visual confirmation, not just build passes\n\n## Prerequisites\n\nRequires Claude Code (already running if you're reading this).\n\n## Setup & Installation\n\n```bash\n# Check if expect-cli is installed\nif ! command -v expect-cli &>\u002Fdev\u002Fnull; then\n  echo \"Installing expect-cli...\"\n  npm install -g expect-cli@latest\nfi\n\n# Verify\nexpect-cli --version\n```\n\n### First-run Playwright browser install (gotcha)\n\nexpect-cli ships with its OWN pinned `playwright-core` version. The system-wide `npx playwright install` installs whatever's LATEST, which expect-cli won't find. First-run symptom:\n\n```\nFailed to launch browser: ... Executable doesn't exist at\n...\u002Fchromium_headless_shell-1217\u002F...\n```\n\n**Fix** — install via expect-cli's bundled playwright:\n\n```bash\n# Headed mode (uses chromium)\nnode $(npm root -g)\u002Fexpect-cli\u002Fnode_modules\u002Fplaywright-core\u002Fcli.js install chromium\n\n# Headless mode (uses chromium-headless-shell)\nnode $(npm root -g)\u002Fexpect-cli\u002Fnode_modules\u002Fplaywright-core\u002Fcli.js install chromium-headless-shell\n```\n\nThat downloads the exact version expect-cli pins. **Without this, the TUI's LLM-driven planner will detect the missing binary and wander into setup\u002Finstall tangents that burn 2-3 minutes of tokens on `npx playwright install` + `ls node_modules` + project introspection BEFORE running any actual browser step.**\n\n## Usage\n\n### Auto-detect changes and test\n\n```bash\n# Test unstaged changes (default) — auto-detects dev server\nexpect-cli\n\n# Test with a specific instruction\nexpect-cli -m \"test the login flow works correctly\"\n\n# Test a whole branch diff\nexpect-cli -t branch\n\n# Skip plan review (CI mode)\nexpect-cli -y\n\n# Show browser window during test\nexpect-cli --headed\n\n# Verbose logging\nexpect-cli --verbose\n```\n\n### Targeted testing\n\n```bash\n# Test specific feature\nexpect-cli -m \"verify the Future\u002FPast toggle on events calendar switches views correctly\"\n\n# Test a fix\nexpect-cli -m \"confirm the modal has a solid dark background, not transparent\"\n\n# Test a flow\nexpect-cli -m \"sign in as coach, navigate to perform, check events show only future events\"\n```\n\n## Running in tmux (for agents)\n\n**Always run in tmux** — expect-cli `tui` has an interactive TUI that blocks the terminal.\n\n### CLI syntax changed — use the `tui` subcommand\n\nRecent expect-cli versions (≥0.1.2) moved the test runner under the `tui` subcommand.\n`expect-cli -m \"...\"` without `tui` will print `error: unknown option '-m'`.\n\n```bash\n# ✅ CORRECT — tui subcommand\nexpect-cli tui -m \"short description\" --browser-mode headed -y\n\n# ❌ WRONG on v0.1.2+\nexpect-cli -m \"short description\" --headed -y\n```\n\n### TUI is deprecated — prefer the MCP server going forward\n\nOn first run, expect-cli prints a banner:\n> ⚠ The TUI is deprecated. Use \u002Fexpect from your coding agent instead. Run `npx expect-cli@latest init` to set up.\n\nFor new projects, install the MCP server (`expect-cli init`) so Claude Code can call\nthe atomic tools (`expect open`, `expect screenshot`, `expect network_requests`, etc.)\ndirectly. The `tui` command still works but will be removed in a future release.\n\n### Safe pattern for long missions (the ONLY way that's robust)\n\nPut the mission in a file via heredoc. Inline `-m \"...\"` strings with apostrophes,\nembedded double-quotes, or newlines break the shell chain: `tmux send-keys` →\n`bash -c` → expect-cli argv parser. Symptom: `error: too many arguments for 'tui'. Expected 0 arguments but got 4`.\n\n```bash\nSTAMP=$(date +%s)\nSESSION=\"expect-${STAMP}\"\nLOG=\"\u002Ftmp\u002Fexpect-${STAMP}.log\"\nMISSION_FILE=\"\u002Ftmp\u002Fmission-${STAMP}.txt\"\nURL=\"http:\u002F\u002Flocalhost:5173\"   # or a deployed preview\n\n# 1) Write the mission to a file — heredoc preserves apostrophes\u002Fquotes\u002Fnewlines.\ncat > \"$MISSION_FILE\" \u003C\u003C'EOF'\nNavigate to the page. Wait for network idle.\nVerify the SHOTs tab is selected and 5+ article cards are visible.\nApostrophes (\"don't\"), quotes, and newlines all survive when the mission is a file.\nEOF\n\n# 2) Launch tmux + expect-cli tui — reference the file with $(cat …), NOT inline.\ntmux new-session -d -s \"$SESSION\" \\\n  \"EXPECT_REPLAY_OUTPUT_PATH=\u002Ftmp\u002Fexpect-replays \\\n   expect-cli tui -u '$URL' -m \\\"\\$(cat $MISSION_FILE)\\\" --browser-mode headed -y --verbose \\\n   2>&1 | tee $LOG\"\n\n# 3) First-run prompt: expect-cli asks \"Install the expect skill for your coding agents? (Y\u002Fn)\".\n#    The `-y` flag does NOT cover this prompt. Decline it:\nsleep 5\ntmux send-keys -t \"$SESSION\" \"n\" Enter\n\n# 4) Monitor. DO NOT attach — it blocks the agent.\ntmux capture-pane -t \"$SESSION\" -p | tail -30\n```\n\n### Why the naive inline approach fails\n\n```bash\n# ❌ BROKEN — the double-quote nesting inside bash -c collapses\ntmux send-keys -t \"$S\" \"bash -c 'expect-cli tui -m \\\"$LONG_MISSION\\\" ...'\"\n# With apostrophes and newlines in $LONG_MISSION, bash sees 4+ positional args.\n```\n\n### Quick-check pattern (no AI, just raw browser)\n\nFor smoke checks where you don't need the AI agent:\n\n```bash\nexpect-cli open \"$URL\" --headed --wait-until networkidle\nexpect-cli screenshot             # saves to \u002Ftmp\u002Fexpect-artifacts\u002F\nexpect-cli console_logs\nexpect-cli network_requests --url supabase\nexpect-cli close\n```\n\nThese atomic commands don't need tmux (they return immediately), don't require a mission\nstring, and don't block on the install prompt.