coding-friend

Dispatch code review to a subagent. Use when the user wants code reviewed — e.g. "review this", "review my changes", "check the code", "look over this", "code review", "any issues with this?", "is this code ok?", "review before merge", "review the diff", "what do you think of these changes?". Also triggers on requests to review specific files, commits, or branches.

Dispatch code review to a subagent. Use when the user wants code reviewed — e.g. "review this", "review my changes", "check the code", "look over this", "code review", "any issues with this?", "is this code ok?", "review before merge", "review the diff", "what do you think of these changes?". Also triggers on requests to review specific files, commits, or branches.

Save a concise checkpoint of the current conversation's context — goal, decisions, breaking changes, next steps — into a file for resuming later in a DIFFERENT conversation. Use when the user says things like "save a checkpoint", "checkpoint this conversation", "capture context", "snapshot the conversation", "cf-checkpoint". Unlike /compact (condenses context within the SAME conversation), this produces a durable, resumable snapshot for a fresh conversation. Unlike $cf-remember (durable project-wide facts/conventions for AI recall across all future work), this is a transient per-conversation resume artifact. Does NOT auto-invoke — slash-only.

Brainstorm and write an implementation plan. Use when the user wants to plan, build, create, or implement something — e.g. "let's build", "let's create", "I want to create", "create for me", "build for me", "add feature", "implement", "make a", "set up", "I need a", "can you build", "help me build", "how should we implement", "design a solution", "architect", "scaffold", "plan out", "figure out how to", "what's the best way to build". Also triggers on task descriptions that imply multi-step implementation work requiring upfront planning.

In-depth research on a topic with web search and structured output. Use when the user wants deep research before building — e.g. "research this", "look into this library", "investigate how X works", "compare these options", "what are the best practices for", "deep dive into", "study this technology", "analyze this repo", "explore the ecosystem around". Also triggers when the user needs to understand a technology, library, or architecture pattern in depth before making decisions.

Save a concise checkpoint of the current conversation's context — goal, decisions, breaking changes, next steps — into a file for resuming later in a DIFFERENT conversation. Use when the user says things like "save a checkpoint", "checkpoint this conversation", "capture context", "snapshot the conversation", "cf-checkpoint". Unlike /compact (condenses context within the SAME conversation), this produces a durable, resumable snapshot for a fresh conversation. Unlike /cf-remember (durable project-wide facts/conventions for AI recall across all future work), this is a transient per-conversation resume artifact. Does NOT auto-invoke — slash-only.

Brainstorm and write an implementation plan. Use when the user wants to plan, build, create, or implement something — e.g. "let's build", "let's create", "I want to create", "create for me", "build for me", "add feature", "implement", "make a", "set up", "I need a", "can you build", "help me build", "how should we implement", "design a solution", "architect", "scaffold", "plan out", "figure out how to", "what's the best way to build". Also triggers on task descriptions that imply multi-step implementation work requiring upfront planning.

In-depth research on a topic with web search and structured output. Use when the user wants deep research before building — e.g. "research this", "look into this library", "investigate how X works", "compare these options", "what are the best practices for", "deep dive into", "study this technology", "analyze this repo", "explore the ecosystem around". Also triggers when the user needs to understand a technology, library, or architecture pattern in depth before making decisions.

Answer questions about Coding Friend — skills, agents, workflows, setup. Auto-invoke when the user asks about Coding Friend capabilities, available skills, how to use a skill, what skills/agents/commands exist, how it works, or how to configure it — e.g. "what skills are available?", "how does coding friend work?", "what can you do?", "list all skills", "what agents exist?", "how do I use cf-plan?", "what is cf-tdd?", "what does cf-fix do?", "which skill should I use?", "how do I get started?", "coding friend features/setup", "does X require the CLI?", "what works without coding-friend-cli?", "is the CLI required?". Do NOT auto-invoke for general coding questions unrelated to Coding Friend itself.

Generate a review prompt for an external AI agent. Creates a self-contained document that any AI (Gemini, Codex, ChatGPT, or even a human) can read and use to review your code changes. Use when the user wants an outside review — e.g. "get a second opinion", "external review", "review out", "send for review", "cf-review-out", "outside review", "prepare review for gemini", "review prompt".

