dotai

Your coding agents should remember what you taught them.

dotai gives Claude, Cursor, Gemini, Codex, and other coding agents one shared memory for how you build software. Keep your rules, workflows, and engineering taste in ~/.ai/; sync them into any project with one command.

                         ┌─ CLAUDE.md + native slash commands
~/.ai/ ── dotai sync ───┼─ .cursorrules
                         ├─ GEMINI.md
                         └─ AGENTS.md

No server. No account. No proprietary memory layer. Just portable Markdown that you own.

Stop reteaching every agent

You tell one agent “never call useEffect directly.” Next week, a different agent does exactly that. Your review checklist lives in one tool, your deployment workflow in another, and the correction you made yesterday disappears with the chat.

dotai turns those corrections into durable, reusable context:

• Rules are hard constraints. Capture coding standards and scope them by project or file pattern.
• Skills are repeatable workflows. Plan, investigate, review, verify, and ship the same way every time.
• Roles change how the agent thinks. Review as a security engineer; plan as a product manager.
• Preference packs carry your taste. Share softer choices without weakening hard rules.
• Sync speaks each agent's language. Generate the bootstrap file that Claude, Cursor, Gemini, or Codex already understands.

The useful part is the feedback loop:

agent makes a mistake → capture the correction → sync → every agent learns it

See it in action

# Set up your global knowledge base
dotai init

# Teach every agent a correction you never want to repeat
dotai learn "no-print-debugging" \
  --issue "Debug output was left in production code" \
  --correction "Use structured logging and remove temporary debug output" \
  --globs "*.py"

# Bring that knowledge into a project
cd ~/my-project
dotai sync

Now the rule is available to every supported agent, alongside reusable commands such as:

/run_plan as product-manager
/run_investigate as debugger
/run_review as security-engineer
/run_verify
/run_ship

Installation

From PyPI (recommended)

pip install dotai-cli

With pipx (isolated install)

pipx install dotai-cli

From source

git clone https://github.com/dawalama/dotai.git
cd dotai
pip install -e ".[dev]"

Requires Python 3.11 or later.

Quick Start

dotai init

# See what's available
dotai roles
dotai skills
dotai rules

# Sync agent bootstrap files into a project
cd ~/my-project
dotai sync

This generates agent config files in your project root — CLAUDE.md, .cursorrules, GEMINI.md, and AGENTS.md. Structured rules and freeform conventions are included in full; roles and skills are included as a compact catalog by default. For Claude Code, sync also generates .claude/skills/ entries so skills appear as native slash commands. Use dotai sync --full when an agent needs every role and skill definition inline.

Usage

Daily workflow

# Start of day: sync your knowledge into the project you're working on
cd ~/my-project
dotai sync

# Or leave it running — auto-resyncs when you edit ~/.ai/ files
dotai watch

# Now open your editor — Claude Code, Cursor, Gemini, etc.
# Your roles, skills, and rules are already loaded.

Using skills in Claude Code

After dotai sync, skills appear as native slash commands:

> /run_plan                            # Plan before coding
> /run_plan as product-manager         # Scope with acceptance criteria
> /run_plan as systems-architect       # Architecture deep-dive
> /run_review                          # Review current branch diff
> /run_review as security-engineer     # Security-focused audit
> /run_ship                            # Sync, test, push, create PR
> /run_techdebt                        # Scan for tech debt
> /run_careful                         # Enter production-safety mode
> /run_verify                          # Run tests, types, lint, build
> /run_learn                           # Capture a mistake as a permanent rule
> /run_compress                        # Draft and review smaller rule definitions

Using skills outside Claude Code

Assemble a skill prompt and pipe it into any tool:

# Copy a skill prompt to clipboard
dotai prompt review --role paranoid-reviewer | pbcopy

# Save to a file
dotai prompt ship > /tmp/ship-prompt.md

# List skills filtered by category
dotai skills --category deployment

Recording what you learn

dotai learn creates a structured rule file in ~/.ai/rules/ by default (not a freeform journal scrap).

