[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M72YVHT2BZKEKE5RVBV64PEG":3,"skill-same:01M72YVHT2BZKEKE5RVBV64PEG":99},{"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":16,"sourceHtmlUrl":17,"categories":18,"keywords":19,"publishedAt":20,"linkOnly":21,"markdownContent":22,"markdownBody":23,"frontMatter":24,"refName":31,"commitSha":32,"skillDirectory":33,"forksCount":14,"repositoryHtmlUrl":34,"repositoryHomepageUrl":6,"ownerHtmlUrl":35,"ownerBlogUrl":6,"ownerDescription":36,"updatedAt":37,"install":38,"resources":51},"01M72YVHT2BZKEKE5RVBV64PEG","autobrowse",null,"Self-improving browser automation via the auto-research loop. Iteratively runs a browsing task, reads the trace, and improves the navigation skill (strategy.md) until it reliably passes. Supports parallel runs across multiple tasks using sub-agents. Use when you want to build or improve browser automation skills for specific website tasks.","JOHNadonis\u002Fclaude-code-skills","JOHNadonis","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F31716775?v=4","01M72YRBKCVM4ZYJFF7Z71KVQQ","Hongcai Wang","01M72YRBMFSBZNVRJ4D48MWMZ3",0,"MIT","https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Fautobrowse\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Fautobrowse\u002FSKILL.md",[],[],"2026-11-11 15:11:14",false,"---\nname: autobrowse\ndescription: Self-improving browser automation via the auto-research loop. Iteratively runs a browsing task, reads the trace, and improves the navigation skill (strategy.md) until it reliably passes. Supports parallel runs across multiple tasks using sub-agents. Use when you want to build or improve browser automation skills for specific website tasks.\nlicense: See LICENSE.txt\ncompatibility: \"Requires Node.js 18+, browse CLI, and ANTHROPIC_API_KEY. Run from the autobrowse app directory.\"\nallowed-tools: Bash Read Write Edit Glob Grep Agent\nmetadata:\n  author: browserbase\n  homepage: https:\u002F\u002Fgithub.com\u002Fbrowserbase\u002Fskills\n---\n\n# AutoBrowse — Self-Improving Browser Skill\n\nBuild reliable browser automation skills through iterative experimentation. An inner agent browses the site (`evaluate.ts`). You — the outer agent — read what happened and improve the instructions (`strategy.md`). Repeat until it passes consistently.\n\n## Entry Points\n\nInvocation is flexible — both explicit flags and free-form natural language work:\n\n```\n\u002Fautobrowse --task google-flights\n\u002Fautobrowse --task google-flights --iterations 10 --env remote\n\u002Fautobrowse --tasks google-flights,amazon-add-to-cart\n\u002Fautobrowse --all\n\n# Also fine — parse freely:\n\u002Fautobrowse https:\u002F\u002Fflights.google.com\u002F\n\u002Fautobrowse book a flight on delta.com\n\u002Fautobrowse fix the existing google-flights skill\n```\n\nWhen the user drops a URL or free-form instruction instead of `--task \u003Cname>`:\n- If an existing task in `${WORKSPACE}\u002Ftasks\u002F` clearly matches the site\u002Fintent, use it.\n- Otherwise, pick a short kebab-case name, create `${WORKSPACE}\u002Ftasks\u002F\u003Cname>\u002Ftask.md` from `${CLAUDE_SKILL_DIR}\u002Freferences\u002Fexample-task.md`, fill in the URL\u002Fgoal based on what the user said, and proceed. Tell the user the chosen name in one line.\n\n---\n\n## How to run\n\n### Step 1 — Parse arguments and orient\n\nCheck what was passed:\n- `--task \u003Cname>` → single task mode\n- `--tasks a,b,c` or `--all` → multi-task mode (spawn sub-agents)\n- `--iterations N` → how many evaluate → improve cycles (default: 5)\n- `--env local|remote` → browser environment (default: local; use remote for bot-protected sites)\n\nIf the user passed free-form text instead, map it to one of the above before continuing.\n\n### Step 2 — Set up the workspace\n\nAll training artifacts (task definitions, strategy iterations, traces, reports) live in a workspace directory in the **current working directory** — NOT inside `~\u002F.claude\u002Fskills\u002F`. This keeps the inner agent's file writes out of Claude's home dir and away from permission friction.\n\nDefault workspace: `${CWD}\u002Fautobrowse\u002F`\n\n```bash\nmkdir -p .\u002Fautobrowse\u002Ftasks .