Artifact Catalog
118 artifacts — 63 skills, 28 rules, 21 agents, 6 presets
118 results
codi-agent-creator
Agent creation workflow. Use when the user asks to create, build, or define a specialized agent. Also activate when the user wants to add a code reviewer, security analyzer, test generator, or any autonomous worker role. After creation, use codi-artifact-contributor to share, package as ZIP, or open a PR with the new agent. Do NOT activate for creating a skill (use codi-skill-creator), creating a rule (use codi-rule-creator), or bundling multiple artifacts as a preset (use codi-preset-creator).
codi-algorithmic-art
Generative and algorithmic art with p5.js. Use when the user wants to create generative art, algorithmic art, computational art, creative coding sketches, p5.js sketches, flow fields, particle systems, Perlin noise fields, L-systems, recursive or parametric art, or any interactive artwork driven by seeded randomness. Also activate for phrases like "generative sketch", "noise field", "algorithmic composition", or "generative aesthetic movement". Do NOT activate for static illustrations, logo design, or data visualization — those have their own skills.
codi-artifact-contributor
Contribute or share custom codi artifacts (rules, skills, agents, commands). Use when the user wants to open a PR to the codi repo, submit a PR to a team preset repository, share artifacts as a ZIP package, publish an agent or rule, push a preset to a custom repo, fork and PR, or set up GitHub CLI / GitHub MCP for contributions. Also activate for phrases like "share my skill", "contribute to codi", "open a PR with my rule", "publish my agent", "export preset", "send this to the codi repo". Do NOT activate for ordinary code commits, installing a preset (use codi-preset-creator or \`codi preset install\` directly), or releasing a package to npm.
codi-audio-transcriber
Audio and video transcription via OpenAI Whisper. Use when transcribing an mp3, wav, m4a, mp4, webm, or mov file; converting speech to text; generating a podcast transcript, meeting transcript, interview transcript, or subtitle text; or extracting text from a video recording. Also activate for phrases like "transcribe this audio", "convert audio to text", "get the text from this recording", or when the user wants transcripts written to Google Sheets. Launches a local Flask web UI on port 8765 with concurrent chunk processing, clipboard copy, and .txt download. Do NOT activate for real-time microphone transcription (files only) or for text-to-speech / speech synthesis.
codi-audit-fix
Iterative audit-and-fix workflow. Use when processing a list of audit findings, a batch of fixes, an issue backlog, a lint or static-analysis backlog, a security-fix list, a migration checklist, a TODO list of improvements, or any set of items where each one needs: evidence gathering → fix proposal → explicit user approval → implementation → commit. Also activate for phrases like "work through a list", "go through these one by one", "batch of fixes", "process the backlog", "iterate through findings". Enforces strict one-item-at-a-time discipline with a commit per item. Do NOT activate for a single bug (use codi-debugging), new feature work, or exploratory review without a fix list.
codi-box-validator
Validate and enforce uniform spacing, hierarchy, and structural consistency in HTML layouts. Use ALWAYS when generating HTML for visual designs — Instagram posts, LinkedIn carousels, slide decks, A4 pages, stories, posters, cards, or any fixed-aspect-ratio HTML layout. Also use when the user mentions 'layout', 'design', 'post', 'slide', 'carousel', 'poster', 'social media', 'presentation template', 'validate my design', 'spacing', 'uniform layout', 'box theory', or asks to create any visual HTML content. This skill enforces the 8-rule Box Layout Theory and must run AFTER every HTML generation step — iterate until the layout passes. If in doubt whether a design task needs validation, use this skill.
codi-brainstorming
Design exploration before implementation. Use before starting any non-trivial feature, change, or refactor. Also activate for phrases like "let's brainstorm", "let's think through", "design this feature", "plan this feature", "what's the best way to", "what's the right approach", "before we code", "should we use X or Y", "help me think about". Explores context, asks clarifying questions one at a time, proposes 2-3 approaches with trade-offs, and produces an approved design spec before invoking any implementation skill. Do NOT activate for pure quality tasks (security scan, code review, test coverage), one-liner fixes with zero design ambiguity, content creation tasks (articles, decks, documents, carousels — use codi-content-factory which has its own adaptive intake), or content with fully specified requirements.
codi-branch-finish
Development branch completion. Use when implementation is complete and the branch needs to be merged, submitted as a PR, kept, or discarded. Also activate for phrases like "finish this branch", "done with this branch", "wrap up the branch", "ship this branch", "ready to merge", "ready to open a PR", "clean up the worktree", "what's next for this feature branch". Verifies tests pass, presents four deterministic options (merge locally / open PR / keep as-is / discard), and cleans up worktrees for merge and discard paths. Do NOT activate for a plain commit (use codi-commit), for starting a new branch (use codi-worktrees), or for reviewing an open PR.
codi-brand-creator
Brand identity scaffold for codi-brand-creator. Use when creating any branded deliverable — social cards, slides, documents, posters, email templates, or any visual output that must carry codi-brand-creator brand identity. Also activate when the user mentions "brand guidelines", "brand standard", "corporate identity", "visual identity", "color palette", "brand colors", "brand fonts", "typography pair", "logo placement", "brand voice", or applies this brand's design tokens to new content. Provides design tokens (colors, fonts), voice guidelines, and Content Factory templates. Do NOT override a user-specified palette when the user has not asked for this brand, and do not activate for generic design work unrelated to codi-brand-creator.
codi-canvas-design
Original static visual design — posters, illustrations, album covers, book covers, editorial spreads, zine pages, print design, magazine layouts, gig posters, minimalist art pieces, brutalist or swiss-design compositions. Use when the user asks for any static .png or .pdf design piece, visual art, or museum-quality design-forward output. Also activate for phrases like "poster", "album cover", "book cover", "editorial design", "print design", "illustration", "design-forward piece", "minimalist art", "brutalist poster", "swiss design". Creates originals through a two-step philosophy + canvas approach. Never copies existing artists' work. Do NOT activate for animated content (use algorithmic-art or slack-gif-creator), data visualization (charts, plots, dashboards), logo design as a brand identity (use brand-creator), or UI components.
