feat: improve reproduce-bug skill, sync agent-browser, clean up redundant skills#333
feat: improve reproduce-bug skill, sync agent-browser, clean up redundant skills#333
Conversation
…workflow Address reviewer feedback: remove Rails/AppSignal/Playwright-specific tool calls, add proper issue fetching via gh, add hypothesis phase before investigation, separate UI/backend/manual reproduction routes, require user confirmation before posting to GitHub issues.
…hecklists Search related code before analyzing specs so gaps are informed by what actually exists. Replace generic QA checklist approach with spec-driven analysis. Organize output by severity instead of category. Remove persona fluff.
Replace local copy with exact upstream from vercel-labs/agent-browser. Add upstream-sourced skills section to AGENTS.md documenting that agent-browser is vendored locally and should not have local modifications.
…arallel - create-agent-skill: thin shim that just forwarded to create-agent-skills - create-agent-skills: duplicates official skill-creator; plugin-specific guidance already lives in AGENTS.md compliance checklist - resolve_parallel: legacy skill using deprecated TodoWrite and unqualified agent names, replaced by resolve-todo-parallel Update README tables and best-practices-researcher agent reference.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: a08ca16284
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Expand the investigation checklist with specific things to look for: database state (missing associations, constraint violations) and request/response cycle (status codes, params, middleware behavior). Tighten existing items with actionable details like stack traces, failed network requests, and CORS issues.
…iggers Narrow skill description to only match when a GitHub issue is provided, and add a guard that asks the user for an issue number before calling gh issue view with an empty argument. Addresses P1 review feedback from PR #333.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 31494959bf
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| name: agent-browser | ||
| description: Browser automation using Vercel's agent-browser CLI. Use when you need to interact with web pages, fill forms, take screenshots, or scrape data. Alternative to Playwright MCP - uses Bash commands with ref-based element selection. Triggers on "browse website", "fill form", "click button", "take screenshot", "scrape page", "web automation". | ||
| description: Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction. | ||
| allowed-tools: Bash(npx agent-browser:*), Bash(agent-browser:*) |
There was a problem hiding this comment.
agent-browser workflows
Fresh evidence beyond the earlier install-only thread: this synced body now recommends normal runtime commands like echo "$PASSWORD" | agent-browser auth save ... and echo '[...]' | agent-browser batch --json (SKILL.md lines 89 and 191-199). With allowed-tools limited to Bash(npx agent-browser:*) and Bash(agent-browser:*), those flows start with echo or depend on shell stdin plumbing, so Claude Code blocks them even when agent-browser is already installed. As written, two advertised day-to-day workflows dead-end under the default permission model.
Useful? React with 👍 / 👎.
| Navigate to the affected area and execute the steps from the issue: | ||
|
|
||
| ```bash | ||
| agent-browser open "http://localhost:${PORT}/[affected_route]" |
There was a problem hiding this comment.
Route B verifies the server with http://localhost:${PORT:-3000}, but the very next navigation template drops that fallback and uses ${PORT} directly. In the common case where the app is running on the default local port without exporting PORT, this expands to http://localhost:/[affected_route], so the browser reproduction path breaks as soon as it tries to leave /. Reusing the same default (or telling the agent to substitute the discovered port explicitly) would keep the default setup working.
Useful? React with 👍 / 👎.
1 participant
Summary
vercel-labs/agent-browser. Add upstream-sourced skills policy to AGENTS.md.create-agent-skill(thin shim),create-agent-skills(duplicates officialskill-creator), andresolve_parallel(legacy, replaced byresolve-todo-parallel). Update README and agent references.Test plan
reproduce-bugskill loads correctly and frontmatter is validspec-flow-analyzeragent dispatches correctly fromce-planagent-browserskill matches upstream contentcreate-agent-skill,create-agent-skills,resolve_parallel)bun run release:validateto check plugin/marketplace consistency