shortcuts

Comprehensive reference for CoC (Copilot of Copilot) architecture — server, memory system, LLM tools, SDK wrapper, process store, workflow engine, deep-wiki, and dashboard SPA. Use whenever the user asks anything related to CoC (Copilot of Copilot), forge, deep-wiki, coc-client, the CoC server, dashboard, workflow engine, memory system, LLM tools, process store, or any of their subsystems. Also use whenever you need background knowledge to answer a question, plan a change, or ground your reasoning in this codebase — read the relevant reference file(s) under `references/` before responding. Use when building, modifying, debugging, or explaining CoC features, or when understanding how subsystems interact.

Submit a single commit or a range of commits as a new GitHub pull request. Cherry-picks the commits onto a fresh branch off the latest origin/main, pushes, opens the PR via `gh`, and enables auto-merge by default. On any cherry-pick/rebase conflict the script aborts the entire submit; the AI must NOT attempt to resolve conflicts. Use when the user asks to "open a PR for this commit", "send commits X..Y as a PR", "PR-ify these commits", or similar.

Schedule recurring follow-up messages into the current conversation. Supports fixed-interval monitoring and one-shot wakeups for dynamic self-pacing.

MUST BE LOADED for any code change in this repository. Implements the requested code change and adds comprehensive test coverage, ensuring tests pass across Linux, macOS, and Windows. Use whenever you are about to modify, add, refactor, fix, or delete source code in this repo — including bug fixes, feature work, refactors, and small tweaks. Use when you need to execute implementation work (not just planning) and must include tests. Commit the changes to the repository after the implementation is complete.

Delegate a job from the current chat to a new conversation. Use when the user asks to delegate or hand off work.

Draft a user experience specification for a requested CoC feature. Use when planning new features, designing user flows, or creating UX specs for the CoC dashboard, CLI, server, or Node packages.

Generate optimized YAML pipeline configurations from natural language requirements, or DAG workflow configurations for complex multi-stage processing with fan-out, fan-in, and scripting. Ask clarifying questions about task type, data source, and output format before creating pipelines. Use when users want to create, design, or build a pipeline or workflow for data processing, classification, research, or analysis tasks.

Generate optimized YAML pipeline configurations from natural language requirements, or DAG workflow configurations for complex multi-stage processing with fan-out, fan-in, and scripting. Ask clarifying questions about task type, data source, and output format before creating pipelines. Use when users want to create, design, or build a pipeline or workflow for data processing, classification, research, or analysis tasks.

System prompts for the CoC Dream analyzer and critic internal steps — proposes and validates high-confidence dream card candidates from completed workspace conversations.

Core instruction sets for Ralph autonomous coding loop phases — grill (clarification), synthesis (goal extraction), execution (iteration), iteration (user prompt), and final-check (validation).

Classify every hunk in a pull request or commit diff by change type (logic, mechanical, test, simple, generated) so reviewers can focus on what matters.

Debug and verify web UI issues by opening pages in a real browser, checking for errors, applying code fixes, rebuilding, and re-testing. Restart the coc dev server ONLY when the user explicitly asks for a restart — otherwise stop and ask for confirmation before restarting. Use when the user asks to debug a web page, verify a UI fix in the browser, or test a localhost page after code changes.

Harden a skill from one failure case without overfitting — name the failure class, derive an abstract rule, validate it against three cases, then emit a minimal diff.

Interview the user about a plan or design and produce an autonomy-ready spec artifact. Use when the user wants to stress-test a plan, get grilled on a design, mentions "grill me", or when promoting an ask-mode chat into a Ralph loop.

Generate, read, and iteratively modify Excalidraw diagrams from natural language descriptions. Use when asked to "create a diagram", "make a flowchart", "visualize a process", "draw a system architecture", "create a mind map", or "iterate on an existing diagram". Supports flowcharts, relationship diagrams, mind maps, system architecture, data flow, swimlane, class, sequence, and ER diagrams. Produces interactive previews inline in chat via `excalidraw://` links.

Access, search, analyze, and submit CoC conversation process records via REST API to a running CoC server (chat, run-workflow, run-script, follow-up, stream, search, list, stats). Use when the user asks to find, read, search, summarize, analyze, submit, or send tasks to CoC.

Identify test coverage gaps in a given code area. Supports three modes: unit-test (pure logic), functional-test (integration with real I/O), and mock-e2e (mock-based end-to-end). Use when you want to find missing tests, audit test coverage, or plan new test work for a package or module.

Process a list of items by dispatching ONE sub-agent per item, strictly sequentially (await each before the next). Same per-item sub-task and final summary contract as map-reduce — only the dispatch order differs. Only invoke when the user explicitly asks for the for-each skill or pattern by name. Do not auto-select for general tasks.

Process a list of items by dispatching one sub-agent per item in parallel (up to max_parallel concurrent), then aggregate results into a final summary. Same per-item sub-task and final summary contract as for-each — only the dispatch order differs. Only invoke when the user explicitly asks for the map-reduce skill or pattern by name. Do not auto-select for general tasks.

Synthesizes rough software engineering manager 1:1 notes into themes, follow-ups, commitments, risks, coaching signals, and next-meeting prep. Use in the My Work workspace when working with 1:1 notes or people-management notes.

Ultra-compressed reply mode. Cuts token usage ~50% while keeping full technical accuracy. Use when user asks to "be brief", "be terse", "less tokens", "compact mode", or invokes /terse. Auto-triggers on explicit token-efficiency requests.

Distill recent CoC chat histories into knowledge-base skill improvements. Scans new conversations since the last run, proposes additions/updates/removals, and patches skill files on confirmation. Use when you want to refresh or update a knowledge-base skill from past chats.

Rewrite documents, plans, and notes as if authored fresh each iteration — produce only the final intended state, never patch deltas on top of the previous version. Use whenever you update an existing document, plan, design note, AGENTS.md, README, MEMORY.md, or any persistent text artifact.

Review a bug fix proposal and evaluate whether it is the cleanest solution. Use when you want to critically assess a fix for root cause alignment, simplicity, consistency, technical debt, side effects, and idiomatic alternatives. If improvements exist, propose a better fix and explain why.

Automated code refactoring suggestion. Use when reviewing a commit range or code area to propose critical, high-value, high-confidence refactorings that must be addressed immediately to avoid technical debt. This skill drafts a refactoring plan instead of making direct code changes.

Advanced research and verification methodologies using multi-phase approaches and parallel sub-agents. Use for deep research on complex topics, multi-model verification of arguments, or decomposing features into atomic commit plans.

Create and update Agent Skills following the agentskills.io specification. Use when creating new skills, updating existing skills, or validating skill structure.

Advanced research and verification methodologies using multi-phase approaches and parallel sub-agents. Use for deep research on complex topics, multi-model verification of arguments, or decomposing features into atomic commit plans.

Review commits, pull requests, diffs, or files against repository-specific standards by selecting relevant rule references and delegating each rule to a focused code-review subagent.

Create and update Agent Skills following the agentskills.io specification. Use when creating new skills, updating existing skills, or validating skill structure.

Update AGENTS.md files across the repository to reflect the latest codebase status and create missing ones for critical folders. Uses sub-agents to explore and write in parallel, preserving main context. Use when AGENTS.md files are outdated or missing.