codi-claude-api
Build apps with the Claude API or Anthropic SDK. TRIGGER when code imports anthropic, @anthropic-ai/sdk, or claude_agent_sdk, or user asks to use Claude API. DO NOT TRIGGER for openai imports, general programming, or ML/data-science tasks.
codi-claude-artifacts-builder
Build complex, multi-component claude.ai HTML artifacts. Use when the user asks for a React-based claude.ai artifact, a multi-component artifact with state or routing, a single-file bundle of a React app, or any artifact that needs shadcn/ui or Radix UI components. Also activate for phrases like "claude.ai artifact", "React artifact", "bundle React app to HTML", "multi-component artifact", "single-file bundle", "Vite bundle", "shadcn artifact". Stack: React 18 + TypeScript + Vite + Parcel + Tailwind CSS + shadcn/ui, bundled to a single self-contained HTML. Do NOT activate for simple single-file HTML or JSX artifacts, static posters/visual art, or production Next.js / web app development.
codi-code-review
Structured code review on an uncommitted diff, single file, or local branch — the PRODUCING side. Use when reviewing code changes, auditing code quality, or producing severity-ranked findings against project rules. Activate for phrases like "review my code", "check my changes", "audit this file", "look at this diff", "feedback on implementation". Produces findings with file path, line number, severity (Critical / Warning / Suggestion), and suggested fixes. Do NOT activate when feedback ARRIVED on YOUR work (use codi-receiving-code-review), for full GitHub PR review with gh-CLI posting (use codi-pr-review), for fixing bugs (use codi-debugging), writing new code (use a content/plan/subagent skill), running a full security scan (use codi-security-scan), or measuring test coverage (use codi-test-suite).
codi-codebase-explore
Explore and understand the codebase using the code graph. Use when the user wants to trace dependencies, find callers, navigate unfamiliar code, run an impact analysis, or get an architecture overview. Also activate for phrases like "how does X work", "where is X defined", "find all callers of", "who calls this function", "what does X depend on", "trace the flow", "show relationships", "dependency map", "impact analysis", "before I refactor", "architecture overview". Do NOT activate for fixing bugs (use codi-debugging), writing new code, running tests, or making edits — this skill is read-only exploration.
codi-codebase-onboarding
Codebase onboarding workflow. Use when exploring an unfamiliar project, creating an onboarding guide, setting up project context for future agent sessions, or persisting a Project Context block into CLAUDE.md / AGENTS.md / .cursorrules. Also activate for phrases like "new to this codebase", "first time in this repo", "orient me", "onboard me", "what does this project do", "project overview", "get started with this codebase", "set up project context", "set up CLAUDE.md". Analyzes architecture, conventions, and key files to produce a concise guide for new team members or AI agents. Do NOT activate for fixing bugs, writing new code, single-file reviews, or dependency tracing — those have dedicated skills.
codi-codi-brand
Codi brand identity. Use when creating any branded deliverable for Codi — slides, documents, social cards, reports, or any HTML/PDF/PPTX/DOCX output that must carry Codi visual identity. Also activate when the user mentions 'codi brand', 'codi design system', or asks for Codi-branded output of any kind. Provides design tokens (colors, fonts), voice guidelines, references, and Content Factory templates. Rendering and export are handled by the codi-content-factory skill — this skill ships no preview server or code.
codi-commit
Git commit workflow with conventional commits, pre-commit checks, and staged-change review. Use when the user asks to commit code, stage changes, create a commit, write a commit message, use conventional commit format, split a commit, or troubleshoot a pre-commit hook failure. Also activate for phrases like "commit these changes", "git commit", "stage these files", "write a conventional commit", "pre-commit failed", "hook failed", "commit message rejected", "commit-msg hook". Activates on /commit. Do NOT activate for pushing to remote (use direct \`git push\`), creating pull requests (use codi-artifact-contributor or codi-branch-finish), merging branches (use codi-branch-finish), or resolving merge conflicts.
codi-compare-preset
Compare local Codi artifacts against upstream templates. Use when the user wants to check customizations, review pending upstream updates, or prepare to contribute improvements. Also activate for phrases like "what have I customized", "diff against upstream", "see my changes vs Codi", "pending upstream updates", "show differences", "review my local config", "upgrade Codi artifacts", or on /codi-compare-preset. Produces a report of local improvements (contribution candidates), upstream updates available, and conflicts. Do NOT activate for installing a preset (use \`codi preset install\`), creating a new preset (use codi-preset-creator), or opening the PR itself (use codi-artifact-contributor).
codi-content-factory
Use when the user wants to create content — articles, blog posts, slide decks, social carousels, documents, business deliverables (reports, proposals, one-pagers, case studies, executive summaries), single-format social posts, or multi-format campaigns (blog + deck + carousel about the same topic). Authors substance first (an anchor article) and distills into every requested visual format. Explicit fast path for one-off requests. Generates branded HTML with an interactive web app — live preview and context-aware export (PNG, PDF, PPTX, DOCX, HTML, ZIP). For branded business documents specifically, see references/business-documents.md.
codi-debugging
Root-cause debugging with tiered escalation. Phases 1-4 = first-line workflow (Root Cause → Pattern → Hypothesis → Implementation). Phase 5 = MCP-powered deep diagnosis when standard phases fail (code-graph, docs search, sequential thinking, reasoning-confirmation-execution loop). Use when investigating a bug, test failure, crash, exception, unexpected behavior, build failure, integration issue, flaky test, test pollution, or production incident. Also activate for phrases like "why is X broken", "can't figure out why", "debug this", "error investigation", "stack trace", "root cause", "stuck on this error", "multiple attempts failed", "deep diagnosis", "MCP investigation", "sequential thinking", and on /codi-check. Enforces root cause before any fix — no fixes may be proposed until Phase 1 investigation is complete. Do NOT activate for writing new code or features (use codi-plan-writer), planning architecture (use codi-brainstorming), or running tests without a failure to investigate (use codi-test-suite).
