Skills Export

This repository contains a standalone set of Codex-oriented skills plus portable exported agent configuration files that can be versioned in git and installed into either a local Codex or Claude Code environment.

Each exported skill source lives in skills/<skill>/ with a SKILL.md entry point and optional references or agent metadata. Repository-root agents/*.toml files provide exact tracked backups of custom Codex subagent profiles that can be installed back into a Codex environment as a group or individually.

The exported export/AGENTS.md enforces a plan-first workflow by default, uses PLAN.md for normal single-agent work, switches to visible PLAN.<short-id>.md files when concurrent agents need separate plan surfaces, requires fully completed approved work to delete the chosen plan file by default so stale plans do not look active to later agents, allows clearing the file only as a narrow fallback when deletion cannot be used, treats that fallback completion marker as inactive for future PLAN.md selection, supports an explicit #no-plan escape hatch when the user intentionally wants to skip planning, and allows a narrow chat-approval exception for very simple requests that clearly reduce to one localized mutating command, including tightly bounded deletions when the exact paths are enumerated in chat before approval. It also instructs agents to align generated code with Micah Halter's explicit, reader-oriented coding style, to maintain a compact ## Agent Memory section in project-local AGENTS.md files for durable, evidence-backed project context without overwriting maintainer-managed guidance, to prefer inline edits to existing files over delete-and-recreate workflows during execution, and to use project-local TASK.md files as durable prioritized backlogs that agents can update conservatively and self-assign from only when the user explicitly delegates backlog selection. When a workflow is expected to include mutating git steps such as staging or committing, those exact commands must be listed in the approved plan up front instead of being left implicit. Plan approval covers only post-approval actions listed in the plan and does not replace separate runtime sandbox or platform permission prompts.

The exported export/config.toml is a portable Codex settings file. It keeps only machine-agnostic settings that are intended to be shared across systems, including shared model defaults, [agents] settings, [notice.model_migrations], and [tui], and the installer merges those settings into an existing target config.toml instead of overwriting machine-specific local entries.

The repo-root AGENTS.md is repository-local policy for maintaining this source-of-truth repository. Installed global agent policy is sourced from export/AGENTS.md.

Exported top-level config

• export/AGENTS.md Exported agent policy for Codex and Claude Code.
• export/config.toml Exported portable Codex settings, including shared model defaults, [agents], [notice.model_migrations], and [tui], installed only for Codex targets.
• agents/*.toml Repository-root Codex agent profile backups installed only for Codex targets.

Agent profile roles

• scout Cheapest read-only locator for fast file and call-site lookup when completeness is not the priority.
• explorer Read-only repo explorer for deeper synthesis across multiple files.
• reviewer (medium) Lower-cost read-only first-pass validator for correctness, scope, and validation gaps before escalating to heavier reviewers.
• auditor, auditor (high) Read-only issue-finding auditors for evidence-backed defect discovery, with auditor (high) reserved for subtler or cross-cutting problems.
• spec reviewer Read-only PASS/FAIL reviewer for requirement compliance: nothing missing, nothing extra.
• quality reviewer Read-only PASS/FAIL reviewer for maintainability, correctness risk, and test quality after spec compliance is already satisfied.
• architect Read-only high-reasoning design agent for architecture, contracts, and decomposition work.
• implementer (medium) Lowest-cost write-capable profile for tightly bounded, low-risk patches with a clear file scope.
• implementer Default write-capable profile for normal coding tasks that need balanced rigor and autonomy.
• implementer (xhigh) Highest-rigor write-capable profile for risky refactors, tricky bugs, and other high-uncertainty changes.

Included skills

Google Workspace via gog

• gog Umbrella skill for the gog Google Workspace CLI. Use it as the entry point for Gmail, Calendar, Drive, Docs, Sheets, Slides, Forms, Contacts, Tasks, Workspace admin flows, auth setup, and troubleshooting.
• gog-calendar Calendar-specific gog workflows, including event lookup, creation, updates, invitations, free/busy checks, and scheduling support.
• gog-drive-docs Drive and document-centric gog workflows, including Drive file operations, Sheets, Docs, Slides, Forms, and Apps Script tasks.
• gog-mail Gmail-specific gog workflows, including message search, thread reads, attachments, drafts, replies, labels, filters, and mailbox operations.
• gog-productivity Productivity-oriented gog workflows for Contacts, Tasks, People, and time utilities.
• gog-workspace-admin Workspace-scoped and specialized gog workflows, including Chat, Keep, Groups, Classroom, gog config, and allowlist or sandbox-related setup.

Micah style skills

• micah-email-style Draft or revise emails in Micah Halter's writing style, especially direct, warm, person-to-person emails.
• micah-technical-writing-style Draft or revise technical documents in Micah Halter's technical writing style, especially research papers, reports, abstracts, introductions, methods sections, and style-matched revisions.

