ai-tools

Reply to GitHub PR review threads as the PR author -- someone reviewed your PR and you want to respond (and optionally fix code). Classifies each thread, confirms the plan, then posts replies immediately. Does not create pending reviews. Use when the user says "reply to comments", "respond to review", or invokes /gh-reply-comments. For reviewer-side follow-up after the author pushes fixes, use pr-follow-up instead.

Reviewer-side PR follow-up: after the PR author pushes fixes and replies, validate whether each thread you started was addressed (or reasonably justified), then reply and resolve the good threads in a single batch. Use when the user says "validate the fixes", "did they address my review", "check if fixes addressed my review", or invokes /pr-follow-up.

Staff-engineer-level PR review that fans out to pr-review-toolkit and pr-review-local in parallel, audits automated and human reviewer comments, runs a judge layer that filters false positives, dedups overlapping findings, and scores confidence, then presents structured findings by severity. Use when the user says "staff review", invokes /staff-review, or wants a thorough review that also validates existing reviewer comments. Pass --no-judge to skip the judge layer. Do not use for simple code reviews or when the user explicitly asks for pr-review-toolkit directly.

Append a task item to today's Obsidian daily note from a PR URL or Jira ticket. Activate when the owner asks to append, add, log, or track an item in their daily tasks -- especially when they drop a GitHub PR URL or a Jira ticket key (like PROJ-123) with intent to track it, or to log any free-text task. Trigger phrases include "append this PR to my daily tasks", "add to daily", "log this PR for review", "track this ticket", "add this to my tasks", "add a task to my daily note", "log this to today". Do not trigger for general PR review requests that have no tracking intent.

Interactive triage and posting of PR review comments as a pending GitHub review. Iterates findings one by one with send/edit/skip actions, posts as pending review via gh api. Also accepts a pre-triaged JSON via --from <path> from /review-board, or decided findings from the review-harness DB via --from-db, both skipping the interactive loop. Use when the user says "post review", "post comments", invokes /post-review, or after completing any PR review when there are comments to post.

Migrate one file or route from the deprecated antdV3 import (npm alias antd@3.x) to antd v6, proving no visual or behavior regression with before/after Playwright screenshots of the real route before committing. Use when a Sonar no-restricted-imports MAJOR flags an antdV3 import, or the owner asks to migrate a component to antd v6. Trigger phrases include "migrate antdV3", "antd v6 migration", "migrate this component to antd v6", "kill the antdV3 import". Reusable across projects, parameterized by target file, repo path, route, and storageState filename. Captures the loading spinner, the migrated components, and a behavior check, then stops at a ready-to-commit signed-commit command for the owner to run. Scoped to one target only, never a repo-wide sweep. Not for non-antd refactors, dependency bumps of antd itself, or migrations with no rendered route to screenshot.

Fetch a Jira bug ticket, derive suspected root causes, fan out parallel read-only agents across the codebase, then synthesize a ranked root-cause report. Use when the owner pastes a Jira key or browse URL and asks to investigate a bug's root cause, send agents to investigate a ticket-linked bug, find where a bug originates in code, or trace a reported defect to its source layers. Trigger phrases include "investigate-with-jira", "investigate this ticket", "send agents to investigate", "find the root cause of this bug", "where does this bug come from". Investigation only, never edits code and never posts back to Jira. Takes a required ticket (key or URL), optional free-text context or hypothesis, and an optional repo path that defaults to the current directory. Not for fixing bugs, validating fixes on stage, or tickets with no code-investigation angle.

Prove a merged bug fix holds by driving a real browser against a stage environment with Playwright, capturing video and network evidence, then posting a verification comment to the Jira ticket. Use when the owner asks to validate or verify a bug fix on stage, reproduce a bug scenario in a real browser and prove the fix works, produce video or network evidence that a fix holds, or confirm a stage build resolved a ticket before sign-off. Trigger phrases include "validate this bug fix", "verify the fix on stage", "prove the fix holds", "reproduce and verify on stage", "browser evidence the fix works", "drive Playwright to confirm the fix". Parameterized by ticket, stage URL, target data, and assertion, so it works for any bug, not just one flow. Not for unit tests, local-only checks, or fixes that are not yet on a stage build.