codi-dev-docs-manager
Codi self-documentation skill (for Codi contributors only). Use when building, updating, regenerating, or checking Codi's own docs site. Also activate for phrases like "regenerate docs", "are the docs stale", "update the docs", "rebuild docs site", "validate docs freshness", "skill catalog", "pre-release doc sweep", or when the user runs \`codi docs --generate\` / \`--validate\`. Maintains 20 auto-generated sections across artifacts.md, configuration.md, architecture.md, presets.md, and README.md. Do NOT activate for consumer project documentation (use codi-project-documentation), branded reports or proposals (use codi-content-factory), or user-facing tutorials.
codi-dev-e2e-testing
End-to-end validation of the Codi installation (for Codi contributors). Use when the user asks to test or audit the full Codi feature set — artifacts, presets, hooks, verification, commit workflow, MCP, pre-commit discipline — in a clean test project. Also activate for phrases like "validate codi installation", "e2e sweep", "full feature test", "release validation", "regression sweep", "system test", "Codi installation audit". Do NOT activate for testing a single skill in isolation (use codi-test-suite), consumer application e2e (use codi-webapp-testing), or unit test generation (use codi-tdd).
codi-dev-operations
Unified Codi operations skill. Use when the user wants to add, update, or remove rules, skills, or agents; configure flags, presets, or MCP servers; run \`codi generate\`, \`add\`, \`update\`, \`verify\`, \`doctor\`, \`validate\`, or \`preset\` commands; resolve \`generate\` exit-2 conflicts; check drift; clean or revert backups; or follow the self-dev clean + reinstall flow for source-layer edits. Also activate for phrases like "codi generate didn't propagate", "my source edit isn't showing up", "three-layer pipeline", "conflict resolution". Do NOT activate for writing a new skill from scratch (use codi-skill-creator), creating a new rule (use codi-rule-creator), or scaffolding a new agent (use codi-agent-creator) — those have their own lifecycle workflows.
codi-dispatching-parallel-agents
Parallel agent orchestration across multiple INDEPENDENT failure domains. Use when investigation reveals 3+ unrelated failures (test files with no shared cause, separate subsystems, disjoint migration tasks, unrelated audit findings) so dispatching N concurrent agents beats serial diagnosis. Trigger phrases: "dispatch parallel agents", "parallel debug", "fan out to subagents", "investigate these in parallel", "concurrent investigation", "multiple independent failures". Enforces an iron law — independence must be proven before dispatch (no shared state, no sequential dependency, no shared files). Each agent gets a focused per-domain prompt; integration is mechanical, not a re-diagnosis. Do NOT activate for executing approved plan tasks (use codi-plan-execution SUBAGENT — sequential by contract), for a single failure (use codi-debugging), for tasks with shared state (decouple first), or for a single subagent (call it directly).
codi-docx
Create, edit, read, or analyze Word documents (.docx). Use when the user mentions "Word doc", "Word document", ".docx", "convert doc to docx", or any .docx workflow. Also activate when adding tracked changes, accepting or rejecting changes, adding comments, inserting images, extracting text or tables from a .docx, producing reports, memos, letters, or templates as Word files, or generating a branded DOCX with design tokens. Handles unpack → edit XML → pack for existing docs, or docx-js/python-docx for new docs. Do NOT activate for PDF files (use codi-pdf), spreadsheets (use codi-xlsx), PowerPoint files (use codi-pptx), or plain Markdown docs (use codi-project-documentation).
codi-evidence-gathering
Structured investigation before proposing changes. Use when you need to gather concrete evidence about actual vs intended behavior before suggesting a fix, evaluating an audit item, or validating a step. Also activate for phrases like "gather evidence", "investigate before fixing", "actual vs intended behavior", "structured investigation", "verify my assumption", "check before proposing", "investigate X before we change it". Called by codi-audit-fix and codi-guided-execution, also usable standalone. Do NOT activate when you already have fresh, direct tool output from the current session, when the user wants to write new code (use codi-plan-writer), or when the investigation is already complete and a fix is the next step.
codi-frontend-design
Distinctive, production-grade frontend interfaces. Use when the user asks to build a website, landing page, dashboard, hero section, navbar, or any React / Vue / HTML component. Also activate when styling, restyling, or beautifying an existing web app, redesigning a UI, setting up a design system, using shadcn/ui components, or working on CSS styling. Also trigger for phrases like "beautify this app", "make this look better", "redesign this page", "elevate the UI", "design system", "hero section". Avoids generic AI aesthetics (overused fonts like Inter / Roboto, purple gradients on white, predictable layouts). Do NOT activate for marketing one-pagers, landing pages, or PDF documents whose primary output is exported HTML/PDF — use codi-content-factory (this skill is for app UIs only: Next.js, Vite), for slide decks or presentations (use codi-content-factory or codi-pptx), for static posters / album covers (use codi-canvas-design), or for multi-component claude.ai artifacts (use codi-claude-artifacts-builder).
codi-graph-sync
Synchronize the code knowledge graph. Use when the graph is stale, files have changed significantly, or queries return outdated results. Also activate for phrases like "reindex repo", "refresh graph", "update code graph", "graph is out of date", "graph-code is stale", or on /update_graph and /index_graph. Choose full re-index for major refactors / first-time indexing; choose incremental update for routine post-edit refreshes. Do NOT activate for querying the graph (use codi-codebase-explore) or for regular code edits that do not require fresh graph state.