\u002Fautobrowse\u002Ftraces .\u002Fautobrowse\u002Freports\n```\n\nIf the task directory (`.\u002Fautobrowse\u002Ftasks\u002F\u003Ctask>\u002Ftask.md`) doesn't exist yet, scaffold it:\n\n```bash\nmkdir -p .\u002Fautobrowse\u002Ftasks\u002F\u003Ctask>\ncp ${CLAUDE_SKILL_DIR}\u002Freferences\u002Fexample-task.md .\u002Fautobrowse\u002Ftasks\u002F\u003Ctask>\u002Ftask.md\n# Then edit task.md to describe the URL, inputs, steps, and expected JSON output\n```\n\nThe skill source at `${CLAUDE_SKILL_DIR}` stays read-only — only `.\u002Fautobrowse\u002F` in CWD gets written to during training. Graduation (final step) writes a single file to `~\u002F.claude\u002Fskills\u002F\u003Ctask>\u002FSKILL.md`.\n\nList available tasks:\n```bash\nls .\u002Fautobrowse\u002Ftasks\u002F\n```\n\n### Step 3 — Multi-task: spawn parallel sub-agents\n\nIf running multiple tasks, use the Agent tool to spawn one sub-agent per task simultaneously. Each sub-agent receives a self-contained prompt to run the full autobrowse loop for its task:\n\n> \"You are running the autobrowse skill for task `\u003Cname>`. Workspace: `\u003Cabsolute-path-to-workspace>` (e.g. `\u002Fpath\u002Fto\u002Fproject\u002Fautobrowse`). Run `\u003CN>` iterations of: evaluate → read trace → improve strategy.md → repeat. Use `--env \u003Cenv>`. Pass `--workspace \u003Cworkspace>` to every evaluate.mjs invocation. Follow the autobrowse loop instructions exactly.\n>\n> When graduating, install the skill to `~\u002F.claude\u002Fskills\u002F\u003Ctask-name>\u002FSKILL.md` with proper agentskills frontmatter (name + description). Do not just copy strategy.md — write a self-contained skill.\n>\n> At the end, output a structured summary with: task name, pass\u002Ffail on final run, total cumulative cost, iterations completed, per-iteration table (iter number, turns, cost, status, hypothesis tested), and 2-3 bullet key learnings.\"\n\nSpawn all sub-agents in parallel, wait for all to complete, then collect their summaries and write the session report.\n\n**For single task**, skip this step and run the loop directly below.\n\n---\n\n## The Loop (run this for each task)\n\n### Iteration start\n\nCheck that `.\u002Fautobrowse\u002Ftasks\u002F\u003Ctask>\u002Ftask.md` exists (scaffold it from the template if not — see Step 2). `strategy.md` is auto-created empty by the harness on first run.\n\n### Requirements\n\n- `ANTHROPIC_API_KEY` must be in the environment (or in a `.env` file in CWD — `evaluate.mjs` auto-loads it). If missing, the harness prints a clear error and exits; don't hunt for keys in other paths.\n\n### Run the inner agent\n\n```bash\nnode ${CLAUDE_SKILL_DIR}\u002Fscripts\u002Fevaluate.mjs --task \u003Ctask-name> --workspace .\u002Fautobrowse\n# or for bot-protected sites:\nnode ${CLAUDE_SKILL_DIR}\u002Fscripts\u002Fevaluate.mjs --task \u003Ctask-name> --workspace .\u002Fautobrowse --env remote\n```\n\nThis runs the browser session and writes a full trace to `.\u002Fautobrowse\u002Ftraces\u002F\u003Ctask>\u002Flatest\u002F`.\n\n### Read the trace\n\n```bash\ncat .\u002Fautobrowse\u002Ftraces\u002F\u003Ctask-name>\u002Flatest\u002Fsummary.md\n```\n\nThe summary has duration, cost, turns, the decision log, and the final JSON output.\n\nIf the agent failed or got stuck, look deeper:\n- Read `.\u002Fautobrowse\u002Ftraces\u002F\u003Ctask-name>\u002Flatest\u002Ftrace.json` — search for the failure turn\n- Read screenshots around the failure point with the Read tool\n\n### Form one hypothesis\n\nFind the exact turn where things went wrong. What single heuristic would have prevented it?\n\nExamples:\n- \"After clicking the dropdown, wait 1s — options animate in before they're clickable\"\n- \"Navigate directly to `\u002Fpay-invoice\u002F` — skip the landing page entirely\"\n- \"Use `browse fill #field_3 value` not `browse type` — this field clears on focus\"\n- \"The page shows a spinner at turn 8 — add `browse wait timeout 2000` before snapshot\"\n\n### Update strategy.md\n\nEdit `.\u002Fautobrowse\u002Ftasks\u002F\u003Ctask-name>\u002Fstrategy.md`. Keep everything that worked. Fix the specific failure. Add a concrete heuristic.