# Preview the rule without writing
dotai learn "auth-header" \
  --issue "Forgot Bearer prefix on API token" \
  --correction "Always prepend 'Bearer ' to auth tokens" \
  --dry-run

# Save structured rule + resync agent configs
dotai learn "auth-header" \
  -i "Forgot Bearer prefix on API token" \
  -c "Always prepend 'Bearer ' to auth tokens" \
  --sync

# Import a detailed rule from a file
dotai learn "no-useEffect" --from-file react-rules.md --globs "*.tsx,*.ts"

# Freeform journal entry in rules.md (opt-in)
dotai learn "note" -i "..." -c "..." --append-md

# Or let the agent do it — /run_learn reviews the session and proposes a rule
> /run_learn

# Disable a rule for one legacy project
dotai toggle no-useeffect --off --project legacy-app

# Audit rule quality (duplicates, empty bodies)
dotai rules --check

The feedback loop: vibe-code → agent makes mistake → /run_learn → structured rule → dotai sync → agent never repeats it.

Keeping rules healthy

Knowledge gets noisy as it grows. dotai audit reviews the effective ruleset locally and deterministically—nothing is sent to an external model.

# Find weak, generic, preference-like, oversized, or overlapping rules
dotai audit
dotai audit --project my-app

# Machine-readable output or a CI quality gate
dotai audit --json
dotai audit --fail-on high

The report includes stable finding codes, evidence, suggested actions, an estimated prompt-token cost, and a context-concentration summary showing which rules dominate the budget. Large-rule findings show the rule's share of total context, change risk, reviewable sections, and a conservative manual-savings range. Security and other high-consequence rules receive preservation-first guidance and are deliberately protected from “the model probably knows this” recommendations.

dotai compress turns high-confidence findings into a conservative plan. Semantic rewriting is handled by the active coding agent, not a hidden provider inside the CLI:

# Inspect candidates
dotai compress

# Ask the current coding agent to draft, diff, and review changes
> /run_compress

# Low-level application boundary for an approved versioned proposal
dotai compress apply proposal.json

/run_compress reads the JSON plan, drafts shorter complete rules, explains what it preserved and consolidated, shows the full diff in conversation, and asks Apply this compression? [y/N]. It then passes only approved changes to dotai. Near-duplicates, generic guidance, and rules that may belong in a preference pack remain suggestions unless the developer explicitly approves a concrete proposal.

When no exact duplicates exist, compression says there are no safe automatic savings while still listing semantic review candidates and their estimated savings. These estimates are directional—they help prioritize review. The CLI never invents a semantic rewrite or calls an external provider.

Proposal files are versioned and bound to the original rule with a SHA-256 hash, so stale or renamed rules are rejected. Before applying approved changes, dotai validates every proposed rule and creates a complete timestamped snapshot under the applicable .ai/backups/<timestamp>-compress/. After writing, the manifest is finalized with actual before/after hashes, per-rule token savings, total savings, and completion status. Subsequent audits show recently compressed rules and clearly note when a successful reduction remains above the large-rule threshold. Backups are ordinary local files and are never deleted automatically.

Migrating existing agent files

Already have a CLAUDE.md, .cursorrules, or AGENTS.md? Import user-authored content into ~/.ai/:

# Preview
dotai import-agent CLAUDE.md --dry-run

# Append into project .ai/rules.md (default)
dotai import-agent CLAUDE.md

# One structured rule
dotai import-agent .cursorrules --mode rule --name project-conventions

# One rule per ## section
dotai import-agent AGENTS.md --mode sections --dry-run

# Scan a project directory for known agent files
dotai import-agent . --mode rules_md

dotai strips its own managed marker sections so you don't re-import generated primers. Directory imports combine all discovered agent files into one update, and structured-rule imports choose a new suffixed filename rather than overwrite an existing rule. Preview unfamiliar files with --dry-run before importing.

Preference packs (taste) — borrowable soft style

Hard rules are law. Preference packs are soft taste: CLI stack, design micro-details, export style — things that are too granular or fluid for hard rules. You can author them, pull someone else's, and activate them per project.

Precedence: hard rules → freeform rules.md → active preference packs → model default.

