Agent
Read-only adversarial reviewer for code, PRDs, RFCs, design docs, and other deliverables. Inspects the full pinned scope, reports every BLOCKER, QUESTION, and NOTE in one batch with stable finding IDs, and never edits the artifact. Used by the adversarial-review skill as the isolated Reviewer role.
Skill Claude CodeCodex
Re-explains and re-verifies prior assistant messages with evidence-backed explanations, confidence levels, and ASCII diagrams. Use when the user asks to clarify, justify, get on the same page, or understand the reasoning behind a previous answer.
Skill Claude CodeCodex
Use when the user wants to render Mermaid diagrams as SVG, generate polished diagram visuals, create flowcharts, sequence diagrams, state diagrams, class diagrams, ER diagrams, or XY charts. Use for diagram visualization, theming, or presentation-quality diagram output.
Skill Claude CodeCodex
Explores user intent and design options before implementing a new feature, component, or behavioral change. Use when the user wants to design before coding, requirements are unclear, or mentions brainstorming, ideation, or exploring approaches.
Skill Claude CodeCodex
Diagnose and fix failing tests, CI failures, red builds, pipeline failures, crashes, exceptions, stack traces, timeouts, flaky behavior, regressions, or bug reports. Use when the user reports a failure, asks to fix a bug, asks why tests fail, or asks to make CI pass.
Skill Claude CodeCodex
Runs an uncapped Coordinator-mediated Synthesizer–Challenger loop that develops a comprehensive answer through full-batch internal discussion, asking users only for key choices. Use only when the user explicitly requests iterative adversarial or multi-perspective synthesis of a question, topic, idea, decision, or…
Agent
Independent challenger in a deliberate loop. Reports every visible error, gap, counterpoint, risk, trade-off, and inconsistency in one batch with stable D-IDs, and rechecks every prior ID and topic each pass. Never produces the answer. Used by the deliberate skill as the isolated Challenger role.
Agent
Produces the complete answer in a deliberate loop. Covers the full topic bundle each pass — conclusions, alternatives, uncertainty, and cross-topic effects — and revises the whole answer in one batch when challenged. Never challenges. Used by the deliberate skill as the isolated Synthesizer role.
Skill Claude CodeCodex
Explain a topic like I'm a 5 year old. Use when the user says eli5 with a topic, or asks for a dead-simple picture explainer of how something works.
Skill Claude CodeCodex
Resolve an in-progress Git merge or rebase conflict. Use when Git reports unmerged paths or cannot continue because of conflicts.
Skill Claude CodeCodex
Route an ambiguous coding task to indexed Git repositories using human-maintained business responsibilities, page snapshots, and focused CodeGraph lookup. Use when a task does not name one unambiguous repository, may span repositories, or the user asks to list, query, add, update, or validate locally indexed projects.
Skill Claude CodeCodex
Build a lightweight map and project glossary before exploring an unknown repository or unfamiliar module. Use for quick orientation, entry-point discovery, broad code exploration, feature work in a new codebase, or shared vocabulary for key concepts, types, responsibilities, and call relationships. Do not use for a…
Skill Claude CodeCodex
Provides brief or detailed, source-backed in-chat understanding of any target. Use for TL;DRs, quick overviews, exploration, in-depth explanations, or detailed reports. Do not use for background or durable-file research, reviews, audits, recommendations, tutorials, plans, implementation, or repository mapping.
Skill Claude CodeCodex
Use when the user wants app-icon artwork, a transparent 1024×1024 PNG, Chrome extension icons, or platform icon materials. Offers 10 styles, generates a green-screen source, waits for approval, removes it with Node.js sharp, and validates outputs. Exclude general images, vector-system edits, and…
Skill Claude CodeCodex
Parses a Figma URL and generates a structured text description of the UI layer architecture. Use when the user provides a Figma URL and asks to parse, analyze, describe, or extract design information.
Skill Claude CodeCodex
Manages Zhihu iOS component dependencies and builds via venom-cli. Use when the user mentions Zhihu iOS development, venom-cli, component dependency management, or asks to build/check/make/switch/integrate components in a Zhihu project.
Skill Claude CodeCodex
Create first-time Zhihu Yunxiao One MRs; list open MRs or QA; batch-approve requests; and manage the macOS approval daemon and dashboard. Use for Yunxiao MR creation, listing, QA, approval, or daemon management. Exclude metadata edits, code review, join/restore, and GitHub or GitLab PRs.
Skill Claude CodeCodex
Create, inspect, update, enable, disable, delete, and audit persistent Agent Hooks rules that inject prompts or run scripts at supported hook timings. Use for cross-session user directives, reusable changed-file or command matching, runtime guards, hook-like automation, or trigger inventory and management for Codex…
Skill Claude CodeCodex
Manage persistent Agent Rules. Use when the user wants to add, edit, list, or remove an unconditional rule, change a bundled default rule, or inspect how built-in, user, and project rules are injected. Use Agent Hooks for conditional or scripted automation instead.
Skill Claude CodeCodex
Use when the user wants to list, create, inspect, or follow SOPs, procedures, playbooks, runbooks, workflow rules, or process documents. Also use when the user mentions "SOP", "procedure", "playbook", "runbook", "process doc", or "per our SOP".
Skill Claude CodeCodex
Archives an exact user-selected range of visible User and Agent messages from the current Pi session into tasks/conversations/. Use for /archive or explicit requests to preserve part of the current conversation verbatim. Do not use for summaries, task state, implementation guidance, or external transcripts.
Skill Claude CodeCodex
Validate one or all Agent Skill packages with CSL Agent Kit's deterministic local quality gate. Use after creating or modifying a skill package, or when asked to check Skill frontmatter, resources, context budget, or routing fixtures. Do not use for non-skill rules, subjective portfolio analysis, or runtime project…
Skill Claude CodeCodex
Load, establish, and maintain dispatch-ready workspace context in tasks/context.md. Use at session start, after resume or compaction, when Context is missing or lacks a valid Project Core, before a concrete task needs project orientation, and before ending when confirmed durable project facts changed. Ensure a…
Skill Claude CodeCodex
Grill the user relentlessly about a plan, decision, or topic with task-aware recording. Use when the user wants to stress-test thinking and wants the outcomes handled by the canonical task system instead of polluting an existing task record.