Answer questions about Coding Friend — skills, agents, workflows, setup. Auto-invoke when the user asks about Coding Friend capabilities, available skills, how to use a skill, what skills/agents/commands exist, how it works, or how to configure it — e.g. "what skills are available?", "how does coding friend work?", "what can you do?", "list all skills", "what agents exist?", "how do I use cf-plan?", "what is cf-tdd?", "what does cf-fix do?", "which skill should I use?", "how do I get started?", "coding friend features/setup", "does X require the CLI?", "what works without coding-friend-cli?", "is the CLI required?". Do NOT auto-invoke for general coding questions unrelated to Coding Friend itself.

Generate a review prompt for an external AI agent. Creates a self-contained document that any AI (Gemini, Codex, ChatGPT, or even a human) can read and use to review your code changes. Use when the user wants an outside review — e.g. "get a second opinion", "external review", "review out", "send for review", "cf-review-out", "outside review", "prepare review for gemini", "review prompt".

Quick Q&A about codebase — explores code to answer, saves to docs/memory. Use when the user asks a focused question about the project — e.g. "how does X work?", "where is Y defined?", "what's the flow for Z?", "explain this module", "how are these connected?", "what pattern does this use?", "why is this done this way?". Unlike /cf-research (deep multi-doc output), this gives a single focused answer.

Load a previously saved checkpoint file back into a fresh conversation so work can continue with the prior context. Use when the user wants to resume from a saved checkpoint — e.g. "resume from checkpoint", "load checkpoint", "continue from checkpoint", "cf-checkpoint-from", "pick up from the saved context", "restore conversation context". Requires an existing checkpoint file — does NOT create checkpoints (use /cf-checkpoint for that).

Extract knowledge for human learning from conversation (configurable output, language, categories). Auto-invoke this skill when the conversation contains substantial new technical knowledge, problem-solving insights, or non-trivial concepts worth preserving — such as debugging breakthroughs, architecture decisions, best practices, or explanations of complex topics. Do NOT auto-invoke for trivial tasks like simple file edits, typo fixes, or routine operations. Unlike /cf-remember (project context for AI recall), this creates educational notes for the human to learn from.

Structured optimization workflow — baseline, analyze, optimize, measure, compare. Auto-invoke this skill when the conversation involves performance optimization, speed improvements, or the user mentions something is slow — e.g. "this is slow", "make it faster", "optimize", "performance", "bottleneck", "too many queries", "high latency", "memory leak", "reduce load time", "speed up", "takes too long", "timeout", "O(n²)", "N+1". Do NOT auto-invoke for minor refactors or style changes that are not performance-related.

Systematic 4-phase debugging — root cause investigation, hypothesis testing, regression-guarded fix, mandatory bug doc. Auto-invoke when a bug is non-trivial or a previous fix didn't stick — signals: bug recurs ("still broken", "same error again", "came back"), flaky / intermittent / race / timing, "used to work, now broken", regression after update or dependency bump, hard-to-reproduce, works locally but fails in CI (or vice versa), multi-stage pipeline failure, stack trace into a library, user already tried fixes that didn't help, or user asks to "find the root cause", "investigate", "diagnose", "why is this happening", "trace this", "bisect", "deep dive". Also load BEFORE touching code whenever the cause is unclear and a hypothesis must be stated first. Prefer over cf-fix for hard, recurring, or investigation-heavy bugs (cf-fix may delegate here). Do NOT auto-invoke for trivial typos, one-line fixes, or config errors with an obvious cause.

Extract project knowledge from conversation to docs/memory — for AI recall in future sessions. Use when the user wants to save project context, capture decisions, document conventions, or preserve knowledge — e.g. "remember this", "save this to memory", "document what we did", "capture this decision", "write this down", "note this", "record this convention". Auto-invoke when: (1) a bug fix had a non-obvious root cause, a workaround, or took multiple attempts — save to bugs/; (2) an explicit architecture or approach decision was made — save to decisions/; (3) a new project-wide convention or pattern was established — save to conventions/; (4) a substantial session ends with new feature flows, non-obvious API behavior, or gotchas — save to features/. Do NOT auto-invoke for trivial fixes, simple config changes, or purely educational exchanges (use /cf-learn for those). Unlike /cf-learn (educational notes for humans), this saves project context for AI recall.

Collect and act on review results from an external AI agent. Reads the result file written by an external reviewer and presents findings, then offers to fix issues. Use when the user wants to collect an outside review — e.g. "review in", "collect review", "check review results", "cf-review-in", "read external review", "import review".