# Create a pack
dotai prefs new "CLI Conventions" --domain cli
# Edit ~/.ai/preferences/cli-conventions.md, then activate:
dotai prefs use cli-conventions
dotai sync

# Borrow / pull someone else's taste (local file or git repo)
dotai prefs pull ./design-eng-taste.md --name design-eng --domain design
dotai prefs pull https://github.com/org/taste-packs -n cli
dotai prefs use design-eng

# Session overlay without changing active list
dotai sync --with-prefs design-eng,cli
dotai primer --with-prefs design-eng

# List / show / deactivate
dotai prefs
dotai prefs show cli-conventions
dotai prefs unuse design-eng

# Project operations stay inside that project; global removal is explicit
dotai prefs use cli-conventions --project my-app
dotai prefs remove cli-conventions --global

Preference lookup and removal are scope-safe: a project command cannot mutate a pack owned by another registered project. Removing a pack also removes it from the activation list for the pack's owning scope.

Example pack (~/.ai/preferences/cli.md):

---
name: CLI Conventions
id: cli
description: How I build CLIs
domain: cli
tags: typescript, commander
---

## Soft preferences

Hard rules always take precedence if they conflict.

### Stack
- TypeScript + tsup
- Commander.js
- Vitest
- pnpm (npm link for local bins)

### Style
- Lowercase `-v` for version
- Start version at `0.0.1`
- Commands live under `commands/`

Searching your knowledge

# Search by keyword
dotai search "useEffect"

# Filter by type
dotai search "review" --type skill

# Filter by tag
dotai search --tag security

# Combine filters
dotai search "auth" --type rule --tag react

Multi-project setup

# Initialize project-local overrides
dotai init ~/my-project
dotai init ~/other-project

# Each project can have its own rules, roles, and skills in .ai/
# Project-level rules override global ones during sync

Directory Structure

~/.ai/                          # Global (cross-project)
├── rules.md                    # Inline rules, conventions, and lessons learned
├── rules/                      # Structured rules (hard constraints)
│   └── no-useeffect.md         # Example: ban useEffect in React
├── preferences/                # Soft taste packs (borrowable style priors)
│   └── cli.md                  # Example: CLI stack / micro-style
├── preferences-active.json     # Which preference packs are active
├── backups/                    # Local snapshots created before compression writes
├── roles/                      # Cognitive modes
│   ├── reviewer.md             # Paranoid staff engineer
│   ├── architect.md            # Systems thinker
│   ├── qa.md                   # Methodical tester
│   ├── founder.md              # Product visionary
│   ├── ship.md                 # Release engineer
│   ├── writer.md               # Documentation specialist
│   ├── debugger.md             # Root cause analyst
│   ├── security.md             # Application security specialist
│   ├── mentor.md               # Patient teacher and pair programmer
│   └── product-manager.md      # Scoping, acceptance criteria, prioritization
├── skills/                     # Reusable workflows
│   ├── review.md               # Code review (/run_review)
│   ├── commit-helper.md        # Conventional commit messages (/run_commit)
│   ├── context-dump.md         # Context dump for new sessions (/run_context)
│   ├── parallel-work.md        # Git worktree parallel dev (/run_parallel)
│   ├── ship.md                 # Sync, test, push, PR (/run_ship)
│   ├── techdebt.md             # Find tech debt (/run_techdebt)
│   ├── careful.md              # Production-safety mode (/run_careful)
│   ├── investigate.md          # Root-cause analysis (/run_investigate)
│   ├── scaffold.md             # Boilerplate generation (/run_scaffold)
│   ├── verify.md               # Run tests, types, lint (/run_verify)
│   ├── plan.md                 # Structured planning workflow (/run_plan)
│   ├── learn.md                # Capture learnings as rules (/run_learn)
│   ├── compress.md             # Review semantic rule compression (/run_compress)
│   └── deploy/                 # Folder-based skill (with scripts & assets)
│       ├── main.md
│       ├── scripts/
│       ├── assets/
│       └── config.json
└── tools/                      # Python tool implementations
    └── *.py

<project>/.ai/                  # Per-project overrides
├── rules.md                    # Project-specific rules and conventions
├── rules/                      # Project-specific structured rules
├── roles/                      # Project-specific roles
├── skills/                     # Project-specific skills
└── tools/                      # Project-specific tools

Skills

Skills are reusable AI workflows organized into 9 categories:

Category	Purpose	Example
reference	Library/CLI documentation lookups	API docs, framework guides
verification	Testing, validation, type-checking	Test runners, lint checks
data	Dashboards, queries, monitoring	Query templates
workflow	Multi-step automation	/run_parallel, /run_context
scaffolding	Boilerplate / code generation	Project templates
code-quality	Review, linting, style enforcement	/run_review, /run_techdebt
deployment	CI/CD, release, ship	/run_ship
debugging	Investigation, root-cause analysis	Root-cause workflows
maintenance	Operational procedures, migrations	Runbook skills