\n\nGood strategies have:\n- **Fast path**: direct URL or shortcuts to skip exploration\n- **Step-by-step workflow**: exact sequence with timing notes\n- **Site-specific knowledge**: selector IDs, form field names, success indicators\n- **Failure recovery**: what to do when X goes wrong\n\n### Judge the result\n\nRead the new summary. Did it pass? Make clear progress?\n- **Pass or progress** → keep, next iteration\n- **No progress or regression** → revert strategy.md to the previous version and try a different hypothesis\n\n### After all iterations — publish if ready\n\nIf the task passed on 2+ of the last 3 iterations **or has reached the max iteration limit**, install it as a Claude Code skill. **Do not just copy strategy.md** — the skill must be self-contained and useful to someone who has never seen this codebase. If graduating at max iterations without a clean pass, note the known failure point but still document everything learned.\n\nInstall by writing to `~\u002F.claude\u002Fskills\u002F\u003Ctask-name>\u002FSKILL.md`:\n\n```bash\nmkdir -p ~\u002F.claude\u002Fskills\u002F\u003Ctask-name>\n```\n\nUse this structure for the SKILL.md:\n\n```markdown\n---\nname: \u003Ctask-name>\ndescription: \u003C1-2 sentences describing what this skill does and when to use it. Include trigger keywords.>\n---\n\n# \u003CTask Title> — Browser Skill\n\n## Purpose\n\u003C1-2 sentences: what this automates and why it exists.>\n\n## When to Use\n\u003CWhen should someone reach for this skill.>\n\n## Browse CLI Reference\nThe inner agent uses the `browse` CLI. Key commands for this task:\n- `browse stop` — kill existing session (always run before switching to remote)\n- `browse env remote` — start a fresh Browserbase cloud session\n- `browse newpage \u003Curl>` — open URL in a new tab (required in remote mode — `browse open` fails with \"no page available\")\n- `browse open \u003Curl>` — navigate existing tab (local mode only)\n- `browse wait load` — wait for page to finish loading\n- `browse wait timeout \u003Cms>` — wait a fixed amount of time for spinners or animations\n- `browse wait selector \"\u003Cselector>\"` — wait for an element to become visible\n- `browse get title` — verify you're on the right page\n- `browse get text body` — extract all visible text (preferred for content extraction)\n- `browse snapshot` — get accessibility tree; each node has a ref in `[X-Y]` format (e.g. `[0-5]`, `[2-147]`)\n- `browse click [X-Y]` — click element by ref from the latest snapshot (include the brackets)\n\n**Never use `--session \u003Cname>` flags in SKILL.md.** Named sessions are a parallel-run workaround — they contaminate skills with infrastructure concerns. Skills must work in isolation with the default session.\n\n## Workflow\n\n### Step 1 — Start session\n\u003Cexact browse commands in order>\n\n### Step 2 — Navigate\n\u003Cexact URL and verification steps>\n\n### Step 3 — Extract\n\u003Cexact extraction commands>\n\n### Step 4 — Output\n\u003Cwhat JSON to emit, referencing the schema below>\n\n## Site-Specific Gotchas\n\u003CBullet list of every hard-won heuristic from the iterations. This is the core value of the skill.>\n\n## Failure Recovery\n\u003CWhat to do when navigation fails, session is contaminated, or extraction returns garbage>\n\n## Expected Output\n```json\n\u003Cpaste the exact expected output schema from task.md>\n```\n```\n\nAfter writing the SKILL.md, confirm it's installed:\n```bash\nls ~\u002F.claude\u002Fskills\u002F\u003Ctask-name>\u002FSKILL.md\n```\n\nThe skill is now available as `\u002F\u003Ctask-name>` in Claude Code.\n\n---\n\n## Final report (multi-task mode)\n\nAfter all sub-agents complete, print a markdown table:\n\n| Task | Iterations | Final Status | Graduated | Cost |\n|------|-----------|--------------|-----------|------|\n| google-flights | 5 | ✅ pass | yes | $0.42 |\n| amazon-add-to-cart | 5 | ❌ fail | no | $1.20 |\n\nThen write a persistent session report to `.\u002Fautobrowse\u002Freports\u002F` so there's a durable record of the run inside the workspace:\n\n```bash\nmkdir -p .\u002Fautobrowse\u002Freports\n```\n\nWrite the file `.