keppo-old

Review recent code changes and broad codebase areas for critical and high severity security vulnerabilities. Use when Codex needs to clear `./out-security-review`, select 25 non-doc, non-test files split across past-day commits, past-week commits, and broad codebase coverage, group them into 6-8 review buckets, spawn one security-review sub-agent per review bucket, persist each candidate finding to `./out-security-review/<starting_filename>_<timestamp>.md` using a filename-safe UTC ISO 8601 basic timestamp like `20260331T214512Z`, re-verify every candidate with fresh sub-agents, and emit final confirmed findings for responsible disclosure.

Record a short Playwright-based product demo video for significant UI or user-visible functionality changes. Use when Codex needs to show a workflow in motion for a PR, either by recording an existing targeted E2E spec or by creating a disposable one-off Playwright spec that captures the changed behavior in under a minute, then uploading the final video to Vercel Blob and preparing a PR comment body that says "Demo at commit {hash}" with a 1-2 sentence summary and durable hosted link.

Review the last 7 days of recent commits for large architecture improvements that dramatically improve maintainability. Use when the agent needs to clear `./out-code-architect`, select 25 non-doc, non-test files from recent commits across different parts of the repo, spawn one architecture-review sub-agent per starting file, persist each candidate finding to `./out-code-architect/<starting_filename>_<timestamp>.md` using a filename-safe UTC ISO 8601 basic timestamp like `20260331T214512Z`, re-verify every candidate with fresh sub-agents, and emit final confirmed findings for issue filing.

Review the last 7 days of recent commits for critical and high severity bugs that break user experience. Use when the agent needs to clear `./out-bug-finder`, select 25 non-doc, non-test files from recent commits across different parts of the repo, spawn one bug-review sub-agent per starting file, persist each candidate finding to `./out-bug-finder/<starting_filename>_<timestamp>.md` using a filename-safe UTC ISO 8601 basic timestamp like `20260331T214512Z`, re-verify every candidate with fresh sub-agents, and emit final confirmed findings for issue filing.

Review recent GitHub Actions workflow runs, separate expected noise from actionable failures, and write a machine-readable workflow health report for deterministic post-agent issue handling.

Print the full host-side `scripts/pr.sh` command needed to create or switch to a feature branch, commit current changes, push, and open or update a PR outside the Docker sandbox using explicit flag-style inputs.

Commit all working-tree and untracked changes in git, then push to the current branch using an automatically generated Conventional Commit message.

Commit all working-tree and untracked changes in git using an automatically generated Conventional Commit message.

Create an implementation plan in plans/ with clear goal, actionable checklists, and specific file references.

Automatically gather flaky E2E tests from recent CI runs on the main branch and from recent PRs by wwwillchen/wwwillchen-bot/keppo-bot, then deflake them.

Triage and fix targeted E2E failures or flakes using CI artifacts, traces, local repros, and the smallest real product, fixture, or harness change.

Investigate flaky GitHub Actions end-to-end test runs for the "E2E main" workflow on push events to main, cluster warnings, errors, and failures across runs, identify the dominant root causes, implement fixes in application or test infrastructure code, validate the result, and then hand off the ready branch to the pr-push skill when that skill is available.

Systematic UI screenshot critique. Analyzes interface screenshots and delivers specific, actionable feedback organized by visual design, interface design, interaction consistency, and user context.

Generate an image using the Dyad engine API and save it locally.

Run the full E2E suite and fix root-cause failures in application code (not test hacks).

Multi-agent code review system that spawns three independent Claude sub-agents to review PR diffs. Each agent receives files in different randomized order to reduce ordering bias. Issues are validated using reasoned analysis rather than simple vote counting. Reports merge verdict (YES / NOT SURE / NO). Automatically deduplicates against existing PR comments. Always posts a summary (even if no new issues), with low priority issues in a collapsible section.

Fix failing CI checks and GitHub Actions on a Pull Request.

Read all unresolved GitHub PR comments from trusted authors and address or resolve them appropriately.

Address all outstanding issues on a GitHub Pull Request by handling both review comments and failing CI checks.

Commit all working-tree and untracked changes on a feature branch, push the branch, and open a GitHub pull request with a Conventional Commit-style title, a concise summary, and a collapsed rationale section. If the current branch is `main`, create and switch to a feature branch first. When a PR already exists, fetch its current title and description, synthesize them with the new changes, and update the PR — never silently overwrite useful existing content. Use when Codex needs to finish the full git handoff flow after implementing changes and the user wants a branch-safe commit, push, and PR creation sequence.

Rebase the current branch on the latest upstream changes, resolve conflicts, and push.

Turn a bug description, feature request, rough idea, or issue summary into a short implementation prompt with exact file references and a precise description of what to change. Use when the user wants a tighter prompt for another agent or coding pass and values brevity over planning prose.

Make existing code simpler to read, understand, and maintain without changing functionality. By default, target the current uncommitted code unless the user points to a different file, diff, or area. Use when the user asks to simplify, clean up, refactor for clarity, reduce duplication, or improve code structure while preserving behavior.

Turborepo monorepo build system guidance. Triggers on: turbo.json, task pipelines, dependsOn, caching, remote cache, the "turbo" CLI, --filter, --affected, CI optimization, environment variables, internal packages, monorepo structure/best practices, and boundaries. Use when user: configures tasks/workflows/pipelines, creates packages, sets up monorepo, shares code between apps, runs changed/affected packages, debugs cache, or has apps/packages directories.