Scan project and populate memory with knowledge — architecture, conventions, tech stack, key features, infrastructure. Use when a project has no memories yet, or when the user wants to refresh/rebuild project understanding — e.g. "scan the project", "scan the codebase", "bootstrap memory", "initialize memory with project knowledge", "populate memory", "build project understanding", "scan this aspect", "analyze the project". This is a token-heavy operation — always warn the user before proceeding.

Verify, commit, push, and create PR. Use when the user wants to ship their work — e.g. "ship it", "push and create PR", "let's ship", "deploy this", "send it", "push my changes", "create a PR", "open a pull request", "ready to merge", "let's get this merged". Also triggers when the user is done with a feature branch and wants the full verify → commit → push → PR workflow. Supports --dry-run to preview without pushing.

Personal teacher skill — after any task, explains what happened in a conversational, storytelling way (like a sharp friend over coffee). Covers: approach taken, alternatives rejected, how parts connect, tools used, tradeoffs, mistakes made, pitfalls to watch, expert observations, and transferable lessons. Unlike /cf-learn (concise structured reference notes), this creates a narrative deep-dive for the human to truly understand what happened and why.

Catch up after absence — summarize git history for a specific user. Use when a user returns after time away from a project and needs to understand what happened — e.g. "warm up", "catch me up", "what happened while I was away", "what did I work on", "summarize recent changes", "what changed since I left", "warm up the project".

Quick Q&A about codebase — explores code to answer, saves to docs/memory. Use when the user asks a focused question about the project — e.g. "how does X work?", "where is Y defined?", "what's the flow for Z?", "explain this module", "how are these connected?", "what pattern does this use?", "why is this done this way?". Unlike $cf-research (deep multi-doc output), this gives a single focused answer.

Load a previously saved checkpoint file back into a fresh conversation so work can continue with the prior context. Use when the user wants to resume from a saved checkpoint — e.g. "resume from checkpoint", "load checkpoint", "continue from checkpoint", "cf-checkpoint-from", "pick up from the saved context", "restore conversation context". Requires an existing checkpoint file — does NOT create checkpoints (use $cf-checkpoint for that).

Smart conventional commit with diff analysis. Use when the user wants to commit changes — e.g. "commit this", "commit my changes", "save my work", "create a commit", "git commit", "commit what we did", "stage and commit". Also triggers when the user finishes a task and wants to commit the result.

UI design workflow — scan existing UI patterns, design new UI from a description, or modify existing UI while keeping it visually consistent with the rest of the project. Use when the user wants to: design or redesign a component, page, or section; change colors, typography, spacing, or visual style; understand the project's current design system; add new UI that should match existing patterns; or ensure a UI change doesn't look out of place. Use when the user says things like "make it look like", "update the styling", "redesign this", "match the existing style", "what design patterns does this project use", "extract the design system", "add a dark mode", "make it more minimal/bold/clean", or describes a desired visual outcome. Invoke via $cf-design — helps Codex produce intentional, consistent UI.

Quick bug fix workflow. Use when the user reports a bug, error, or broken behavior — e.g. "fix this", "it's broken", "not working", "there's a bug", "I'm getting an error", "this crashes", "something is wrong", "why does this fail", "debug this", "it throws", "fix the issue", "resolve this error", "help me fix", "can you fix", "this doesn't work", "stopped working", "regression", "unexpected behavior", "failing test", "broken after update". Also triggers on stack traces, error messages, or descriptions of incorrect program behavior.

Extract knowledge for human learning from conversation (configurable output, language, categories). Auto-invoke this skill when the conversation contains substantial new technical knowledge, problem-solving insights, or non-trivial concepts worth preserving — such as debugging breakthroughs, architecture decisions, best practices, or explanations of complex topics. Do NOT auto-invoke for trivial tasks like simple file edits, typo fixes, or routine operations. Unlike $cf-remember (project context for AI recall), this creates educational notes for the human to learn from.

Structured optimization workflow — baseline, analyze, optimize, measure, compare. Auto-invoke this skill when the conversation involves performance optimization, speed improvements, or the user mentions something is slow — e.g. "this is slow", "make it faster", "optimize", "performance", "bottleneck", "too many queries", "high latency", "memory leak", "reduce load time", "speed up", "takes too long", "timeout", "O(n²)", "N+1". Do NOT auto-invoke for minor refactors or style changes that are not performance-related.