codi-guided-execution
Collaborative step-by-step execution for first-time technical processes. Use when the user wants to perform a setup, configuration, infrastructure task, deployment, or operational workflow for the first time and needs structured guidance with documentation at each step. Also activate for phrases like "first-time setup", "step by step", "walk me through", "guide me through", "OAuth configuration", "cloud provider setup", "CI/CD setup", "interactive deployment", "configure from scratch". Agent and user work as a team: agent reasons, plans, and executes what it can; user performs actions that require credentials, browser access, or external systems. Every completed step produces a written document. Do NOT activate for pure code changes (use codi-plan-execution), audits of existing code (use codi-audit-fix), or design-only tasks (use codi-brainstorming → codi-plan-writer).
codi-guided-qa-testing
Step-by-step QA testing methodology. Use when the user wants to test, validate, or QA a project systematically. Also activate for phrases like "QA this project", "validate the release", "test the whole app", "systematic testing", "pre-release QA", "regression pass", "user acceptance testing", "UAT", "check if everything works". Coding agent drives automatable [AGENT] phases, flags [HUMAN]-only phases, and tracks all results in a living QA report document. Do NOT activate for running a single test or measuring coverage (use codi-test-suite), or auditing the codi installation itself (use codi-dev-e2e-testing).
codi-html-live-inspect
Use when the user wants to open a local HTML file or static website in "live inspect" mode so the coding agent can see which elements the user is clicking on in real time. Also activate when the user says "open this html and watch what I click", "start html live inspect", "inspect this page with me", "collaboratory mode for this site", "let me select elements in the browser and tell you what they are", or wants the agent to drive a local web page. Runs a local HTTP server that injects a DOM inspector into every served page, exposes REST endpoints for the agent to read the current selection, recent user interactions, and (optionally) to run JavaScript inside the page.
codi-humanizer
Remove AI writing patterns from generated content. Use when the user says text sounds robotic, formulaic, AI-generated, too formal, stiff, corporate, or "like ChatGPT / Claude". Also activate for phrases like "humanize this", "sounds too AI", "dehumanize AI content", "make this natural", "remove AI tells", "natural tone", "strip AI voice". Applies a two-pass rewrite against 29 documented AI patterns (vocabulary, structure, style, communication, filler) plus a Soul Check for substance and point-of-view. Do NOT activate for writing new content from scratch (use codi-content-factory), code comments (use codi-refactoring or direct edits), translation, or copy-editing for grammar only.
codi-internal-comms
Write internal communications — status reports, 3P updates (Progress / Plans / Problems), leadership updates, team newsletters, company-wide newsletters, FAQ responses, incident reports, internal memos, internal emails, all-hands summaries, weekly/bi-weekly/monthly updates. Use when the user says "write the 3P update", "draft the newsletter", "write the team update", "answer this FAQ", "internal memo", "leadership email", "status update for the team". Follows structured guideline files per comm type. Do NOT activate for external press releases, customer-facing emails, marketing copy (use a copywriting skill), or personal 1:1 communications.
codi-mcp-ops
MCP (Model Context Protocol) operations. Use when the user wants to set up, configure, call, debug, or build an MCP server. Also activate for phrases like "add MCP server", "configure MCP", "MCP tool call failed", "MCP Inspector", "stdio transport", "http MCP", "build custom MCP", "connect external API as MCP", or when the user mentions mcp.yaml, Model Context Protocol, or MCP tools. Handles mcp.yaml config, tool discovery and calling, scaffolding (\`codi add mcp-server\`), building custom servers with @modelcontextprotocol/sdk, and distribution via \`codi generate\`. Do NOT activate for the Claude API or Anthropic SDK (use codi-claude-api) or generic REST API design (use a general api-designer flow).
codi-mobile-development
Mobile development for iOS (SwiftUI / UIKit), Android (Jetpack Compose / XML), and cross-platform (React Native, Flutter, Kotlin Multiplatform). Use when the user is building, reviewing, or architecting a mobile app. Also activate for phrases like "iOS app", "Android app", "SwiftUI view", "Jetpack Compose screen", "@Observable", "React Native component", "Flutter widget", "KMP module", "App Store", "Play Store", "mobile testing", "mobile accessibility", "mobile performance", "deep linking", "push notifications", "mobile architecture". Do NOT activate for web UI (use codi-frontend-design), desktop apps, game engines, or pure backend API work.
codi-notebooklm
Google NotebookLM research assistant with browser automation. Use when the user wants to query NotebookLM notebooks, manage a notebook library, authenticate with Google, generate Audio Overviews (podcasts), or get source-grounded answers from uploaded documents. Also activate when the user mentions NotebookLM, shares a NotebookLM URL, or uses phrases like "ask my docs", "check my notebook", "query my documentation".
codi-pdf
PDF processing — read, create, edit, merge, split, rotate, watermark, encrypt, OCR. Use when the user mentions a .pdf file, asks to extract text, tables, or images from a PDF, fill a PDF form, sign a PDF, merge or split PDFs, convert a page range, or make a scanned PDF searchable via OCR. Also activate for phrases like "PDF to text", "extract tables", "OCR scanned PDF", "fill this form", "merge PDFs", "rotate pages", "add watermark", "encrypt PDF", "PDF metadata". Handles both Python (pypdf, pypdfium2) and JavaScript (pdf-lib). Do NOT activate for Word documents (use codi-docx), Excel spreadsheets (use codi-xlsx), HTML-to-PDF rendering from scratch (use codi-content-factory), or plain Markdown docs (use codi-project-documentation).
codi-plan-execution
Execute an approved implementation plan in one of two modes — INLINE (sequential checkpoint-driven execution by the primary agent) or SUBAGENT (fresh subagent per task with two-stage review). Use after codi-plan-writer produces an approved plan. Always asks the user to pick the mode — never auto-selects. Activates on phrases like "execute this plan", "implement the plan", "run the plan", "walk through the plan step by step", "TDD per task", "dispatch subagents for each task", "subagent orchestration", "two-stage review", "plan execution", "multi-file implementation". Do NOT activate without an approved plan (use codi-brainstorming → codi-plan-writer first), for trivial single-file edits, for bug investigation without a fix plan (use codi-debugging), for parallel dispatch across INDEPENDENT failure domains (use codi-dispatching-parallel-agents), or when the baseline test suite is already failing (fix the baseline first, never execute on red).