\u002Fautobrowse\u002Freports\u002FYYYY-MM-DD-HH-MM-\u003Ctasks>.md` with:\n\n```markdown\n# AutoBrowse Session Report\n**Date:** \u003CISO date>\n**Tasks:** \u003Ccomma-separated list>\n**Environment:** remote|local\n**Total cost:** $X.XX\n\n## Results\n\n| Task | Iterations | Pass Rate | Final Status | Graduated | Cost |\n|------|-----------|-----------|--------------|-----------|------|\n| ... | ... | X\u002F5 | ✅\u002F❌ | yes\u002Fno | $X.XX |\n\n## Per-Task Learnings\n\n### \u003Ctask-name>\n- **Key insight 1:** \u003Cwhat the agent learned>\n- **Key insight 2:** \u003Canother heuristic>\n- **Failure mode fixed:** \u003Cwhat was failing and how it was resolved>\n\n## Iteration Log\n\n### \u003Ctask-name>\n| Iter | Turns | Cost | Status | Hypothesis tested |\n|------|-------|------|--------|-------------------|\n| 1 | 79 | $18.75 | ❌ fail | baseline |\n| 2 | 9 | $0.26 | ✅ pass | session contamination fix |\n| ... | ... | ... | ... | ... |\n```\n\n---\n\n## Rules\n\n- **Only edit `strategy.md`** — never touch `task.md` (unless creating it from the template) or `evaluate.mjs`\n- **Stay in the workspace** — all training writes go to `.\u002Fautobrowse\u002F`, never to `~\u002F.claude\u002Fskills\u002Fautobrowse\u002F`. The skill source is read-only.\n- **One hypothesis per iteration** — test one change at a time\n- **Build on wins** — keep what worked, add to it\n- **Trust the trace** — the inner agent shows exactly what it saw and did\n- **Graduate to `~\u002F.claude\u002Fskills\u002F`** — the only file you write there is the final graduated `SKILL.md`\n","# AutoBrowse — Self-Improving Browser Skill\n\nBuild reliable browser automation skills through iterative experimentation. An inner agent browses the site (`evaluate.ts`). You — the outer agent — read what happened and improve the instructions (`strategy.md`). Repeat until it passes consistently.\n\n## Entry Points\n\nInvocation is flexible — both explicit flags and free-form natural language work:\n\n```\n\u002Fautobrowse --task google-flights\n\u002Fautobrowse --task google-flights --iterations 10 --env remote\n\u002Fautobrowse --tasks google-flights,amazon-add-to-cart\n\u002Fautobrowse --all\n\n# Also fine — parse freely:\n\u002Fautobrowse https:\u002F\u002Fflights.google.com\u002F\n\u002Fautobrowse book a flight on delta.com\n\u002Fautobrowse fix the existing google-flights skill\n```\n\nWhen the user drops a URL or free-form instruction instead of `--task \u003Cname>`:\n- If an existing task in `${WORKSPACE}\u002Ftasks\u002F` clearly matches the site\u002Fintent, use it.\n- Otherwise, pick a short kebab-case name, create `${WORKSPACE}\u002Ftasks\u002F\u003Cname>\u002Ftask.md` from `${CLAUDE_SKILL_DIR}\u002Freferences\u002Fexample-task.md`, fill in the URL\u002Fgoal based on what the user said, and proceed. Tell the user the chosen name in one line.\n\n---\n\n## How to run\n\n### Step 1 — Parse arguments and orient\n\nCheck what was passed:\n- `--task \u003Cname>` → single task mode\n- `--tasks a,b,c` or `--all` → multi-task mode (spawn sub-agents)\n- `--iterations N` → how many evaluate → improve cycles (default: 5)\n- `--env local|remote` → browser environment (default: local; use remote for bot-protected sites)\n\nIf the user passed free-form text instead, map it to one of the above before continuing.\n\n### Step 2 — Set up the workspace\n\nAll training artifacts (task definitions, strategy iterations, traces, reports) live in a workspace directory in the **current working directory** — NOT inside `~\u002F.claude\u002Fskills\u002F`. This keeps the inner agent's file writes out of Claude's home dir and away from permission friction.\n\nDefault workspace: `${CWD}\u002Fautobrowse\u002F`\n\n```bash\nmkdir -p .\u002Fautobrowse\u002Ftasks .\u002Fautobrowse\u002Ftraces .\u002Fautobrowse\u002Freports\n```\n\nIf the task directory (`.\u002Fautobrowse\u002Ftasks\u002F\u003Ctask>\u002Ftask.md`) doesn't exist yet, scaffold it:\n\n```bash\nmkdir -p .\u002Fautobrowse\u002Ftasks\u002F\u003Ctask>\ncp ${CLAUDE_SKILL_DIR}\u002Freferences\u002Fexample-task.md .\u002Fautobrowse\u002Ftasks\u002F\u003Ctask>\u002Ftask.md\n# Then edit task.md to describe the URL, inputs, steps, and expected JSON output\n```\n\nThe skill source at `${CLAUDE_SKILL_DIR}` stays read-only — only `.