\n\n### Scripted-walkthrough pattern (RECOMMENDED for agent flows)\n\n**This is the most reliable agent mode** — fully deterministic, no LLM planning step, no token burn on irrelevant tangents. Drive the whole flow via `expect-cli playwright \u003Ccode>` atomic commands.\n\n```bash\n# 1) Open browser to starting URL\nexpect-cli open http:\u002F\u002Flocalhost:5160\u002Flogin --browser chromium --headed --wait-until domcontentloaded\n\n# 2) Drive interactions via raw playwright code\nexpect-cli playwright \"await page.fill('input[type=email]', 'coach1@shot.com'); \\\n  await page.fill('input[type=password]', 'th3SHOT123'); \\\n  await page.click('button[type=submit]'); \\\n  await page.waitForURL(\u002F\\\u002Fhome|\\\u002Fperform\u002F, {timeout: 15000})\" \\\n  --description \"Login coach1\"\n\n# 3) Navigate + assert via page.evaluate (must use page.evaluate — NOT bare document.x)\nexpect-cli playwright \"await page.goto('http:\u002F\u002Flocalhost:5160\u002Fperform?tab=team', {waitUntil: 'domcontentloaded'}); \\\n  await page.waitForTimeout(2500); \\\n  return await page.evaluate(() => ({ \\\n    title: document.title, \\\n    bodyText: (document.body.innerText || '').slice(0, 200), \\\n    playerCount: (document.body.innerText.match(\u002FPlayer \u002Fgi) || []).length \\\n  }));\"\n\n# 4) Capture screenshot\nexpect-cli screenshot --full-page\n\n# 5) Inspect what happened\nexpect-cli console_logs\nexpect-cli network_requests --url supabase\n\n# 6) Close at end\nexpect-cli close\n```\n\n**Why this beats the TUI mission flow:**\n\n| Concern | TUI mission flow | Scripted atomic flow |\n|---|---|---|\n| Token cost | High (LLM plans + executes + reports) | Low (deterministic shell) |\n| Reproducibility | Variable (LLM may classify failures differently) | Exact (same script same result) |\n| Time-to-first-action | 2-5 min (planner reads SKILL.md, greps source) | \u003C10s (instant) |\n| Failure mode clarity | Opaque (`category=app-bug` with no detail) | Direct (you control the assertion) |\n| Browser install drama | LLM may try `npx playwright install` mid-test | Solved upfront (see prerequisite) |\n\n**Critical gotchas in `expect-cli playwright \u003Ccode>`:**\n\n1. **MUST use `page.evaluate(() => {...})`** to access DOM — bare `document.x` outside `evaluate` errors with `document is not defined` (the code runs in Node, not the page).\n2. **Return value must be JSON-serialisable** — pass arrays\u002Fobjects, not DOM nodes or Promises.\n3. **For text matching, use `document.body.innerText`** then regex against the string. Selector-based existence checks via `document.querySelector` work too.\n4. **Anchor-or-button discovery**: many SHOT UI elements are `\u003Cbutton>` not `\u003Ca>` (per perform-redesign walk 2026-05-19). Probe for BOTH before assuming click target shape:\n   ```js\n   const anchors = Array.from(document.querySelectorAll('a[href]')).filter(a => a.offsetParent !== null);\n   const buttons = Array.from(document.querySelectorAll('button')).filter(b => b.offsetParent !== null);\n   ```\n5. **`networkidle` is flaky against apps with realtime sockets** (Supabase, Firebase live queries) — they keep the network busy indefinitely. Use `domcontentloaded` + `waitForTimeout(2500)` or wait for a specific element instead.\n\n## CLI Reference\n\n| Flag | Description |\n|------|-------------|\n| `-m, --message \u003Ctext>` | Natural language test instruction |\n| `-f, --flow \u003Cslug>` | Reuse a previously saved test flow |\n| `-y, --yes` | Skip plan review, run immediately |\n| `-a, --agent \u003Cprovider>` | AI provider: `claude` or `codex` (default: claude) |\n| `-t, --target \u003Ctarget>` | What to test: `unstaged`, `branch`, or `changes` |\n| `--headed` | Show visible browser window |\n| `--verbose` | Enable verbose logging |\n| `--replay-host \u003Curl>` | Override replay viewer host |\n\n| Subcommand | Description |\n|------------|-------------|\n| `expect-cli init` | First-time global install + agent skill setup |\n| `expect-cli audit` | Run lint\u002Ftype\u002Fformat checks and suggest browser tests |\n\n## How It Works\n\n1. **Auto-detects dev server** — scans localhost ports via `lsof`, recognizes Vite\u002FNext\u002FReact\u002FAngular\n2. **Reads git diff** — scopes test plan to what actually changed\n3. **AI generates test plan** — step-by-step browser actions + assertions\n4. **Playwright executes** — real browser, real clicks, real screenshots\n5. **Reports pass\u002Ffail** — with session recording for replay\n\n## Replay Recordings & Evidence\n\nExpect-cli automatically saves session recordings to disk:\n\n```bash\n# Default location\nls \u002Ftmp\u002Fexpect-replays\u002F\n\n# Override with env var\nEXPECT_REPLAY_OUTPUT_PATH=\u002Fpath\u002Fto\u002Fdir expect-cli -m \"test X\" -y\n```\n\n**Artifacts saved per session:**\n\n| File | Format | Usage |\n|------|--------|-------|\n| `*.html` | HTML replay viewer | Open in browser to replay the full test session |\n| `*.ndjson` | rrweb event recording | Raw event data for programmatic analysis |\n| `*.webm` | Playwright video | Actual browser screen recording |\n| `*-steps.json` | Step metadata | Pass\u002Ffail per step with timestamps |\n\n**To collect evidence after a test run:**\n```bash\n# Find latest replay\nLATEST=$(ls -t \u002Ftmp\u002Fexpect-replays\u002F*.html 2>\u002Fdev\u002Fnull | head -1)\necho \"Replay: $LATEST\"\n\n# Open replay in browser\nopen \"$LATEST\"\n\n# Copy video for PR evidence\ncp \u002Ftmp\u002Fexpect-replays\u002F*.webm .