Skill Triggers

All skill triggers use the run_ prefix to avoid collisions with built-in agent commands:

Skill	Trigger	Description
Code Review	/run_review	Diff-based structural review
Commit Helper	/run_commit	Conventional commit messages
Context Dump	/run_context	Session context for onboarding
Parallel Work	/run_parallel	Git worktree management
Ship	/run_ship	Sync, test, push, create PR
Find Tech Debt	/run_techdebt	Identify debt and duplication
Careful Mode	/run_careful	Production-safety guardrails
Investigate	/run_investigate	Systematic root-cause analysis
Scaffold	/run_scaffold	Generate boilerplate from patterns
Verify	/run_verify	Run tests, types, lint, build
Plan	/run_plan	Structured planning before coding
Learn	/run_learn	Capture mistakes as permanent rules

Skill Format

Skills support both structured and runbook formats:

---
name: Code Review
trigger: /run_review
role: reviewer
category: code-quality
allowed-tools: Read, Grep, Glob, Bash
context: local, ci
tags: review, quality
---

Analyze the current branch's diff for issues that tests don't catch.

## Gotchas

- Large diffs (>500 lines) should be split into per-file reviews
- Always verify the base branch before diffing

## Steps

1. Detect the base branch
2. Read the full diff
3. Check for security issues, error swallowing, race conditions
4. Generate a structured report with file:line references

Gotchas

Skills should document common failure points — things that push the agent out of its normal way of thinking. Gotchas are rendered with warning markers in the agent prompt so they get extra attention.

Conditional Contexts

Skills can specify which contexts they're active in:

context: production, sensitive

This lets you create production-safety skills that activate extra caution for operations affecting live systems.

Folder-Based Skills

For complex workflows, skills can be directories with helper scripts and assets:

~/.ai/skills/deploy/
├── main.md          # Skill definition (frontmatter + body)
├── scripts/         # Shell/Python scripts the agent can invoke
│   ├── health-check.sh
│   └── rollback.py
├── assets/          # Templates, reference docs, configs
│   └── pr-template.md
└── config.json      # User-specific configuration

Folder-based skills enable progressive disclosure — the agent reads main.md for the workflow, then pulls in scripts only when needed. This keeps token usage efficient.

Roles

A role is a cognitive mode — a persona that frames how the AI approaches a task.

Role Format

---
name: Paranoid Reviewer
description: Staff engineer focused on production safety
tags: review, security, quality
---

You are a paranoid staff engineer reviewing code before it lands in production.
Your job is to find bugs that tests don't catch...

## Principles
- Assume every external input is hostile
- Check error paths, not just the happy path

## Anti-patterns
- Commenting on style — that's the linter's job

Skills reference roles by ID — when a skill runs, the role's full persona is injected into the prompt.

Composing Skills with Roles

Skills and roles can be composed inline. Pass as <role> to run any skill with a specific persona:

/run_plan as product-manager         # Scope a feature with acceptance criteria
/run_plan as systems-architect      # Architecture deep-dive with tradeoffs
/run_review as paranoid-reviewer    # Security-focused code review
/run_review as security-engineer    # Full OWASP-style security audit
/run_techdebt as debugger           # Hunt tech debt with a debugger's mindset
/run_review as mentor               # Review that teaches, not just critiques
/run_review                         # Uses the skill's default role, or none

From the CLI:

dotai prompt review --role paranoid-reviewer | pbcopy
dotai prompt review --role qa > /tmp/prompt.md

Rules

Rules are structured coding standards that agents enforce automatically.

Structured Rules

Individual rule files live in ~/.ai/rules/ with frontmatter:

---
name: no-useEffect
description: Never call useEffect directly
globs: "*.tsx, *.ts"
tags: react, hooks
enabled: true
---

All useEffect usage must be replaced with declarative patterns...