\u002Fautobrowse\u002F` in CWD gets written to during training. Graduation (final step) writes a single file to `~\u002F.claude\u002Fskills\u002F\u003Ctask>\u002FSKILL.md`.\n\nList available tasks:\n```bash\nls .\u002Fautobrowse\u002Ftasks\u002F\n```\n\n### Step 3 — Multi-task: spawn parallel sub-agents\n\nIf running multiple tasks, use the Agent tool to spawn one sub-agent per task simultaneously. Each sub-agent receives a self-contained prompt to run the full autobrowse loop for its task:\n\n> \"You are running the autobrowse skill for task `\u003Cname>`. Workspace: `\u003Cabsolute-path-to-workspace>` (e.g. `\u002Fpath\u002Fto\u002Fproject\u002Fautobrowse`). Run `\u003CN>` iterations of: evaluate → read trace → improve strategy.md → repeat. Use `--env \u003Cenv>`. Pass `--workspace \u003Cworkspace>` to every evaluate.mjs invocation. Follow the autobrowse loop instructions exactly.\n>\n> When graduating, install the skill to `~\u002F.claude\u002Fskills\u002F\u003Ctask-name>\u002FSKILL.md` with proper agentskills frontmatter (name + description). Do not just copy strategy.md — write a self-contained skill.\n>\n> At the end, output a structured summary with: task name, pass\u002Ffail on final run, total cumulative cost, iterations completed, per-iteration table (iter number, turns, cost, status, hypothesis tested), and 2-3 bullet key learnings.\"\n\nSpawn all sub-agents in parallel, wait for all to complete, then collect their summaries and write the session report.\n\n**For single task**, skip this step and run the loop directly below.\n\n---\n\n## The Loop (run this for each task)\n\n### Iteration start\n\nCheck that `.\u002Fautobrowse\u002Ftasks\u002F\u003Ctask>\u002Ftask.md` exists (scaffold it from the template if not — see Step 2). `strategy.md` is auto-created empty by the harness on first run.\n\n### Requirements\n\n- `ANTHROPIC_API_KEY` must be in the environment (or in a `.env` file in CWD — `evaluate.mjs` auto-loads it). If missing, the harness prints a clear error and exits; don't hunt for keys in other paths.\n\n### Run the inner agent\n\n```bash\nnode ${CLAUDE_SKILL_DIR}\u002Fscripts\u002Fevaluate.mjs --task \u003Ctask-name> --workspace .\u002Fautobrowse\n# or for bot-protected sites:\nnode ${CLAUDE_SKILL_DIR}\u002Fscripts\u002Fevaluate.mjs --task \u003Ctask-name> --workspace .\u002Fautobrowse --env remote\n```\n\nThis runs the browser session and writes a full trace to `.\u002Fautobrowse\u002Ftraces\u002F\u003Ctask>\u002Flatest\u002F`.\n\n### Read the trace\n\n```bash\ncat .\u002Fautobrowse\u002Ftraces\u002F\u003Ctask-name>\u002Flatest\u002Fsummary.md\n```\n\nThe summary has duration, cost, turns, the decision log, and the final JSON output.\n\nIf the agent failed or got stuck, look deeper:\n- Read `.\u002Fautobrowse\u002Ftraces\u002F\u003Ctask-name>\u002Flatest\u002Ftrace.json` — search for the failure turn\n- Read screenshots around the failure point with the Read tool\n\n### Form one hypothesis\n\nFind the exact turn where things went wrong. What single heuristic would have prevented it?\n\nExamples:\n- \"After clicking the dropdown, wait 1s — options animate in before they're clickable\"\n- \"Navigate directly to `\u002Fpay-invoice\u002F` — skip the landing page entirely\"\n- \"Use `browse fill #field_3 value` not `browse type` — this field clears on focus\"\n- \"The page shows a spinner at turn 8 — add `browse wait timeout 2000` before snapshot\"\n\n### Update strategy.md\n\nEdit `.\u002Fautobrowse\u002Ftasks\u002F\u003Ctask-name>\u002Fstrategy.md`. Keep everything that worked. Fix the specific failure. Add a concrete heuristic.\n\nGood strategies have:\n- **Fast path**: direct URL or shortcuts to skip exploration\n- **Step-by-step workflow**: exact sequence with timing notes\n- **Site-specific knowledge**: selector IDs, form field names, success indicators\n- **Failure recovery**: what to do when X goes wrong\n\n### Judge the result\n\nRead the new summary. Did it pass? Make clear progress?