Resume an existing implementation plan created by $cf-plan, continuing from where execution last stopped. Use when the user wants to continue, resume, pick up, or finish a previously saved plan — e.g. "resume the plan", "continue the plan", "pick up where we left off", "finish the plan", "continue implementing the plan", "resume <slug>". Requires an existing plan file under docs/plans/. Does NOT create new plans — use $cf-plan for that.

Extract project knowledge from conversation to docs/memory — for AI recall in future sessions. Use when the user wants to save project context, capture decisions, document conventions, or preserve knowledge — e.g. "remember this", "save this to memory", "document what we did", "capture this decision", "write this down", "note this", "record this convention". Auto-invoke when: (1) a bug fix had a non-obvious root cause, a workaround, or took multiple attempts — save to bugs/; (2) an explicit architecture or approach decision was made — save to decisions/; (3) a new project-wide convention or pattern was established — save to conventions/; (4) a substantial session ends with new feature flows, non-obvious API behavior, or gotchas — save to features/. Do NOT auto-invoke for trivial fixes, simple config changes, or purely educational exchanges (use $cf-learn for those). Unlike $cf-learn (educational notes for humans), this saves project context for AI recall.

Collect and act on review results from an external AI agent. Reads the result file written by an external reviewer and presents findings, then offers to fix issues. Use when the user wants to collect an outside review — e.g. "review in", "collect review", "check review results", "cf-review-in", "read external review", "import review".

Scan project and populate memory with knowledge — architecture, conventions, tech stack, key features, infrastructure. Use when a project has no memories yet, or when the user wants to refresh/rebuild project understanding — e.g. "scan the project", "scan the codebase", "bootstrap memory", "initialize memory with project knowledge", "populate memory", "build project understanding", "scan this aspect", "analyze the project". This is a token-heavy operation — always warn the user before proceeding.

Verify, commit, push, and create PR. Use when the user wants to ship their work — e.g. "ship it", "push and create PR", "let's ship", "deploy this", "send it", "push my changes", "create a PR", "open a pull request", "ready to merge", "let's get this merged". Also triggers when the user is done with a feature branch and wants the full verify → commit → push → PR workflow. Supports --dry-run to preview without pushing.

Systematic 4-phase debugging — root cause investigation, hypothesis testing, regression-guarded fix, mandatory bug doc. Auto-invoke when a bug is non-trivial or a previous fix didn't stick — signals: bug recurs ("still broken", "same error again", "came back"), flaky / intermittent / race / timing, "used to work, now broken", regression after update or dependency bump, hard-to-reproduce, works locally but fails in CI (or vice versa), multi-stage pipeline failure, stack trace into a library, user already tried fixes that didn't help, or user asks to "find the root cause", "investigate", "diagnose", "why is this happening", "trace this", "bisect", "deep dive". Also load BEFORE touching code whenever the cause is unclear and a hypothesis must be stated first. Prefer over cf-fix for hard, recurring, or investigation-heavy bugs (cf-fix may delegate here). Do NOT auto-invoke for trivial typos, one-line fixes, or config errors with an obvious cause.

Use when writing new production code, adding features, implementing changes, or refactoring existing code — e.g. "implement this", "build this feature", "create a function", "add a new endpoint", "write the implementation", "refactor this", "clean up this code", "extract into a module", "simplify this function", "scaffold", "write a test", "add tests", "create a component", "implement the API", "add a route", "write a service", "create the handler". Also auto-invoke when the conversation transitions from planning/discussion to actual code writing — any time production code is about to be written, this skill MUST be loaded first. By default uses direct implementation (no new tests written). TDD is enabled when `--add-tests` is present or config `tdd: true`. This is the gate for all code writing in the project. Do NOT auto-invoke for documentation-only changes, config edits, non-code file updates, or when the user is only asking questions about code without requesting changes.

Personal teacher skill — after any task, explains what happened in a conversational, storytelling way (like a sharp friend over coffee). Covers: approach taken, alternatives rejected, how parts connect, tools used, tradeoffs, mistakes made, pitfalls to watch, expert observations, and transferable lessons. Unlike $cf-learn (concise structured reference notes), this creates a narrative deep-dive for the human to truly understand what happened and why.

Verify before claiming work is complete

Catch up after absence — summarize git history for a specific user. Use when a user returns after time away from a project and needs to understand what happened — e.g. "warm up", "catch me up", "what happened while I was away", "what did I work on", "summarize recent changes", "what changed since I left", "warm up the project".