\u002Ftest-evidence\u002F\n```\n\n**Note:** Screenshots are NOT saved to disk (base64 in AI context only). Use the `.webm` video for evidence extraction (see below).\n\n## Publishing Interactive Replay (Recommended)\n\nThe HTML replay is the best evidence format — reviewers can scrub through the entire session interactively. Publish via `\u002Fhere-now`:\n\n```bash\nSESSION_ID=$(ls -t \u002Ftmp\u002Fexpect-replays\u002F*.html | head -1 | xargs basename .html)\nmkdir -p \u002Ftmp\u002Freplay\ncp \u002Ftmp\u002Fexpect-replays\u002F${SESSION_ID}.html \u002Ftmp\u002Freplay\u002F\ncp \u002Ftmp\u002Fexpect-replays\u002F${SESSION_ID}.ndjson.js \u002Ftmp\u002Freplay\u002F\ncp \u002Ftmp\u002Fexpect-replays\u002F${SESSION_ID}.ndjson \u002Ftmp\u002Freplay\u002F\n\n# Publish (3 files: HTML shell + ndjson.js recording + ndjson raw data)\n\u002F.claude\u002Fskills\u002Fhere-now\u002Fscripts\u002Fpublish.sh \u002Ftmp\u002Freplay\u002F --client claude\n\n# The replay URL is: https:\u002F\u002F{slug}.here.now\u002F{SESSION_ID}.html\n# Post to PR comment with the direct link\n```\n\n**Important**: The HTML file loads the `.ndjson.js` by deriving the filename from its own name. All 3 files must be published together, and the HTML must be accessed by its original filename (not as `index.html`).\n\n**Expiry**: Anonymous publishes expire in 24h. Include the claim URL in the PR comment so the user can make it permanent.\n\n## Extracting Screenshots from Recordings\n\nSince expect-cli doesn't save per-step screenshots to disk, extract key frames from the `.webm` video using ffmpeg. This produces shareable PNG evidence for PR comments.\n\n```bash\nWEBM=$(ls -t \u002Ftmp\u002Fexpect-replays\u002F*.webm | head -1)\nmkdir -p \u002Ftmp\u002Fscreenshots\n\n# Extract frames at key timestamps (adjust based on test step timing from logs)\n# Timing guide: step 1 finish ≈ step1_duration, step 2 ≈ step1 + step2, etc.\nffmpeg -i \"$WEBM\" -ss 25 -frames:v 1 \u002Ftmp\u002Fscreenshots\u002F01-login.png -y 2>\u002Fdev\u002Fnull\nffmpeg -i \"$WEBM\" -ss 50 -frames:v 1 \u002Ftmp\u002Fscreenshots\u002F02-home.png -y 2>\u002Fdev\u002Fnull\nffmpeg -i \"$WEBM\" -ss 80 -frames:v 1 \u002Ftmp\u002Fscreenshots\u002F03-feature.png -y 2>\u002Fdev\u002Fnull\n\n# Validate screenshots show real content (not splash\u002Fblank)\nfor f in \u002Ftmp\u002Fscreenshots\u002F*.png; do\n  dims=$(magick identify -format \"%wx%h\" \"$f\" 2>\u002Fdev\u002Fnull)\n  echo \"$(basename $f): $dims ($(wc -c \u003C \"$f\") bytes)\"\ndone\n```\n\n**Timestamp estimation**: Sum step durations from the test log. If step 1 took 25s and step 2 took 17s, the step 2 completion frame is at ~42s.\n\n## Uploading Evidence for PR Comments\n\n### Option 1: imgbb (preferred for small batches)\n\n```bash\nIMGBB_API_KEY=\"${IMGBB_API_KEY:-006dfde8d5037a1e366db2bb24e915d3}\"\nfor file in \u002Ftmp\u002Fscreenshots\u002F*.png; do\n  sleep 3  # Rate limit\n  url=$(curl -s -X POST \"https:\u002F\u002Fapi.imgbb.com\u002F1\u002Fupload?key=${IMGBB_API_KEY}\" \\\n    --form \"image=@${file}\" | jq -r '.data.url \u002F\u002F empty')\n  echo \"$(basename $file) → $url\"\ndone\n```\n\n### Option 2: GitHub Draft Release (when imgbb rate-limits)\n\n```bash\ngh release create pr${PR}-evidence --repo OWNER\u002FREPO \\\n  --title \"PR #${PR} Evidence\" --notes \"Browser test screenshots\" \\\n  --draft \u002Ftmp\u002Fscreenshots\u002F*.png\n\n# Get download URLs\ngh release view pr${PR}-evidence --repo OWNER\u002FREPO --json assets \\\n  --jq '.assets[] | \"\\(.name) → \\(.url)\"'\n```\n\n### Posting to PR\n\n```bash\nBASE=\"https:\u002F\u002Fgithub.com\u002FOWNER\u002FREPO\u002Freleases\u002Fdownload\u002FTAG\"\ngh pr comment $PR --repo OWNER\u002FREPO --body \"$(cat \u003C\u003CEOF\n## ✅ Browser Test Evidence\n\n![login](${BASE}\u002F01-login.png)\n![feature](${BASE}\u002F02-feature.png)\n\n\u003Cdetails>\u003Csummary>All screenshots\u003C\u002Fsummary>\n\n![step3](${BASE}\u002F03-step.png)\n\n\u003C\u002Fdetails>\nEOF\n)\"\n```\n\n## Full Evidence Workflow (end-to-end)\n\n```bash\n# 1. Run test\nSESSION=\"expect-$(date +%s)\"\ntmux new-session -d -s \"$SESSION\"\ntmux send-keys -t \"$SESSION\" \"EXPECT_REPLAY_OUTPUT_PATH=\u002Ftmp\u002Fexpect-replays expect-cli -m 'test description' --headed -y --verbose 2>&1 | tee \u002Ftmp\u002Fexpect.log\" C-m\n\n# 2. Wait for completion\nsleep 120 && tmux capture-pane -t \"$SESSION\" -p | tail -20\n\n# 3. Extract frames\nWEBM=$(ls -t \u002Ftmp\u002Fexpect-replays\u002F*.webm | head -1)\nffmpeg -i \"$WEBM\" -ss 30 -frames:v 1 \u002Ftmp\u002Fscreenshots\u002Fstep1.png -y 2>\u002Fdev\u002Fnull\nffmpeg -i \"$WEBM\" -ss 60 -frames:v 1 \u002Ftmp\u002Fscreenshots\u002Fstep2.png -y 2>\u002Fdev\u002Fnull\n\n# 4. Validate (read with Claude to verify content)\n# Use Read tool on each PNG to verify it shows the expected UI state\n\n# 5a. Publish interactive replay (PREFERRED)\nSESSION_ID=$(ls -t \u002Ftmp\u002Fexpect-replays\u002F*.html | head -1 | xargs basename .html)\nmkdir -p \u002Ftmp\u002Freplay\ncp \u002Ftmp\u002Fexpect-replays\u002F${SESSION_ID}.