Repository and agent configuration

• agent-config Manage repository-first updates for skills and global configuration files, always resolving the source-of-truth repository first by using an explicit user path or the default $GIT_PATH/code.mehalter.com/mehalter/skills location before syncing approved repository-to-local changes into Codex or Claude environments.
• agent-orchestrator Coordinate large coding tasks through a phase-gated multi-agent loop that first follows the active project's planning policy, front-loads read-only discovery to build a complete execution plan, delegates implementation by cluster, and runs reviewer_medium before spec_reviewer and quality_reviewer.
• parse-recipe Parse plain-text recipes, recipe URLs, HTML recipe content, or existing Recipe JSON into a schema.org/Recipe JSON object while preserving ingredients and instruction wording and preferring fully structured recipeInstructions.
• mealie Interact with a self-hosted Mealie instance through its REST API for agent-driven recipe import and direct recipe authoring, recipe retrieval, meal plan operations, shopping list and shopping item workflows, recipe media updates, and general raw API requests using an exported Bash helper plus environment-provided credentials.

Installation

install.sh supports Codex and Claude Code targets.

Use Codex by default, or select Claude Code explicitly with --target claude.

The installer supports these install workflows:

1. Install the full Codex environment: all exported skills plus export/AGENTS.md and export/config.toml.
2. Install the full Claude Code environment: all exported skills plus export/AGENTS.md as CLAUDE.md.
3. Install all exported skills only.
4. Install export/AGENTS.md only.
5. Install all repository-root agent profiles only for Codex.
6. Install export/config.toml only for Codex.
7. Install only a specified set of skills.
8. Install only specified repository-root agent profiles for Codex.

For Codex, the installer writes skills into ${CODEX_HOME:-$HOME/.codex}/skills, writes export/AGENTS.md to ${CODEX_HOME:-$HOME/.codex}/AGENTS.md, and writes export/config.toml to ${CODEX_HOME:-$HOME/.codex}/config.toml. For Codex, the installer writes repository-root agents/*.toml files into ${CODEX_HOME:-$HOME/.codex}/agents. When a target config.toml already exists, the installer merges portable exported settings such as model, [agents], [notice.model_migrations], and [tui] into that file instead of replacing unrelated machine-specific settings.

For Claude Code, the installer:

• copies the exported skill directories into ${CLAUDE_HOME:-$HOME/.claude}/skills
• generates Claude subagent files in ${CLAUDE_HOME:-$HOME/.claude}/agents
• installs export/AGENTS.md as ${CLAUDE_HOME:-$HOME/.claude}/CLAUDE.md

Install the full environment:

./install.sh

Explicitly install the full environment:

./install.sh --env

Install the full environment for Claude Code:

./install.sh --target claude --env

Install all exported skills only:

./install.sh --skills

Install all exported skills only for Claude Code:

./install.sh --target claude --skills

Install export/AGENTS.md only:

./install.sh --agents

Install export/AGENTS.md only for Claude Code as CLAUDE.md:

./install.sh --target claude --agents

Install all repository-root agent profiles for Codex:

./install.sh --agent-profiles

Install export/config.toml only for Codex:

./install.sh --config

List available skills:

./install.sh --list

List available agent profiles:

./install.sh --list-agent-profiles

Install only selected skills:

./install.sh gog gog-mail micah-technical-writing-style

Install only selected agent profiles for Codex:

./install.sh --agent-profile architect --agent-profile implementer

Install only selected skills for Claude Code:

./install.sh --target claude gog gog-mail

Use a custom destination:

CODEX_HOME="$HOME/.codex" ./install.sh --env

Use a custom Claude destination:

CLAUDE_HOME="$HOME/.claude" ./install.sh --target claude --env

Show usage:

./install.sh --help

Installer behavior

• The installer discovers available skills by scanning skills/ subdirectories that contain SKILL.md.
• With no arguments, it installs the full Codex environment: all exported skills plus export/AGENTS.md and export/config.toml.
• --target codex installs into Codex paths.
• --target claude installs into Claude Code paths.
• --env installs the full environment explicitly for the selected target.
• --skills installs all exported skills without touching export/AGENTS.md.
• --agents installs only export/AGENTS.md.
• --agent-profiles installs all repository-root agents/*.toml files for Codex.
• --list-agent-profiles prints the available repository-root agent profile names.
• --agent-profile <name> installs one selected repository-root agent profile for Codex per flag occurrence.
• --config installs only the portable export/config.toml settings for Codex.
• With one or more skill names, it installs only those skills.
• For Codex, selected skills are copied into ${CODEX_HOME:-$HOME/.codex}/skills.
• For Claude Code, selected skills are copied into ${CLAUDE_HOME:-$HOME/.claude}/skills and a subagent file is generated for each selected skill in ${CLAUDE_HOME:-$HOME/.claude}/agents.
• Installing export/AGENTS.md replaces ${CODEX_HOME:-$HOME/.codex}/AGENTS.md for Codex or ${CLAUDE_HOME:-$HOME/.claude}/CLAUDE.md for Claude Code.
• Installing repository-root agents/*.toml files copies them into ${CODEX_HOME:-$HOME/.codex}/agents for Codex.
• Installing export/config.toml creates ${CODEX_HOME:-$HOME/.codex}/config.toml if missing, or merges portable exported settings such as model, [agents], [notice.model_migrations], and [tui] into the existing file for Codex.
• For each selected skill, the installer replaces any existing installed copy at the target path.
• For each selected agent profile, the installer replaces any existing installed copy at the target path.
• The installer exits with an error if any requested skill name is unknown or if incompatible mode flags and skill names are combined.