codi-plan-writer
Implementation plan generator. Use after codi-brainstorming produces an approved design spec and the user wants the spec broken into atomic 2-5 minute TDD tasks with exact file paths, complete code, and runnable verification commands. Also activate for phrases like "break this into tasks", "write the plan", "plan from spec", "TDD task breakdown", "atomic tasks", "implementation task list", "turn this spec into an execution plan". Produces an executable plan document at \`docs/YYYYMMDD_HHMMSS_[PLAN]_<feature>-impl.md\`. Do NOT activate without an approved spec (use codi-brainstorming first), for executing an existing plan (use codi-plan-execution), or for single-file edits that don't warrant a plan.
codi-pptx
Create, edit, or read PowerPoint (.pptx) files. Use when the user mentions a .pptx file, "deck", "slides", "presentation", "PowerPoint", "pitch deck", "investor deck", "sales deck", "keynote file", or when the user wants to extract speaker notes, insert a slide, edit a deck, or build a branded deck. Also activate for phrases like "update slide N", "add a slide to the deck", "pull text from the pptx", "combine two decks". Stack: python-pptx + pptxgenjs via runtime-detected ts/python scripts. Do NOT activate for PDF slide exports (use codi-pdf), HTML slide decks (use codi-content-factory), Word documents (use codi-docx), or Excel spreadsheets (use codi-xlsx).
codi-pr-review
End-to-end pull request review — the PRODUCING side. Use when reviewing a PR, auditing a pull request, commenting on a PR, posting a review on GitHub, or documenting PR findings. Trigger phrases: "review PR #N", "review this pull request", "review my PR", "audit the PR", "post review on PR", "PR review doc", "check PR against claims". Produces a severity-ranked findings document (OWASP-aligned Critical/High/Medium/Low + Conventional Comments labels), saves it under docs/, and posts it to the PR via gh CLI. Do NOT activate when the user RECEIVED feedback on their own PR and is deciding how to respond (use codi-receiving-code-review), for reviewing a single file or uncommitted diff (use codi-code-review), for fixing issues already found (use codi-debugging), for a dedicated security audit on the whole codebase (use codi-security-scan), or for general project documentation (use codi-project-documentation).
codi-preset-creator
Guided creation of Codi presets. Use when the user wants to create, package, scaffold, fork, share, publish, or customize a Codi preset — a reusable bundle of rules, skills, agents, commands, flags, and MCP configs. Also activate for phrases like "bundle my rules and skills", "make a preset for my team", "org-wide preset", "fork the balanced preset", "share preset via ZIP", "publish preset to GitHub". Do NOT activate for installing an existing preset (use \`codi preset install\` directly), diffing local vs upstream (use codi-compare-preset), or contributing artifacts back to Codi upstream (use codi-artifact-contributor).
codi-project-documentation
Documentation creation and maintenance for consumer projects. Use when the user asks to write or update a README, API reference, OpenAPI spec, ADR, contributing guide, changelog, RFC, spec document, JSDoc, docstrings, or inline code documentation. Also activate for phrases like "write the README", "document this API", "technical writing", "draft a proposal", "write an RFC", "author a decision doc", "co-author a spec". Applies the Diataxis framework (tutorial / how-to / reference / explanation). Do NOT activate for branded PDF reports or proposals (use codi-content-factory), Word documents (use codi-docx), slide decks (use codi-content-factory or codi-pptx), or Codi's own self-documentation (use codi-dev-docs-manager).
codi-project-quality-guard
Audit and enforce project quality infrastructure. Use when the user sets up a new project, scaffolds from scratch, audits CI/CD, checks for missing .gitignore / .gitattributes / .editorconfig, adds pre-commit hooks, linting, testing, security scanning, Docker, or environment config. Also activate for phrases like "project audit", "quality infrastructure", "pipeline continuity", "CI passes locally but not in CI", "setup from scratch", "new project setup", "what's missing in my config". Covers 9 categories (git / security / pre-commit / Python / TypeScript / testing / CI/CD / Docker / env). Do NOT activate for writing application code (use codi-plan-writer), debugging runtime errors (use codi-debugging), or one-off formatter questions.
codi-receiving-code-review
Receiving feedback on YOUR OWN work. Use when a reviewer, teammate, the user, or an automated tool returns feedback on a diff, PR, commit, or code you wrote. Trigger phrases: "the reviewer said", "review feedback came back", "my PR got comments", "they suggested I change", "responding to PR comments", "addressing reviewer concerns", "should I apply this suggestion". Iron law — external feedback is suggestions to evaluate, not orders to follow. Forbids performative agreement ("you're absolutely right!", "good catch!", blind implementation). Requires verifying every claim against the code, distinguishing reviewer error from real issues, and pushing back with evidence when wrong. Do NOT activate for PRODUCING a review on an uncommitted diff (use codi-code-review), for PRODUCING a full PR review (use codi-pr-review), or for the initial completeness check on your own work before review (use codi-verification).
codi-refactoring
Safe dead-code removal and behavior-preserving refactoring workflow. Use when the user wants to remove dead code, cleanup unused imports, delete unused exports, consolidate duplicates, DRY up a module, merge near-identical functions, or simplify code without changing behavior. Also activate for phrases like "remove dead code", "cleanup unused", "simplify this module", "DRY up", "delete unused imports", or when the user runs dead-code detection tools (knip, vulture, deadcode, udeps). Classifies each finding as SAFE / CAUTION / DANGER and re-runs tests after every deletion. Do NOT activate for new features (use codi-plan-writer), bug fixes (use codi-debugging), behavior-changing rewrites (use codi-plan-writer with tests), or performance tuning.