Managing Rules

# List active rules
dotai rules

# Audit duplicates / empty bodies
dotai rules --check

# Import a rule from an external file
dotai learn "no-useEffect" --from-file react-rule.md --globs "*.tsx,*.ts"

# Record a learning as a structured rule (default)
dotai learn "auth-header" --issue "Forgot Bearer prefix" --correction "Always prepend Bearer to tokens"

# Preview first
dotai learn "auth-header" -i "..." -c "..." --dry-run

# Disable a rule globally
dotai toggle no-useeffect --off

# Disable a rule for a specific project only
dotai toggle no-useeffect --off --project my-legacy-app

# Re-enable
dotai toggle no-useeffect --on

Structured rules are included inline in every synced agent file. Freeform rules.md is included too. Skills/roles stay as catalogs unless you pass dotai sync --full.

Agent Sync

dotai sync generates tool-specific bootstrap files with:

• Full structured rule bodies (high-value constraints)
• Freeform rules.md conventions (previously referenced but not inlined)
• Role and skill catalogs (names, triggers, descriptions — not every skill novel)
• Claude Code: compact pointers + native .claude/skills/ slash commands

Use --full only when you need every role persona and skill definition dumped inline (larger context).

These files contain machine-specific context and may include absolute paths. Treat them as generated per-user artifacts unless your team intentionally commits shared agent instructions. dotai-managed sections are marker-delimited, so syncing preserves user-authored content outside those markers.

# Generate all (CLAUDE.md + .cursorrules + GEMINI.md + AGENTS.md)
dotai sync

# Generate for specific agents
dotai sync --agents claude,cursor
dotai sync --agents gemini

# Legacy-style full dump (roles + skill definitions inline)
dotai sync --full

# Print primer to stdout (for piping)
dotai primer
dotai primer --compact
dotai primer --full

Supported Agents

Agent	Output File	Sync Command	Notes
Claude Code	CLAUDE.md + .claude/skills/	dotai sync --agents claude	Marker-based merging + native slash commands
Cursor	.cursorrules	dotai sync --agents cursor	Full rules and preferences; role/skill catalog by default
Gemini CLI	GEMINI.md	dotai sync --agents gemini	Auto-discovered in project root and ~/.gemini/GEMINI.md
Generic	AGENTS.md	dotai sync --agents generic	Works with Codex, Copilot, and any agent that reads it

For Claude Code, dotai sync also generates .claude/skills/<trigger>/SKILL.md files. These register as native slash commands — they appear in autocomplete and work as real /run_* commands with role composition support built in.

Tip: Run dotai primer --full | pbcopy to copy the complete context to your clipboard for pasting into an agent's system prompt.

Installing Skills

Install skills from git repos or local directories. Claude-native SKILL.md files are auto-detected and converted to dotai format with triggers and categories inferred from the content. New installs are security-vetted before use.

# Install from a GitHub repo — auto-detects and converts Claude-native skills
dotai install https://github.com/slavingia/skills

# Install a specific skill from a repo
dotai install https://github.com/user/ai-skills -s deploy

# Install from a local directory
dotai install ~/my-skills/review

# Install to a specific project instead of global
dotai install https://github.com/team/skills -p my-project

After install, run dotai sync to make them available as slash commands in Claude Code, Gemini, etc.

Refresh skills that were installed from a tracked source (team conventions repo):

dotai install --update
dotai install --update -s mvp

Updates are staged and security-vetted before installation. A blocked or declined update leaves the existing skill unchanged. Multi-skill repositories refresh each recorded skill independently. --skip-vet is an explicit trust override and should only be used for a source you have reviewed.

Converting Claude-native skills

If you have a Claude-native SKILL.md and want to convert it manually:

# Preview what would be converted
dotai convert /path/to/SKILL.md --dry-run

# Convert to global skills
dotai convert /path/to/SKILL.md

# Convert to a specific project
dotai convert /path/to/SKILL.md -p my-project

# Convert to a custom directory
dotai convert /path/to/skill-dir/ -o ~/my-skills/

The converter maps Claude-native fields (compatibility, license) to dotai format, auto-generates a /run_* trigger, and infers a category from the skill's name and description.

Importing from plugin ecosystems

dotai can import skills from Claude plugins, Cursor plugins, and Gemini extensions — converting them into universal dotai format so they work across all tools.