Push current branch and create or update a GitHub PR with an intent-driven description. Use when the user says "ship", "ship it", "create PR", "push and PR", or wants to push their branch and open a pull request. Runs pre-flight safety checks, pushes to remote, and generates PR title/body from branch name and commit history. Detects stacked PRs via gh-stack or base-chain walk and offers to include a Stack section in the body. Supports --draft and --ready flags to skip the draft/ready prompt, and --base to override the target branch.

Obsidian knowledge architect. Produces complete, copy-paste-ready Obsidian outputs: notes with YAML frontmatter, valid .canvas JSON, valid .base YAML, folder structures as tree + bash script, CSS snippets, URI schemes, Tasks plugin queries. Trigger whenever the user mentions: Obsidian, vault, canvas, base, wikilinks, PKM, second brain, daily notes, note template, MOC, Map of Content, backlinks, note structure, folder structure, knowledge management, markdown note, Obsidian plugin, graph view, properties, frontmatter, callout, embed, slash commands, obsidian publish, web clipper, obsidian URI, obsidian CLI, obsidian sync, workspaces, Tasks plugin. Also trigger for: "organize my notes", "build a vault", "create a note system", "make a canvas", "set up a base", "import my notes". Do not withhold output while explaining. Produce the artifact first, then add any brief notes.

Post-merge cleanup after a PR ships. Verifies the branch's PR is merged on remote, switches to the default branch, pulls, and prompts for safe local branch deletion. Use whenever the user says "PR merged", "merged", "merged it", "done", "post-merge cleanup", or invokes /post-merged right after shipping a PR, even if they don't explicitly ask for cleanup. Trust `gh pr view` state=MERGED as authoritative (handles squash/rebase merges correctly). Guardrails - never force-delete, never push/reset/clean, aborts on dirty tree.

End-of-session command that updates auto-memory, optionally writes session knowledge to Basic Memory, and creates an Obsidian session note via a .notes/ symlink. Use when the user says "wrap up", "end session", "finish up", "done for the day", invokes /wrap-up, or is clearly finishing a work session. Also use after completing a major task when the user wants to capture what happened. If in doubt, offer to run it.

Run the 3-subagent PR review (Security, Regression, Performance) against your current branch diff vs the resolved base branch, locally, before pushing. Use when the user says "review my branch", "local PR review", "review before push", "pre-ship review", or invokes /pr-review-local. Does not post to any PR. Output is stdout markdown only.

Multi-agent PR reviewer. Use ONLY when explicitly asked to review a pull request: "review PR #N", "review this PR", "code review". Do NOT trigger automatically during coding or general questions.

Scaffolds a Claude Code /review GitHub Actions workflow into the current repo. Use when the user says "set up Claude PR review", "install PR review workflow", "add /review to this repo", "scaffold PR review", or invokes /claude-pr-review-setup. Supports flags `--upgrade-runtime` (overwrite only the runtime skill) and `--force` (overwrite all three installed files). Always use this skill when the user wants automated PR review via GitHub Actions in a new repo. Do not trigger automatically.

Scaffold agent team infrastructure (Helen + Jess, SQLite state, inbox/outbox, CLAUDE.md injection) into any project. Use when the user says "init team", "set up agent team", "bootstrap agent infrastructure", "add Helen and Jess", or invokes /init-team. Always use this skill when the user wants to set up a multi-agent team in a project, even if they don't use those exact words.

Quick-capture and manage project todos during work sessions. Subcommands: add, list, done, remove, help. Todos are stored as markdown files in .todos/ at the project root. Use when the user says "add todo", "show todos", "mark done", "remove todo", "todo help", or invokes /todo:add, /todo:list, /todo:done, /todo:remove, /todo:help.