{html,ndjson.js,ndjson} \u002Ftmp\u002Freplay\u002F\n\u002F.claude\u002Fskills\u002Fhere-now\u002Fscripts\u002Fpublish.sh \u002Ftmp\u002Freplay\u002F --client claude\n# Post the replay URL to PR\n\n# 5b. Upload static screenshots (FALLBACK if replay not suitable)\ngh release create prN-evidence --draft \u002Ftmp\u002Fscreenshots\u002F*.png\ngh pr comment N --body \"## Evidence\\n![step1](url)\\n![step2](url)\"\n```\n\n## Integration with \u002Fvalidate\n\nWhen `\u002Fvalidate` is invoked, this skill adds browser verification as Step 2.5:\n\n```\nStep 2.5: Browser Verification (via expect-test)\n- Auto-detect running dev server\n- Generate browser tests from the implementation plan\n- Execute tests and capture screenshots\n- Report visual verification results alongside code validation\n```\n\n## Result Interpretation\n\n```\n✔ = Test passed\n✘ = Test failed (with description of what went wrong)\n```\n\nResults include:\n- Pass\u002Ffail status per test step\n- Screenshots at each step\n- Session recording URL for replay\n- Specific failure descriptions (e.g., \"element not found\", \"wrong color value\")\n\n### Failure classification taxonomy (TUI mode)\n\nThe TUI agent attaches a `category=...; domain=...` label to each ✘ failure. Decoded:\n\n| Category | Meaning | Examples | Agent action |\n|---|---|---|---|\n| `app-bug` | Real application behavior mismatch | Element missing, wrong nav target, broken render | File a bead; this is the test's job |\n| `env-issue` | Infrastructure \u002F setup \u002F auth problem | Dev server down, login failed, network 5xx, fixture data absent | Fix the env, don't blame the app |\n| `timeout` | Assertion timeout | Element never appeared, page never settled | Often = data not loading; check the env first |\n\nWhen you see `env-issue` consecutively across early steps (login + landing), it almost always means **the test infrastructure isn't ready** (dev server pointing at wrong DB, missing test users, missing test data). Fix the env before chasing `app-bug` reports.\n\n**The scripted atomic-flow pattern (above) sidesteps this entirely** — you control the assertions so failure classification is just your shell exit code + the returned object.\n\n## Limitations\n\n- Early stage tool (v0.0.x) — may crash on complex scenarios\n- macOS\u002FLinux only (uses `lsof` for port detection)\n- Requires a running dev server on localhost\n- No custom Playwright config — browser is fully managed\n- Interactive TUI — must use `-y` flag or run in tmux for automation\n- **Headed mode steals window focus** — when the user has Chrome open for other work, prefer `--browser-mode headless` (TUI) or `expect-cli open … --browser chromium` without `--headed` (atomic). Only use `--headed` when the user explicitly wants to watch.\n- **Per-step screenshots NOT saved to disk by default** — `expect-cli screenshot` saves under `\u002Ftmp\u002Fexpect-artifacts\u002Fscreenshot-{n}.png`, but mid-`expect-cli playwright` steps don't auto-capture. Use the `.webm` video + ffmpeg, OR call `expect-cli screenshot` explicitly between steps in scripted flows.\n- **TUI mission planner can wander 2-5 min on first run** — bundled Playwright browser install + project introspection. Pre-install via the Prerequisites section's command, or skip the TUI entirely and use the scripted atomic-flow pattern.\n\n## Environment\n\n| Variable | Description |\n|----------|-------------|\n| `NO_TELEMETRY=1` | Disable analytics\u002Ftelemetry |\n","# Expect Test — AI Browser Testing\n\nRun AI-generated browser tests against your dev server using [expect-cli](https:\u002F\u002Fgithub.com\u002Fmillionco\u002Fexpect). No test files to write — the AI reads your git diff, generates a test plan, and executes it in a real Playwright-driven browser.\n\n## When to Use\n\n- After making UI changes — verify they work visually\n- After fixing bugs — confirm the fix in a real browser\n- To test user flows (login, navigation, forms)\n- As part of `\u002Fvalidate` to add browser verification\n- When you need visual confirmation, not just build passes\n\n## Prerequisites\n\nRequires Claude Code (already running if you're reading this).\n\n## Setup & Installation\n\n```bash\n# Check if expect-cli is installed\nif ! command -v expect-cli &>\u002Fdev\u002Fnull; then\n  echo \"Installing expect-cli...\"\n  npm install -g expect-cli@latest\nfi\n\n# Verify\nexpect-cli --version\n```\n\n### First-run Playwright browser install (gotcha)\n\nexpect-cli ships with its OWN pinned `playwright-core` version. The system-wide `npx playwright install` installs whatever's LATEST, which expect-cli won't find. First-run symptom:\n\n```\nFailed to launch browser: ... Executable doesn't exist at\n...\u002Fchromium_headless_shell-1217\u002F...\n```\n\n**Fix** — install via expect-cli's bundled playwright:\n\n```bash\n# Headed mode (uses chromium)\nnode $(npm root -g)\u002Fexpect-cli\u002Fnode_modules\u002Fplaywright-core\u002Fcli.js install chromium\n\n# Headless mode (uses chromium-headless-shell)\nnode $(npm root -g)\u002Fexpect-cli\u002Fnode_modules\u002Fplaywright-core\u002Fcli.js install chromium-headless-shell\n```\n\nThat downloads the exact version expect-cli pins. **Without this, the TUI's LLM-driven planner will detect the missing binary and wander into setup\u002Finstall tangents that burn 2-3 minutes of tokens on `npx playwright install` + `ls node_modules` + project introspection BEFORE running any actual browser step.