codi-refine-rules
Review and refine rules from collected feedback. Two modes: REVIEW (read-only summary of \`.codi/feedback/\` grouped by artifact with top 3 actions) and REFINE (interactive one-at-a-time approval flow that edits rule files and runs \`codi generate\`). Use when the user asks to review collected observations, show accumulated feedback, audit \`.codi/feedback/\`, improve rules, refine rules, process rule feedback, apply rule updates, or fix outdated rules. Also activates on /codi-refine-rules or phrases like "show feedback summary", "what's accumulated", "review observations", "process rule feedback", "update outdated rule". Do NOT activate for creating a brand-new rule (use codi-rule-creator), emitting new observations during work (that happens automatically via codi-rule-feedback), or general quality audits without collected feedback (use codi-compare-preset).
codi-roadmap
Create a structured roadmap or persistent todo list as a JSON file in \`docs/roadmaps/\`. Use when the user plans features, tracks milestones, breaks down an epic, organizes phase-based work, or drafts a product roadmap. Also activate for phrases like "create a roadmap", "plan next steps", "feature roadmap", "Q3 plan", "milestone tracking", "epic breakdown", "phase-based plan", "multi-phase rollout", "product roadmap". Output is JSON (priority, status, dependencies). Do NOT activate for in-session task tracking (use TaskCreate), implementation plans (use codi-plan-writer after codi-brainstorming), or daily progress logs (use codi-session-log).
codi-rule-creator
Rule creation workflow. Use when the user asks to create, write, or define a coding rule, standard, or convention. Also activate when the user wants to enforce behavior, set constraints, or establish coding standards. After creation, use codi-artifact-contributor to share, package as ZIP, or open a PR with the new rule. Do NOT activate for creating a skill (use codi-skill-creator), creating an agent (use codi-agent-creator), or refining an existing rule from collected feedback (use codi-refine-rules).
codi-rule-feedback
Rule observation skill. Activates automatically when the agent notices a gap, outdated guidance, missing example, or user correction related to a loaded rule. Emits an inline \`[CODI-OBSERVATION: rule | category | text]\` marker — the Stop hook collects and structures it to \`.codi/feedback/rules/\`. Categories: \`missing-step\`, \`outdated-rule\`, \`missing-example\`, \`user-correction\`, \`trigger-miss\`, \`trigger-false\`, \`wrong-output\`. Background observer — the agent does not write files. Do NOT emit for single-occurrence anecdotes (require 2+ evidence points), do NOT directly edit rule files (use codi-refine-rules), and do NOT emit during time-critical operations (bug incidents, blocking hot fixes).
codi-security-scan
Security analysis workflow. Use when the user wants to audit the codebase for vulnerabilities, hardcoded secrets, OWASP Top 10 risks, dependency CVEs, input-validation gaps, or supply-chain issues. Also activate for phrases like "audit security", "vulnerability scan", "check for secrets", "CVE scan", "SAST", "security review", "pentest findings", "threat model", "OWASP audit", "supply chain check". Produces severity-ranked findings (Critical / High / Medium / Low) with actionable fixes. Do NOT activate for general code review (use codi-code-review), fixing a specific bug (use codi-debugging), or project-setup CI/CD scaffolding (use codi-project-quality-guard).
codi-session-log
Session log and handoff — unified markdown work journal in \`docs/sessions/\`. Three modes. HANDOFF: user is switching to a new chat or context is near limit; writes a dense ready-to-paste handoff markdown file. LOG (close-day): end-of-day summary appended to today's log markdown. RESUME (open-day): start-of-day recap reading the latest session files. Activates on phrases like "context is full", "switch conversation", "handoff to next session", "wrap up the session", "end of day summary", "what did I do yesterday", "pick up where we left off", "/session-handoff", "/open_day", "/close_day". Do NOT activate for recovery from repeated in-session agent mistakes (use codi-session-recovery), commit messages (use codi-commit), architecture decision records (use codi-project-documentation), or PR descriptions.
codi-session-recovery
Agent self-diagnosis after repeated mistakes. Activate when you have corrected your own mistakes 2 or more times in the current conversation, when the user flags the same mistake more than once, or when you notice yourself reverting changes or trying a third approach. Also activate for phrases like "after multiple mistakes", "self-diagnosis", "stop and reflect", "error pattern", "session degraded", "repeated corrections", "context contaminated". Produces a REVIEW-category diagnostic report under \`docs/\` and recommends a context reset. Do NOT activate for a single first-time mistake (just fix and continue), user-reported codebase bugs (use codi-debugging), end-of-day logs (use codi-session-log /close_day), or session handoff to a new chat (use codi-session-log /session-handoff).
codi-skill-creator
Skill creation, improvement, and migration workflow. Use when the user asks to create, build, scaffold, add, or improve a skill. Also activate for phrases like "new skill from scratch", "add a skill", "optimize skill description", "tune skill triggers", "write SKILL.md", "skill evals", "skill testing", "benchmark skill", "package skill", "import external skill", "migrate skill from <url/path/zip>", "security-review a skill", or on skill.test.json / SKILL.md file mentions. Do NOT activate for creating a rule (use codi-rule-creator), creating an agent (use codi-agent-creator), packaging multiple artifacts as a preset (use codi-preset-creator), or writing general project documentation (use codi-project-documentation).
codi-slack-gif-creator
Create animated GIFs optimized for Slack (custom emoji and message reactions). Use when the user asks to create a Slack GIF, custom Slack emoji, animated emoji, animated reaction, 128x128 GIF, 480x480 GIF, or any looped animation for a Slack channel. Also activate for phrases like "make an emoji", "Slack GIF for #channel", "animate this concept for Slack", "progress bar GIF". Provides Slack size constraints (128x128 emoji / 480x480 message), validation tools, easing functions, and PIL drawing helpers. Detects ts/python runtime automatically. Do NOT activate for static images (use codi-canvas-design), generative p5.js art (use codi-algorithmic-art), or screen recordings.