\n- **Pass or progress** → keep, next iteration\n- **No progress or regression** → revert strategy.md to the previous version and try a different hypothesis\n\n### After all iterations — publish if ready\n\nIf the task passed on 2+ of the last 3 iterations **or has reached the max iteration limit**, install it as a Claude Code skill. **Do not just copy strategy.md** — the skill must be self-contained and useful to someone who has never seen this codebase. If graduating at max iterations without a clean pass, note the known failure point but still document everything learned.\n\nInstall by writing to `~\u002F.claude\u002Fskills\u002F\u003Ctask-name>\u002FSKILL.md`:\n\n```bash\nmkdir -p ~\u002F.claude\u002Fskills\u002F\u003Ctask-name>\n```\n\nUse this structure for the SKILL.md:\n\n```markdown\n---\nname: \u003Ctask-name>\ndescription: \u003C1-2 sentences describing what this skill does and when to use it. Include trigger keywords.>\n---\n\n# \u003CTask Title> — Browser Skill\n\n## Purpose\n\u003C1-2 sentences: what this automates and why it exists.>\n\n## When to Use\n\u003CWhen should someone reach for this skill.>\n\n## Browse CLI Reference\nThe inner agent uses the `browse` CLI. Key commands for this task:\n- `browse stop` — kill existing session (always run before switching to remote)\n- `browse env remote` — start a fresh Browserbase cloud session\n- `browse newpage \u003Curl>` — open URL in a new tab (required in remote mode — `browse open` fails with \"no page available\")\n- `browse open \u003Curl>` — navigate existing tab (local mode only)\n- `browse wait load` — wait for page to finish loading\n- `browse wait timeout \u003Cms>` — wait a fixed amount of time for spinners or animations\n- `browse wait selector \"\u003Cselector>\"` — wait for an element to become visible\n- `browse get title` — verify you're on the right page\n- `browse get text body` — extract all visible text (preferred for content extraction)\n- `browse snapshot` — get accessibility tree; each node has a ref in `[X-Y]` format (e.g. `[0-5]`, `[2-147]`)\n- `browse click [X-Y]` — click element by ref from the latest snapshot (include the brackets)\n\n**Never use `--session \u003Cname>` flags in SKILL.md.** Named sessions are a parallel-run workaround — they contaminate skills with infrastructure concerns. Skills must work in isolation with the default session.\n\n## Workflow\n\n### Step 1 — Start session\n\u003Cexact browse commands in order>\n\n### Step 2 — Navigate\n\u003Cexact URL and verification steps>\n\n### Step 3 — Extract\n\u003Cexact extraction commands>\n\n### Step 4 — Output\n\u003Cwhat JSON to emit, referencing the schema below>\n\n## Site-Specific Gotchas\n\u003CBullet list of every hard-won heuristic from the iterations. This is the core value of the skill.>\n\n## Failure Recovery\n\u003CWhat to do when navigation fails, session is contaminated, or extraction returns garbage>\n\n## Expected Output\n```json\n\u003Cpaste the exact expected output schema from task.md>\n```\n```\n\nAfter writing the SKILL.md, confirm it's installed:\n```bash\nls ~\u002F.claude\u002Fskills\u002F\u003Ctask-name>\u002FSKILL.md\n```\n\nThe skill is now available as `\u002F\u003Ctask-name>` in Claude Code.\n\n---\n\n## Final report (multi-task mode)\n\nAfter all sub-agents complete, print a markdown table:\n\n| Task | Iterations | Final Status | Graduated | Cost |\n|------|-----------|--------------|-----------|------|\n| google-flights | 5 | ✅ pass | yes | $0.42 |\n| amazon-add-to-cart | 5 | ❌ fail | no | $1.20 |\n\nThen write a persistent session report to `.\u002Fautobrowse\u002Freports\u002F` so there's a durable record of the run inside the workspace:\n\n```bash\nmkdir -p .\u002Fautobrowse\u002Freports\n```\n\nWrite the file `.\u002Fautobrowse\u002Freports\u002FYYYY-MM-DD-HH-MM-\u003Ctasks>.md` with:\n\n```markdown\n# AutoBrowse Session Report\n**Date:** \u003CISO date>\n**Tasks:** \u003Ccomma-separated list>\n**Environment:** remote|local\n**Total cost:** $X.XX\n\n## Results\n\n| Task | Iterations | Pass Rate | Final Status | Graduated | Cost |\n|------|-----------|-----------|--------------|-----------|------|\n| ... | ... | X\u002F5 | ✅\u002F❌ | yes\u002Fno | $X.XX |\n\n## Per-Task Learnings\n\n### \u003Ctask-name>\n- **Key insight 1:** \u003Cwhat the agent learned>\n- **Key insight 2:** \u003Canother heuristic>\n- **Failure mode fixed:** \u003Cwhat was failing and how it was resolved>\n\n## Iteration Log\n\n### \u003Ctask-name>\n| Iter | Turns | Cost | Status | Hypothesis tested |\n|------|-------|------|--------|-------------------|\n| 1 | 79 | $18.75 | ❌ fail | baseline |\n| 2 | 9 | $0.26 | ✅ pass | session contamination fix |\n| ... | ... | ... | ... | ... |\n```\n\n---\n\n## Rules\n\n- **Only edit `strategy.md`** — never touch `task.md` (unless creating it from the template) or `evaluate.mjs`\n- **Stay in the workspace** — all training writes go to `.