**\n\n## Usage\n\n### Auto-detect changes and test\n\n```bash\n# Test unstaged changes (default) — auto-detects dev server\nexpect-cli\n\n# Test with a specific instruction\nexpect-cli -m \"test the login flow works correctly\"\n\n# Test a whole branch diff\nexpect-cli -t branch\n\n# Skip plan review (CI mode)\nexpect-cli -y\n\n# Show browser window during test\nexpect-cli --headed\n\n# Verbose logging\nexpect-cli --verbose\n```\n\n### Targeted testing\n\n```bash\n# Test specific feature\nexpect-cli -m \"verify the Future\u002FPast toggle on events calendar switches views correctly\"\n\n# Test a fix\nexpect-cli -m \"confirm the modal has a solid dark background, not transparent\"\n\n# Test a flow\nexpect-cli -m \"sign in as coach, navigate to perform, check events show only future events\"\n```\n\n## Running in tmux (for agents)\n\n**Always run in tmux** — expect-cli `tui` has an interactive TUI that blocks the terminal.\n\n### CLI syntax changed — use the `tui` subcommand\n\nRecent expect-cli versions (≥0.1.2) moved the test runner under the `tui` subcommand.\n`expect-cli -m \"...\"` without `tui` will print `error: unknown option '-m'`.\n\n```bash\n# ✅ CORRECT — tui subcommand\nexpect-cli tui -m \"short description\" --browser-mode headed -y\n\n# ❌ WRONG on v0.1.2+\nexpect-cli -m \"short description\" --headed -y\n```\n\n### TUI is deprecated — prefer the MCP server going forward\n\nOn first run, expect-cli prints a banner:\n> ⚠ The TUI is deprecated. Use \u002Fexpect from your coding agent instead. Run `npx expect-cli@latest init` to set up.\n\nFor new projects, install the MCP server (`expect-cli init`) so Claude Code can call\nthe atomic tools (`expect open`, `expect screenshot`, `expect network_requests`, etc.)\ndirectly. The `tui` command still works but will be removed in a future release.\n\n### Safe pattern for long missions (the ONLY way that's robust)\n\nPut the mission in a file via heredoc. Inline `-m \"...\"` strings with apostrophes,\nembedded double-quotes, or newlines break the shell chain: `tmux send-keys` →\n`bash -c` → expect-cli argv parser. Symptom: `error: too many arguments for 'tui'. Expected 0 arguments but got 4`.\n\n```bash\nSTAMP=$(date +%s)\nSESSION=\"expect-${STAMP}\"\nLOG=\"\u002Ftmp\u002Fexpect-${STAMP}.log\"\nMISSION_FILE=\"\u002Ftmp\u002Fmission-${STAMP}.txt\"\nURL=\"http:\u002F\u002Flocalhost:5173\"   # or a deployed preview\n\n# 1) Write the mission to a file — heredoc preserves apostrophes\u002Fquotes\u002Fnewlines.\ncat > \"$MISSION_FILE\" \u003C\u003C'EOF'\nNavigate to the page. Wait for network idle.\nVerify the SHOTs tab is selected and 5+ article cards are visible.\nApostrophes (\"don't\"), quotes, and newlines all survive when the mission is a file.\nEOF\n\n# 2) Launch tmux + expect-cli tui — reference the file with $(cat …), NOT inline.\ntmux new-session -d -s \"$SESSION\" \\\n  \"EXPECT_REPLAY_OUTPUT_PATH=\u002Ftmp\u002Fexpect-replays \\\n   expect-cli tui -u '$URL' -m \\\"\\$(cat $MISSION_FILE)\\\" --browser-mode headed -y --verbose \\\n   2>&1 | tee $LOG\"\n\n# 3) First-run prompt: expect-cli asks \"Install the expect skill for your coding agents? (Y\u002Fn)\".\n#    The `-y` flag does NOT cover this prompt. Decline it:\nsleep 5\ntmux send-keys -t \"$SESSION\" \"n\" Enter\n\n# 4) Monitor. DO NOT attach — it blocks the agent.\ntmux capture-pane -t \"$SESSION\" -p | tail -30\n```\n\n### Why the naive inline approach fails\n\n```bash\n# ❌ BROKEN — the double-quote nesting inside bash -c collapses\ntmux send-keys -t \"$S\" \"bash -c 'expect-cli tui -m \\\"$LONG_MISSION\\\" ...'\"\n# With apostrophes and newlines in $LONG_MISSION, bash sees 4+ positional args.\n```\n\n### Quick-check pattern (no AI, just raw browser)\n\nFor smoke checks where you don't need the AI agent:\n\n```bash\nexpect-cli open \"$URL\" --headed --wait-until networkidle\nexpect-cli screenshot             # saves to \u002Ftmp\u002Fexpect-artifacts\u002F\nexpect-cli console_logs\nexpect-cli network_requests --url supabase\nexpect-cli close\n```\n\nThese atomic commands don't need tmux (they return immediately), don't require a mission\nstring, and don't block on the install prompt.\n\n### Scripted-walkthrough pattern (RECOMMENDED for agent flows)\n\n**This is the most reliable agent mode** — fully deterministic, no LLM planning step, no token burn on irrelevant tangents. Drive the whole flow via `expect-cli playwright \u003Ccode>` atomic commands.\n\n```bash\n# 1) Open browser to starting URL\nexpect-cli open http:\u002F\u002Flocalhost:5160\u002Flogin --browser chromium --headed --wait-until domcontentloaded\n\n# 2) Drive interactions via raw playwright code\nexpect-cli playwright \"await page.fill('input[type=email]', 'coach1@shot.com'); \\\n  await page.fill('input[type=password]', 'th3SHOT123'); \\\n  await page.click('button[type=submit]'); \\\n  await page.waitForURL(\u002F\\\u002Fhome|\\\u002Fperform\u002F, {timeout: 15000})\" \\\n  --description \"Login coach1\"\n\n# 3) Navigate + assert via page.evaluate (must use page.evaluate — NOT bare document.x)\nexpect-cli playwright \"await page.goto('http:\u002F\u002Flocalhost:5160\u002Fperform?tab=team', {waitUntil: 'domcontentloaded'}); \\\n  await page.waitForTimeout(2500); \\\n  return await page.evaluate(() => ({ \\\n    title: document.title, \\\n    bodyText: (document.body.innerText || '').slice(0, 200), \\\n    playerCount: (document.body.innerText.match(\u002FPlayer \u002Fgi) || []).length \\\n  }));\"\n\n# 4) Capture screenshot\nexpect-cli screenshot --full-page\n\n# 5) Inspect what happened\nexpect-cli console_logs\nexpect-cli network_requests --url supabase\n\n# 6) Close at end\nexpect-cli close\n```\n\n**Why this beats the TUI mission flow:**\n\n| Concern | TUI mission flow | Scripted atomic flow |\n|---|---|---|\n| Token cost | High (LLM plans + executes + reports) | Low (deterministic shell) |\n| Reproducibility | Variable (LLM may classify failures differently) | Exact (same script same result) |\n| Time-to-first-action | 2-5 min (planner reads SKILL.md, greps source) | \u003C10s (instant) |\n| Failure mode clarity | Opaque (`category=app-bug` with no detail) | Direct (you control the assertion) |\n| Browser install drama | LLM may try `npx playwright install` mid-test | Solved upfront (see prerequisite) |\n\n**Critical gotchas in `expect-cli playwright \u003Ccode>`:**\n\n1. **MUST use `page.evaluate(() => {...})`** to access DOM — bare `document.x` outside `evaluate` errors with `document is not defined` (the code runs in Node, not the page).\n2. **Return value must be JSON-serialisable** — pass arrays\u002Fobjects, not DOM nodes or Promises.\n3. **For text matching, use `document.body.innerText`** then regex against the string. Selector-based existence checks via `document.querySelector` work too.\n4. **Anchor-or-button discovery**: many SHOT UI elements are `\u003Cbutton>` not `\u003Ca>` (per perform-redesign walk 2026-05-19). Probe for BOTH before assuming click target shape:\n   ```js\n   const anchors = Array.from(document.querySelectorAll('a[href]')).filter(a => a.offsetParent !== null);\n   const buttons = Array.from(document.querySelectorAll('button')).filter(b => b.offsetParent !== null);\n   ```\n5. **`networkidle` is flaky against apps with realtime sockets** (Supabase, Firebase live queries) — they keep the network busy indefinitely. Use `domcontentloaded` + `waitForTimeout(2500)` or wait for a specific element instead.\n\n## CLI Reference\n\n| Flag | Description |\n|------|-------------|\n| `-m, --message \u003Ctext>` | Natural language test instruction |\n| `-f, --flow \u003Cslug>` | Reuse a previously saved test flow |\n| `-y, --yes` | Skip plan review, run immediately |\n| `-a, --agent \u003Cprovider>` | AI provider: `claude` or `codex` (default: claude) |\n| `-t, --target \u003Ctarget>` | What to test: `unstaged`, `branch`, or `changes` |\n| `--headed` | Show visible browser window |\n| `--verbose` | Enable verbose logging |\n| `--replay-host \u003Curl>` | Override replay viewer host |\n\n| Subcommand | Description |\n|------------|-------------|\n| `expect-cli init` | First-time global install + agent skill setup |\n| `expect-cli audit` | Run lint\u002Ftype\u002Fformat checks and suggest browser tests |\n\n## How It Works\n\n1. **Auto-detects dev server** — scans localhost ports via `lsof`, recognizes Vite\u002FNext\u002FReact\u002FAngular\n2. **Reads git diff** — scopes test plan to what actually changed\n3. **AI generates test plan** — step-by-step browser actions + assertions\n4. **Playwright executes** — real browser, real clicks, real screenshots\n5. **Reports pass\u002Ffail** — with session recording for replay\n\n## Replay Recordings & Evidence\n\nExpect-cli automatically saves session recordings to disk:\n\n```bash\n# Default location\nls \u002Ftmp\u002Fexpect-replays\u002F\n\n# Override with env var\nEXPECT_REPLAY_OUTPUT_PATH=\u002Fpath\u002Fto\u002Fdir expect-cli -m \"test X\" -y\n```\n\n**Artifacts saved per session:**\n\n| File | Format | Usage |\n|------|--------|-------|\n| `*.html` | HTML replay viewer | Open in browser to replay the full test session |\n| `*.ndjson` | rrweb event recording | Raw event data for programmatic analysis |\n| `*.webm` | Playwright video | Actual browser screen recording |\n| `*-steps.json` | Step metadata | Pass\u002Ffail per step with timestamps |\n\n**To collect evidence after a test run:**\n```bash\n# Find latest replay\nLATEST=$(ls -t \u002Ftmp\u002Fexpect-replays\u002F*.html 2>\u002Fdev\u002Fnull | head -1)\necho \"Replay: $LATEST\"\n\n# Open replay in browser\nopen \"$LATEST\"\n\n# Copy video for PR evidence\ncp \u002Ftmp\u002Fexpect-replays\u002F*.webm .\u002Ftest-evidence\u002F\n```\n\n**Note:** Screenshots are NOT saved to disk (base64 in AI context only). Use the `.webm` video for evidence extraction (see below).\n\n## Publishing Interactive Replay (Recommended)\n\nThe HTML replay is the best evidence format — reviewers can scrub through the entire session interactively. Publish via `\u002Fhere-now`:\n\n```bash\nSESSION_ID=$(ls -t \u002Ftmp\u002Fexpect-replays\u002F*.html | head -1 | xargs basename .html)\nmkdir -p \u002Ftmp\u002Freplay\ncp \u002Ftmp\u002Fexpect-replays\u002F${SESSION_ID}.html \u002Ftmp\u002Freplay\u002F\ncp \u002Ftmp\u002Fexpect-replays\u002F${SESSION_ID}.ndjson.js \u002Ftmp\u002Freplay\u002F\ncp \u002Ftmp\u002Fexpect-replays\u002F${SESSION_ID}.ndjson \u002Ftmp\u002Freplay\u002F\n\n# Publish (3 files: HTML shell + ndjson.js recording + ndjson raw data)\n\u002F.claude\u002Fskills\u002Fhere-now\u002Fscripts\u002Fpublish.sh \u002Ftmp\u002Freplay\u002F --client claude\n\n# The replay URL is: https:\u002F\u002F{slug}.here.now\u002F{SESSION_ID}.html\n# Post to PR comment with the direct link\n```\n\n**Important**: The HTML file loads the `.ndjson.js` by deriving the filename from its own name. All 3 files must be published together, and the HTML must be accessed by its original filename (not as `index.html`).