codi-step-documenter
Step completion document generator. Invoked after each validated step in a codi-guided-execution workflow to produce a structured, reusable guide under \`docs/executions/<workflow-name>/\`. Also usable standalone to document a completed task retroactively. Activate for phrases like "document this step", "write up what we just did", "turn this into a reusable procedure", "document the completed task", "step doc", "runbook entry", "reproducible guide". Do NOT activate for README / ADR / general project docs (use codi-project-documentation), branded reports (use codi-content-factory), or daily progress logs (use codi-session-log).
codi-tdd
Test-Driven Development discipline. Use when implementing any feature, bug fix, refactor, or behavior change. Enforces RED-GREEN-REFACTOR with iron-law verification (no production code without a failing test first). Also activate for phrases like "RED-GREEN-REFACTOR", "failing test first", "test-first", "write the test before the code", "TDD workflow", "implement with tests", "write tests as I implement". Do NOT activate for generating missing tests for existing code without implementation (use codi-test-suite), running tests without writing new behavior (use codi-test-suite RUN mode), fixing failing tests when the production code is already correct (use codi-debugging), or audit-style cleanup without behavior change (use codi-refactoring).
codi-test-suite
Unified testing skill with three modes. RUN: detect the framework and execute the test suite, report pass/fail with quick triage. COVERAGE: run with coverage instrumentation, identify files and lines below the 80% threshold, produce a gap report. GENERATE: COVERAGE + generate tests for uncovered code paths (delegates to codi-test-generator and codi-code-reviewer subagents). Auto-detects Jest, Vitest, Pytest, Go, Rust, and .NET. Activates on phrases like "run the tests", "are the tests green", "test results", "npm test", "pytest", "go test", "cargo test", "vitest", "code coverage", "test gaps", "uncovered lines", "measure coverage", "fill coverage gaps", "what's my coverage". Do NOT activate for writing new feature tests (use codi-tdd), fixing a failing test (use codi-debugging), quality-reviewing existing tests (use codi-code-review), full codi installation audits (use codi-dev-e2e-testing), or multi-phase QA sweeps (use codi-guided-qa-testing).
codi-theme-factory
Apply a visual theme (color palette + font pairing) to slides, documents, HTML artifacts, or other deliverables. Use when the user wants to apply a theme, change colors or fonts, rebrand a deck, re-theme an artifact, or pick from 10 pre-set themes (Ocean Depths, Sunset Boulevard, Forest Canopy, Modern Minimalist, Golden Hour, Arctic Frost, Desert Rose, Tech Innovation, Botanical Garden, Midnight Galaxy). Also activate for phrases like "apply a theme", "change colors", "rebrand this deck", "font pair", "visual theme", "theme a slide deck", "custom theme from description". Do NOT activate for a full brand identity system (use codi-brand-creator or codi-codi-brand), a new design from scratch (use codi-canvas-design or codi-frontend-design), or generative art (use codi-algorithmic-art).
codi-verification
Verification gate before completion. Use before claiming any task is done, fixed, passing, or complete. Requires fresh evidence (a command run in this session with its output read) — not assumptions or memory. Also activate for phrases like "about to say done", "verify before claim", "prove it works", "completion gate", "fresh evidence", "verification check", "before marking complete", and whenever you notice weasel words ("should pass", "probably works", "seems correct", "looks good"). Do NOT activate for initial investigation (use codi-evidence-gathering), debugging a specific failure (use codi-debugging), or design / planning phases where nothing is yet implemented.
codi-webapp-testing
Test, debug, or automate a local web application with Playwright. Use when the user wants to verify frontend functionality, capture screenshots, read browser console logs, check network requests, or drive a local dev server. Also activate for phrases like "Playwright", "local server test", "browser automation", "screenshot this page", "check console logs", "test my webapp", "automate UI", "headless browser", "multi-server test", "end-to-end UI check". Runtime-compat helpers manage server lifecycle. Do NOT activate for collaborative local-HTML element inspection (use codi-html-live-inspect), designing a new UI (use codi-frontend-design), generative art sketches (use codi-algorithmic-art), or full multi-phase QA sweeps (use codi-guided-qa-testing).
codi-worktrees
Git workspace setup for feature development. Use before executing an implementation plan, when the user wants a clean workspace, when parallel features need isolation, or when dirty local changes must be preserved. Also activate for phrases like "git worktree", "new branch for feature", "workspace isolation", "parallel feature work", "feature branch setup", "before plan-execution", "clean workspace", "isolate this feature". Evaluates worktree vs simple branch and sets up the chosen option. Do NOT activate for finishing a branch (use codi-branch-finish), committing changes (use codi-commit), executing the plan itself (use codi-plan-execution), or resolving merge conflicts.
codi-xlsx
Create, edit, read, or fix spreadsheet files (.xlsx, .xlsm, .csv, .tsv). Use when the user wants to work with Excel, build a financial model, clean tabular data, generate a pivot table, apply formulas, or convert between tabular formats. Also activate for phrases like "Excel file", "spreadsheet", "financial model", "data cleaning", "pivot table", "formulas", "openpyxl", "csv to xlsx", "xlsx to csv", "messy tabular data". Handles .xlsx / .xlsm / .csv / .tsv via python (openpyxl) and TypeScript runtimes. Do NOT activate for Word documents (use codi-docx), PDF files (use codi-pdf), PowerPoint decks (use codi-pptx), branded HTML reports (use codi-content-factory), or Google Sheets API integrations (use a gspread / Sheets API flow).