\u002Fautobrowse\u002F`, never to `~\u002F.claude\u002Fskills\u002Fautobrowse\u002F`. The skill source is read-only.\n- **One hypothesis per iteration** — test one change at a time\n- **Build on wins** — keep what worked, add to it\n- **Trust the trace** — the inner agent shows exactly what it saw and did\n- **Graduate to `~\u002F.claude\u002Fskills\u002F`** — the only file you write there is the final graduated `SKILL.md`\n",{"name":5,"license":25,"metadata":26,"description":7,"allowed-tools":29,"compatibility":30},"See LICENSE.txt",{"author":27,"homepage":28},"browserbase","https:\u002F\u002Fgithub.com\u002Fbrowserbase\u002Fskills","Bash Read Write Edit Glob Grep Agent","Requires Node.js 18+, browse CLI, and ANTHROPIC_API_KEY. Run from the autobrowse app directory.","main","4ac29a24db9fc8bba16c61b1cdaa3f5058934247","skills\u002Fautobrowse","https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills","https:\u002F\u002Fgithub.com\u002FJOHNadonis","Java\u002FGo\u002FC++，student majoring in computer.","2026-10-26 22:44:45",{"source":8,"skillName":5,"commands":39,"prompt":49,"zipAvailable":50},[40,43,46],{"tool":41,"command":42},"npx","npx skills add JOHNadonis\u002Fclaude-code-skills --skill autobrowse",{"tool":44,"command":45},"bunx","bunx skills add JOHNadonis\u002Fclaude-code-skills --skill autobrowse",{"tool":47,"command":48},"pnpm","pnpm dlx skills add JOHNadonis\u002Fclaude-code-skills --skill autobrowse","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Fautobrowse\u002FSKILL.md",true,[52,58,64,69,73,77,81,85,89,93],{"path":53,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":56,"sourceHtmlUrl":57},".gitignore","OTHER","其他",50,"https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Fautobrowse\u002F.gitignore",{"path":59,"resourceType":60,"resourceTypeDesc":61,"sizeBytes":62,"sourceHtmlUrl":63},"EXAMPLES.md","REFERENCE","参考文档",1894,"https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Fautobrowse\u002FEXAMPLES.md",{"path":65,"resourceType":66,"resourceTypeDesc":67,"sizeBytes":68,"sourceHtmlUrl":16},"LICENSE.txt","LICENSE","许可证",1074,{"path":70,"resourceType":60,"resourceTypeDesc":61,"sizeBytes":71,"sourceHtmlUrl":72},"package-lock.json",16212,"https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Fautobrowse\u002Fpackage-lock.json",{"path":74,"resourceType":60,"resourceTypeDesc":61,"sizeBytes":75,"sourceHtmlUrl":76},"package.json",340,"https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Fautobrowse\u002Fpackage.json",{"path":78,"resourceType":60,"resourceTypeDesc":61,"sizeBytes":79,"sourceHtmlUrl":80},"README.md",2874,"https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Fautobrowse\u002FREADME.md",{"path":82,"resourceType":60,"resourceTypeDesc":61,"sizeBytes":83,"sourceHtmlUrl":84},"REFERENCE.md",1797,"https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Fautobrowse\u002FREFERENCE.md",{"path":86,"resourceType":60,"resourceTypeDesc":61,"sizeBytes":87,"sourceHtmlUrl":88},"references\u002Fexample-skill.md",8569,"https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Fautobrowse\u002Freferences\u002Fexample-skill.md",{"path":90,"resourceType":60,"resourceTypeDesc":61,"sizeBytes":91,"sourceHtmlUrl":92},"references\u002Fexample-task.md",628,"https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Fautobrowse\u002Freferences\u002Fexample-task.md",{"path":94,"resourceType":95,"resourceTypeDesc":96,"sizeBytes":97,"sourceHtmlUrl":98},"scripts\u002Fevaluate.mjs","SCRIPT","脚本",22753,"https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Fautobrowse\u002Fscripts\u002Fevaluate.mjs",[100,108,116,125,132,140],{"skillNo":101,"name":102,"nameZh":6,"description":103,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":104,"sourceHtmlUrl":105,"categories":106,"keywords":107,"publishedAt":20},"01M7317QQ43D8KSNYCMQ9YSVV4","youtube-clipper","YouTube 视频智能剪辑工具。