\n\n**Expiry**: Anonymous publishes expire in 24h. Include the claim URL in the PR comment so the user can make it permanent.\n\n## Extracting Screenshots from Recordings\n\nSince expect-cli doesn't save per-step screenshots to disk, extract key frames from the `.webm` video using ffmpeg. This produces shareable PNG evidence for PR comments.\n\n```bash\nWEBM=$(ls -t \u002Ftmp\u002Fexpect-replays\u002F*.webm | head -1)\nmkdir -p \u002Ftmp\u002Fscreenshots\n\n# Extract frames at key timestamps (adjust based on test step timing from logs)\n# Timing guide: step 1 finish ≈ step1_duration, step 2 ≈ step1 + step2, etc.\nffmpeg -i \"$WEBM\" -ss 25 -frames:v 1 \u002Ftmp\u002Fscreenshots\u002F01-login.png -y 2>\u002Fdev\u002Fnull\nffmpeg -i \"$WEBM\" -ss 50 -frames:v 1 \u002Ftmp\u002Fscreenshots\u002F02-home.png -y 2>\u002Fdev\u002Fnull\nffmpeg -i \"$WEBM\" -ss 80 -frames:v 1 \u002Ftmp\u002Fscreenshots\u002F03-feature.png -y 2>\u002Fdev\u002Fnull\n\n# Validate screenshots show real content (not splash\u002Fblank)\nfor f in \u002Ftmp\u002Fscreenshots\u002F*.png; do\n  dims=$(magick identify -format \"%wx%h\" \"$f\" 2>\u002Fdev\u002Fnull)\n  echo \"$(basename $f): $dims ($(wc -c \u003C \"$f\") bytes)\"\ndone\n```\n\n**Timestamp estimation**: Sum step durations from the test log. If step 1 took 25s and step 2 took 17s, the step 2 completion frame is at ~42s.\n\n## Uploading Evidence for PR Comments\n\n### Option 1: imgbb (preferred for small batches)\n\n```bash\nIMGBB_API_KEY=\"${IMGBB_API_KEY:-006dfde8d5037a1e366db2bb24e915d3}\"\nfor file in \u002Ftmp\u002Fscreenshots\u002F*.png; do\n  sleep 3  # Rate limit\n  url=$(curl -s -X POST \"https:\u002F\u002Fapi.imgbb.com\u002F1\u002Fupload?key=${IMGBB_API_KEY}\" \\\n    --form \"image=@${file}\" | jq -r '.data.url \u002F\u002F empty')\n  echo \"$(basename $file) → $url\"\ndone\n```\n\n### Option 2: GitHub Draft Release (when imgbb rate-limits)\n\n```bash\ngh release create pr${PR}-evidence --repo OWNER\u002FREPO \\\n  --title \"PR #${PR} Evidence\" --notes \"Browser test screenshots\" \\\n  --draft \u002Ftmp\u002Fscreenshots\u002F*.png\n\n# Get download URLs\ngh release view pr${PR}-evidence --repo OWNER\u002FREPO --json assets \\\n  --jq '.assets[] | \"\\(.name) → \\(.url)\"'\n```\n\n### Posting to PR\n\n```bash\nBASE=\"https:\u002F\u002Fgithub.com\u002FOWNER\u002FREPO\u002Freleases\u002Fdownload\u002FTAG\"\ngh pr comment $PR --repo OWNER\u002FREPO --body \"$(cat \u003C\u003CEOF\n## ✅ Browser Test Evidence\n\n![login](${BASE}\u002F01-login.png)\n![feature](${BASE}\u002F02-feature.png)\n\n\u003Cdetails>\u003Csummary>All screenshots\u003C\u002Fsummary>\n\n![step3](${BASE}\u002F03-step.png)\n\n\u003C\u002Fdetails>\nEOF\n)\"\n```\n\n## Full Evidence Workflow (end-to-end)\n\n```bash\n# 1. Run test\nSESSION=\"expect-$(date +%s)\"\ntmux new-session -d -s \"$SESSION\"\ntmux send-keys -t \"$SESSION\" \"EXPECT_REPLAY_OUTPUT_PATH=\u002Ftmp\u002Fexpect-replays expect-cli -m 'test description' --headed -y --verbose 2>&1 | tee \u002Ftmp\u002Fexpect.log\" C-m\n\n# 2. Wait for completion\nsleep 120 && tmux capture-pane -t \"$SESSION\" -p | tail -20\n\n# 3. Extract frames\nWEBM=$(ls -t \u002Ftmp\u002Fexpect-replays\u002F*.webm | head -1)\nffmpeg -i \"$WEBM\" -ss 30 -frames:v 1 \u002Ftmp\u002Fscreenshots\u002Fstep1.png -y 2>\u002Fdev\u002Fnull\nffmpeg -i \"$WEBM\" -ss 60 -frames:v 1 \u002Ftmp\u002Fscreenshots\u002Fstep2.png -y 2>\u002Fdev\u002Fnull\n\n# 4. Validate (read with Claude to verify content)\n# Use Read tool on each PNG to verify it shows the expected UI state\n\n# 5a. Publish interactive replay (PREFERRED)\nSESSION_ID=$(ls -t \u002Ftmp\u002Fexpect-replays\u002F*.html | head -1 | xargs basename .html)\nmkdir -p \u002Ftmp\u002Freplay\ncp \u002Ftmp\u002Fexpect-replays\u002F${SESSION_ID}.{html,ndjson.js,ndjson} \u002Ftmp\u002Freplay\u002F\n\u002F.claude\u002Fskills\u002Fhere-now\u002Fscripts\u002Fpublish.sh \u002Ftmp\u002Freplay\u002F --client claude\n# Post the replay URL to PR\n\n# 5b. Upload static screenshots (FALLBACK if replay not suitable)\ngh release create prN-evidence --draft \u002Ftmp\u002Fscreenshots\u002F*.png\ngh pr comment N --body \"## Evidence\\n![step1](url)\\n![step2](url)\"\n```\n\n## Integration with \u002Fvalidate\n\nWhen `\u002Fvalidate` is invoked, this skill adds browser verification as Step 2.5:\n\n```\nStep 2.5: Browser Verification (via expect-test)\n- Auto-detect running dev server\n- Generate browser tests from the implementation plan\n- Execute tests and capture screenshots\n- Report visual verification results alongside code validation\n```\n\n## Result Interpretation\n\n```\n✔ = Test passed\n✘ = Test failed (with description of what went wrong)\n```\n\nResults include:\n- Pass\u002Ffail status per test step\n- Screenshots at each step\n- Session recording URL for replay\n- Specific failure descriptions (e.g., \"element not found\", \"wrong color value\")\n\n### Failure classification taxonomy (TUI mode)\n\nThe TUI agent attaches a `category=...; domain=...` label to each ✘ failure. Decoded:\n\n| Category | Meaning | Examples | Agent action |\n|---|---|---|---|\n| `app-bug` | Real application behavior mismatch | Element missing, wrong nav target, broken render | File a bead; this is the test's job |\n| `env-issue` | Infrastructure \u002F setup \u002F auth problem | Dev server down, login failed, network 5xx, fixture data absent | Fix the env, don't blame the app |\n| `timeout` | Assertion timeout | Element never appeared, page never settled | Often = data not loading; check the env first |\n\nWhen you see `env-issue` consecutively across early steps (login + landing), it almost always means **the test infrastructure isn't ready** (dev server pointing at wrong DB, missing test users, missing test data). Fix the env before chasing `app-bug` reports.