Grill With Docs
Ground a plan in repo docs, code, Obsidian notes, glossary, ADRs, specs, and tickets, then run grilling to stress-test the decisions one question at a time.
Session Recall
Find prior local Codex or Claude agent sessions with the agent-session-find CLI. Use when context may exist in previous sessions, after compaction or handoff, before asking the user to remember prior work, when searching for old fixes, decisions, commands, errors, file paths, repo context, review outcomes, or "where did we handle this before?" Keep session contents local and use low-token search…
Monitoring Gh Actions
Use when monitoring ongoing GitHub Actions workflow runs or pull request checks through GitHub CLI and needing history-aware waiting that avoids token-heavy polling and rate limits.
Review Surface Map
Map changed runtime surfaces for a PR, branch, commit, or large diff: entrypoints, flows, callers, contracts, side effects, risk, and validation.
Test Audit
Audit or design tests: related coverage, changed tests, missing tests, no-test-needed decisions, useless tests, stale API assertions, brittle mocks, impossible states, and tests to remove.
Reducing Cognitive Load
Review dense, clever, stringly typed, weakly typed, over-abstracted, or hard-to-read code, especially data flows and type boundaries.
Review Until Clean
Run the harness-native review (bare `codex review` in Codex, the built-in code-review workflow in Claude Code) in a fix-and-rerun loop until clean; also used as code-review''s first phase.
Improve Codebase Architecture
Review or plan architecture changes around module depth, interfaces, adapters, locality, dependency direction, testability, and durable architecture notes.
Rag Anything Mcp
An MCP server for rag-anything (A multi-modal RAG system)
Pr Rubbish Audit
Audit messy PR or branch diffs for unrelated rubbish: branch-history artifacts, noisy comments, accidental deletions, generated drift, and unneeded refactors.
Review Guardrails
Bound autonomous review loops with wall-clock, diff-growth, scope, consult-queue, provisional-fix, and tracked-finding rules.
Typescript Discipline
Write or review TypeScript production code, shared domain types, schemas, API/client contracts, exported functions, typed React, casts, any/unknown, ts-ignore, narrowing, and package-script verification.
Skill Cleaner
Audit agent skills: loaded roots, duplicate skills, unused skills, prompt-budget costs, compact descriptions.
Finding Discipline
Filter code-review observations into actionable findings for PR, cold, security, and noisy-diff reviews; remove style nits and vague risks.
Openclaw Local Test
Spin up and manage a local OpenClaw manual-test Gateway in a browser, with optional local Atlassian AIGW, Slack, Discord, and fresh ports.
Html Explanations
Create standalone HTML for complex explanations, code/diff walkthroughs, plans, reports, handoffs, diagrams, research notes, and concepts.
To Tickets
Break a plan, spec, or conversation into tracer-bullet tickets with blocking edges, then publish them as Obsidian Issues.
Wait Efficiently
Wait without token-heavy polling. Use when the user asks to wait, pause, check back later, babysit a command, monitor a long-running process, wait for CI or GitHub Actions, or avoid repeated status polling while external work completes.
Frontend Design
Design and implement production frontend UI: pages, screens, components, layout, hierarchy, typography, color, states, responsiveness, design systems, and polish. Use for broad visual work; route motion to design-engineering.
Review Animations
Review changed animation, transition, gesture, spring, or motion code for purpose, timing, easing, coherence, interruption, performance, reduced motion, and input behavior. Use as a focused code-review lens.
Clawsweeper Until Clean
Drive Clawsweeper PR re-reviews in a fix-and-rerun loop: comment `/clawsweeper re-review`, monitor, fix findings, and stop only when clean.
Wayfinder
Plan a huge chunk of work, more than one agent session can hold, as a shared Obsidian issue map, and resolve tickets one at a time until the way to the destination is clear.
Cold Pr Review
Run an independent cold review of a PR, branch, or major feature using changed-flow mapping, relevant review lenses, and finding-discipline.
Clawhub Local Test
Start and manage a local ClawHub manual-test instance backed by a development Convex database refreshed from production. Use when testing ClawHub UI, moderation, API, search, publisher, package, skill, or Convex-backed behavior locally with prod-like data instead of hand-running setup, prod export, dev/local import, dev auth fixture, and app server commands.
Tdd
Test-driven development. Use when the user wants to build features or fix bugs test-first, mentions red-green, or wants integration tests.
Research
Investigate a question against high-trust primary sources and capture cited findings in a Markdown note. Use when the user wants docs, API facts, standards, source-code behavior, or implementation context researched before planning or building.
Cold Pr Review Until Clean
Run independent cold-review subagents on a PR, branch, or diff, fix actionable findings, and repeat until the configured clean stop condition is met.
Code Review
Review a PR, branch, commit, or diff with a two-phase until-clean workflow: map changed flows, run native review, run cold review, fix findings, and verify.
Parallel Slice Orchestration
Implement an existing PRD, slice plan, or feature plan with parallel subagents, disjoint slice ownership, integration, and verification.
Frontend Ui Validation
Validate rendered frontend UI with Playwright screenshots, bounding boxes, console checks, layout audits, responsive states, interaction states, and Figma or reference comparisons. Use after visible web UI changes; use repo-owned native proof for Expo and React Native screens.
Grilling
Grill the user relentlessly about a plan or design. Use when the user wants to stress-test a plan before building, or uses any grill trigger phrase.
Acpx Frontend Delegation
Delegate substantial Codex frontend work to Claude Code through acpx. Use when running in Codex and the task includes visual design direction, dense CSS/layout work, responsive UI polish, ambiguous frontend implementation, or the user explicitly asks to use Claude/acpx for frontend work.
To Spec
Turn the current conversation into a spec and publish it to Obsidian Specs. Use after grilling has already resolved the important decisions.
Pr Proof Pack
Create or refresh reviewer-checkable PR proof when creating a pull request, updating a PR body, modifying proof, or changing a branch after proof was written. Use for net diff summaries, PR-visible screenshots, Mermaid diagrams, API examples, behavioral verification, and proof cleanup.
Design Engineering
Implement or refine UI motion and interaction: animations, transitions, micro-interactions, component feel, popovers, drawers, sheets, toasts, drag or swipe gestures, springs, easing, timing, and performance. Use for responsive, physically coherent behavior.
Diagnose
Debug bugs, failing tests, regressions, flaky behavior, unexpected output, or broken behavior with a reproducible evidence loop.
Handoff
Compact the current conversation into a handoff document for another agent to pick up. Prefer a fresh session/thread for standalone, parallel, background, or aside work; fork only when the next agent truly needs the raw current conversation history.
Writing Great Skills
Reference for writing and editing skills well: invocation choices, information hierarchy, progressive disclosure, leading words, pruning, and failure modes.