下载视频和字幕，AI 分析生成精细章节（几分钟级别）， 用户选择片段后自动剪辑、翻译字幕为中英双语、烧录字幕到视频，并生成总结文案。 使用场景：当用户需要剪辑 YouTube 视频、生成短视频片段、制作双语字幕版本时。 关键词：视频剪辑、YouTube、字幕翻译、双语字幕、视频下载、clip video","https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Fyoutube-clipper\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Fyoutube-clipper\u002FSKILL.md",[],[],{"skillNo":109,"name":110,"nameZh":6,"description":111,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":112,"sourceHtmlUrl":113,"categories":114,"keywords":115,"publishedAt":20},"01M7316R2JBWWT24DEVFGVZ1WW","yao-meta-skill","Create, refactor, evaluate, and package agent skills from workflows, prompts, transcripts, docs, or notes. Use when asked to create a skill, turn a repeated process into a reusable skill, improve an existing skill, add evals, or package a skill for team reuse.","https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Fyao-meta-skill\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Fyao-meta-skill\u002FSKILL.md",[],[],{"skillNo":117,"name":118,"nameZh":6,"description":119,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":120,"licenseFileUrl":121,"sourceHtmlUrl":122,"categories":123,"keywords":124,"publishedAt":20},"01M7314XDMF1VN8H6GE0RXWX3V","webapp-testing","Toolkit for interacting with and testing local web applications using Playwright. Supports verifying frontend functionality, debugging UI behavior, capturing browser screenshots, and viewing browser logs.","Apache-2.0","https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwebapp-testing\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwebapp-testing\u002FSKILL.md",[],[],{"skillNo":126,"name":127,"nameZh":6,"description":128,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":129,"categories":130,"keywords":131,"publishedAt":20},"01M73142X503419G1YEBQZVRWJ","vercel-react-best-practices","React and Next.js performance optimization guidelines from Vercel Engineering. This skill should be used when writing, reviewing, or refactoring React\u002FNext.js code to ensure optimal performance patterns. Triggers on tasks involving React components, Next.js pages, data fetching, bundle optimization, or performance improvements.","https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Fvercel-react-best-practices\u002FSKILL.md",[],[],{"skillNo":133,"name":134,"nameZh":6,"description":135,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":136,"sourceHtmlUrl":137,"categories":138,"keywords":139,"publishedAt":20},"01M7313E33KEXPQSJ6JBSW7CT5","ui-test","AI-powered adversarial UI testing via the browse CLI. Analyzes git diffs to test only what changed, or explores the full app to find bugs. Tests functional correctness, accessibility, responsive layout, and UX heuristics. Use when the user asks to test UI changes, QA a pull request, audit accessibility, or run exploratory testing. Supports local browser (localhost) and remote Browserbase (deployed sites).","https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Fui-test\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Fui-test\u002FSKILL.md",[],[],{"skillNo":141,"name":142,"nameZh":6,"description":143,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":120,"licenseFileUrl":6,"sourceHtmlUrl":144,"categories":145,"keywords":146,"publishedAt":20},"01M73123QNJHSG1XT7XJPQD0HK","technical-seo-checker","Use when the user asks to \"technical SEO audit\", \"check page speed\", \"crawl issues\", \"Core Web Vitals\", \"site indexing problems\", \"my site is slow\", \"Google cannot crawl my site\", \"mobile issues\", or \"indexing problems\". Performs technical SEO audits covering site speed, crawlability, indexability, mobile-friendliness, security, and structured data. Identifies technical issues preventing optimal search performance. For content and heading element issues, see on-page-seo-auditor. For link architecture, see internal-linking-optimizer.","https:\u002F\u002Fgithub.com\u002FJOHNadonis\u002Fclaude-code-skills\u002Fblob\u002Fmain\u002Fskills\u002Ftechnical-seo-checker\u002FSKILL.md",[],[]]