\n\n**The scripted atomic-flow pattern (above) sidesteps this entirely** — you control the assertions so failure classification is just your shell exit code + the returned object.\n\n## Limitations\n\n- Early stage tool (v0.0.x) — may crash on complex scenarios\n- macOS\u002FLinux only (uses `lsof` for port detection)\n- Requires a running dev server on localhost\n- No custom Playwright config — browser is fully managed\n- Interactive TUI — must use `-y` flag or run in tmux for automation\n- **Headed mode steals window focus** — when the user has Chrome open for other work, prefer `--browser-mode headless` (TUI) or `expect-cli open … --browser chromium` without `--headed` (atomic). Only use `--headed` when the user explicitly wants to watch.\n- **Per-step screenshots NOT saved to disk by default** — `expect-cli screenshot` saves under `\u002Ftmp\u002Fexpect-artifacts\u002Fscreenshot-{n}.png`, but mid-`expect-cli playwright` steps don't auto-capture. Use the `.webm` video + ffmpeg, OR call `expect-cli screenshot` explicitly between steps in scripted flows.\n- **TUI mission planner can wander 2-5 min on first run** — bundled Playwright browser install + project introspection. Pre-install via the Prerequisites section's command, or skip the TUI entirely and use the scripted atomic-flow pattern.\n\n## Environment\n\n| Variable | Description |\n|----------|-------------|\n| `NO_TELEMETRY=1` | Disable analytics\u002Ftelemetry |\n",{"name":5,"description":7,"argument-hint":24,"user-invocable":25},"[-m message] [--headed] [-y]",true,"main","9cf32d1e636db09530158af14ba301bfa5d401e9","packages\u002Fqe-agent-pack\u002F.apm\u002Fskills\u002Fexpect-test","https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit","https:\u002F\u002Fgithub.com\u002Fstevengonsalvez","stevengonsalvez.com","2026-11-12 00:50:59",{"source":8,"skillName":5,"commands":34,"prompt":44,"zipAvailable":25},[35,38,41],{"tool":36,"command":37},"npx","npx skills add stevengonsalvez\u002Fainb-toolkit --skill expect-test",{"tool":39,"command":40},"bunx","bunx skills add stevengonsalvez\u002Fainb-toolkit --skill expect-test",{"tool":42,"command":43},"pnpm","pnpm dlx skills add stevengonsalvez\u002Fainb-toolkit --skill expect-test","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Fpackages\u002Fqe-agent-pack\u002F.apm\u002Fskills\u002Fexpect-test\u002FSKILL.md",[],[47,54,61,67,74,81],{"skillNo":48,"name":49,"nameZh":6,"description":50,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":51,"categories":52,"keywords":53,"publishedAt":19},"01M75WK9B8DKVJBJJ4CX2QMJMX","well-formed","Fixture skill that satisfies every mechanical rule, used to prove the conformance check passes a clean tree instead of failing everything.","https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fagentskills\u002Fclean\u002Fwell-formed\u002FSKILL.md",[],[],{"skillNo":55,"name":56,"nameZh":6,"description":57,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":58,"categories":59,"keywords":60,"publishedAt":19},"01M75WK6ZGDXJJN8879K49QEYG","underscore-tools","Fixture skill that spells the tool allowlist key with an underscore instead of the hyphen the spec requires.","https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fagentskills\u002Fbroken\u002Funderscore-tools\u002FSKILL.md",[],[],{"skillNo":62,"name":63,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":64,"categories":65,"keywords":66,"publishedAt":19},"01M75WK524MVT2H9E27ZCEA3HR","no-frontmatter","https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fagentskills\u002Fbroken\u002Fno-frontmatter\u002FSKILL.md",[],[],{"skillNo":68,"name":69,"nameZh":6,"description":70,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":71,"categories":72,"keywords":73,"publishedAt":19},"01M75WJWZBMKR8R60PG3T3MCCC","long-description","Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling. Fixture skill with a deliberately oversized description used to prove the conformance check rejects anything past the 1024 character ceiling.","https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fagentskills\u002Fbroken\u002Flong-description\u002FSKILL.md",[],[],{"skillNo":75,"name":76,"nameZh":6,"description":77,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":78,"categories":79,"keywords":80,"publishedAt":19},"01M75WJVRH3J22WKED7E5D01S8","some-other-name","Fixture skill whose frontmatter name does not match its parent directory name.","https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fagentskills\u002Fbroken\u002Fdir-mismatch\u002FSKILL.md",[],[],{"skillNo":82,"name":83,"nameZh":6,"description":84,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":85,"categories":86,"keywords":87,"publishedAt":19},"01M75WJTC7H5FRP6S129VBSMZC","bad_Name","Fixture skill whose name uses an underscore and a capital letter, so it violates the agentskills.io name pattern.","https:\u002F\u002Fgithub.com\u002Fstevengonsalvez\u002Fainb-toolkit\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fagentskills\u002Fbroken\u002Fbad_Name\u002FSKILL.md",[],[]]