# Import from a Claude Code plugin
dotai import-plugin /path/to/claude-plugin/
dotai import-plugin https://github.com/user/claude-plugin

# Import from a Cursor plugin
dotai import-plugin /path/to/cursor-plugin/

# Import from a Gemini CLI extension
dotai import-plugin /path/to/gemini-extension/

# Import only a specific skill from a plugin
dotai import-plugin /path/to/plugin/ -s review-code

# Also convert agents to dotai roles
dotai import-plugin /path/to/plugin/ --include-agents

# Also convert Cursor .mdc rules to dotai rules
dotai import-plugin /path/to/cursor-plugin/ --include-rules

Supported plugin formats:

• Claude plugins — directories with .claude-plugin/plugin.json
• Cursor plugins — directories with .cursor-plugin/plugin.json
• Gemini extensions — directories with gemini-extension.json

The importer discovers skills, commands, agents, and rules within the plugin and converts each to dotai format. Hooks and MCP server configs are noted but not converted (they're tool-specific).

Removing skills

# Remove a skill by name (will ask for confirmation)
dotai remove mvp

# Remove without confirmation
dotai remove mvp --force

# Remove from a specific project
dotai remove mvp -p my-project

Tracking sources

Installed skills remember where they came from. Use dotai skills to see the source column:

dotai skills
# Shows: Name, Trigger, Category, Role, Scope, Source, Description
# Source shows "slavingia/skills" for GitHub imports, path for local installs

Creating Skills

# Simple single-file skill
dotai new-skill "Deploy" -t /run_deploy -c deployment

# Folder-based skill with scripts and assets
dotai new-skill "Deploy" -t /run_deploy -c deployment --folder

CLI Reference

dotai init [project-path]              # Initialize ~/.ai/ or project .ai/
dotai roles                            # List available roles
dotai role <name>                      # Output a role's full prompt to stdout
dotai skills [-c category]             # List skills (optionally filter by category)
dotai prompt <skill> [--role <role>]   # Assemble skill + role prompt for any agent
dotai sync [path] [--agents ...] [--full]  # Sync ~/.ai/ into agent config files
dotai primer [--project <name>] [--full|--compact]  # Print agent context to stdout
dotai rules [-p project] [-a] [--check]    # List rules (or audit quality)
dotai audit [-p project] [--json] [--fail-on severity]  # Read-only rule audit
dotai compress [-p project] [--json]       # Plan compression candidates
dotai compress apply <proposal.json> [--yes]  # Validate, back up, and apply proposals
dotai toggle <rule-id> --on/--off      # Enable/disable rules globally or per-project
dotai learn "title" -i "..." -c "..."  # Create structured rule (default)
dotai learn "title" -i "..." -c "..." --dry-run|--sync|--force|--append-md
dotai learn "title" --from-file <f>    # Import structured rule from a file
dotai import-agent <path> [-m mode] [--dry-run]  # Non-destructive agent-file migration
dotai prefs [list|show|new|pull|use|unuse|remove]  # Preference / taste packs
dotai prefs new "CLI" --domain cli     # Create a soft style pack
dotai prefs pull <path|url> [-n id]    # Borrow / install a pack
dotai prefs use <id> [-p project|--global]  # Activate pack in an explicit scope
dotai sync --with-prefs a,b            # Session overlay of preference packs
dotai install <source> [-s skill]      # Install skills from git repo or local path
dotai install --update [-s skill] [--skip-vet]  # Vetted refresh from recorded sources
dotai import-plugin <source>          # Import from Claude/Cursor/Gemini plugin
dotai remove <skill> [-p project]     # Remove an installed skill
dotai convert <path> [--dry-run]      # Convert Claude-native SKILL.md to dotai format
dotai new-skill <name> [-t trigger]    # Create a new skill from template
dotai watch [path] [--agents ...]       # Auto-resync on ~/.ai/ changes
dotai search "query" [--type t] [--tag] # Search knowledge index
dotai index [--refresh]                # Build/show knowledge index
dotai tree                             # Show knowledge tree

Philosophy

• Model-agnostic — Works with Claude, Cursor, Gemini, Codex, Copilot, Ollama, or any LLM
• File-based — No databases, no servers, just markdown files in ~/.ai/
• Composable — Roles are reusable across skills, skills across projects
• Portable — Your knowledge travels with you, not locked into one tool
• Gotcha-first — Skills document failure points, not just happy paths
• Progressive disclosure — Folder-based skills keep token usage efficient

License

MIT — see LICENSE.