codi-agent-usage
Guidelines for spawning and managing sub-agents — when to use, best practices, parallel execution
codi-api-design
API design conventions, multi-protocol guidance, and schema-first development
codi-architecture
Architecture guidelines, design patterns, and module boundaries
codi-code-style
Code style and formatting conventions
codi-csharp
C# conventions — records, async patterns, LINQ, nullable types, testing
codi-django
Django conventions — fat models, QuerySet optimization, DRF serializers, migrations, testing
codi-documentation
Documentation standards and practices
codi-error-handling
Error handling, resilience patterns, and observability integration
codi-git-workflow
Git workflow and commit conventions
codi-golang
Go conventions — error handling, interfaces, testing, concurrency patterns
codi-improvement-dev
Continuous artifact improvement — observe patterns, emit CODI-OBSERVATION markers, propose rule/skill improvements with evidence and user approval
codi-java
Java conventions — records, streams, Optional, testing, dependency injection
codi-kotlin
Kotlin conventions — null safety, sealed classes, coroutines, testing patterns
codi-nextjs
Next.js App Router conventions — server components, ISR, route handlers, middleware, SEO
codi-output-discipline
Token-efficient output — anti-sycophancy, concise responses, formatting safety, scope discipline
codi-performance
Performance optimization, Core Web Vitals, and observability-driven tuning
codi-production-mindset
Production-grade standards — no shortcuts, observability, reliability, safe deployments
codi-python
Python-specific conventions — typing, data models, testing, resource management
codi-react
React-specific conventions — components, hooks, state management, performance
codi-rust
Rust conventions — ownership, error handling, traits, testing patterns
codi-security
Security best practices and vulnerability prevention
codi-simplicity-first
Simplicity-first development — avoid over-engineering, prefer direct solutions, minimize complexity
codi-spanish-orthography
Spanish language orthography enforcement — accents, tildes, punctuation
codi-spring-boot
Spring Boot conventions — constructor injection, JPA, security, testing, error handling
codi-swift
Swift conventions — value types, protocols, concurrency, testing patterns
codi-testing
Testing strategy, TDD workflow, and test infrastructure
codi-typescript
TypeScript-specific conventions — strict typing, immutability, async patterns
codi-workflow
Human-in-the-loop workflow — baby steps, self-evaluation, MCP-first exploration, no unsolicited commits
codi-ai-engineering-expert
Use when integrating LLMs into products. Covers RAG systems, embeddings, prompt engineering, guardrails, evaluation frameworks, multi-agent orchestration, and cost optimization.
codi-api-designer
Use when designing REST or GraphQL APIs. Reviews naming conventions, pagination patterns, error handling, versioning, and produces OpenAPI-compatible documentation.
codi-code-reviewer
Use when reviewing PRs or code changes. Provides severity-ranked findings (CRITICAL/HIGH/MEDIUM/LOW) with confidence filtering across multiple files in parallel.
codi-codebase-explorer
Use when navigating an unfamiliar codebase. Traces call graphs, maps dependencies, finds callers, and produces architecture overviews using a code knowledge graph.
codi-data-analytics-bi-expert
Use when designing dashboards, KPI frameworks, or analytical SQL. Covers data visualization best practices, metric definitions, and BI tool selection.
codi-data-engineering-expert
Use when designing ETL/ELT pipelines, lakehouse/warehouse architecture, or data orchestration. Produces architecture reports with Airflow/Dagster/dbt recommendations.
codi-data-intensive-architect
Use when choosing databases or designing for scale. Evaluates storage engines, replication, partitioning, and consistency trade-offs for high-throughput workloads.
codi-data-science-specialist
Use when designing ML pipelines, feature engineering, or model evaluation. Researches best practices for experiment tracking, hypothesis testing, and interpretability.
codi-docs-lookup
Use before coding to verify API signatures, check deprecations, and find working examples. Searches official docs and returns authoritative references, not guesses.
codi-legal-compliance-eu
Use when building EU-facing features. Reviews GDPR compliance, conducts DPIAs, evaluates lawful bases, data retention policies, and EU AI Act requirements.
codi-marketing-seo-specialist
Use when optimizing for search or planning content strategy. Researches current SEO best practices, keyword strategies, technical SEO for Next.js, and conversion optimization.
codi-mlops-engineer
Use when deploying models or building ML infrastructure. Designs deployment architectures, monitors drift, selects MLOps platforms, and sets up ML CI/CD pipelines.
codi-nextjs-researcher
Use when making Next.js architecture decisions. Researches current best practices for rendering strategy (SSR/SSG/ISR), caching, performance, and production deployment.
codi-openai-agents-specialist
Use when building with the OpenAI Agents SDK. Covers agent orchestration, tool design, handoff patterns, guardrails, sessions, tracing, and multi-agent architectures.
codi-payload-cms-auditor
Use when building or auditing a Payload CMS app. Reviews collection design, access control, hook patterns, performance optimization, and migration strategies.
codi-performance-auditor
Use when diagnosing performance problems. Profiles bottlenecks, detects N+1 queries, bundle size issues, memory leaks, and Core Web Vitals regressions.
codi-python-expert
Use when optimizing Python systems. Covers asyncio/threading/multiprocessing trade-offs, GIL implications, Numba/C extensions, and enterprise-grade Python workflows.
codi-refactorer
Use when cleaning up code without changing behavior. Removes dead code, consolidates duplicates, reduces cyclomatic complexity, and preserves test coverage.
codi-scalability-expert
Use when designing for traffic growth or resolving scaling bottlenecks. Covers database sharding, caching, CDN strategy, load balancing, and FinOps cost optimization.
codi-security-analyzer
Use when reviewing auth, payments, or data handling code. Runs OWASP-based vulnerability analysis with trust boundary mapping and produces a remediation plan.
codi-test-generator
Use when writing tests or improving coverage. Generates unit, integration, and E2E tests with meaningful assertions, edge cases, and test doubles where needed.
codi-minimal
Permissive — security off, no test requirements, all actions allowed
codi-balanced
Recommended — security on, type-checking strict, no force-push
codi-strict
Enforced — security locked, tests required, delete restricted, no force-push
codi-fullstack
Comprehensive web/app development — broad rules, testing, and security. Language-agnostic.
codi-dev
Preset for developing the Codi CLI itself — strict TypeScript, anti-hardcoding, safe releases, and full QA tooling
codi-power-user
Daily workflow — graph exploration, day tracking, error diagnosis, enhanced commits