agents-init

Shared instructions for AI coding agents.

From the root of any target repository, run:

curl -fsSL https://raw.githubusercontent.com/dani-polani/agents-init/main/install.sh | sh

The command downloads the current AGENTS.md, CLAUDE.md and COPYRIGHT.md from this repository and replaces files with the same names in the target repository.

If the target repository already uses a task workflow, the command also refreshes its workflow file and the ## Tasks section of PROJECT.md from the matching templates, reading the project values back from the existing **...:** lines and leaving every other section of PROJECT.md alone.

• .agents/tools/linear-tasks.md exists → the Linear workflow is refreshed from init-linear-tasks templates (values: **Linear project:**, **Project id:**, **Project label:**, **Default team:**, **Workspace id:**).
• Otherwise .agents/tools/obsidian-tasks.md exists → the Obsidian workflow is refreshed from init-obsidian-tasks templates (value: **Project name:**); the gitignored obsidian-tasks.local.md is never touched.

If both files are present, only the Linear one is refreshed and a warning is printed. Repositories with neither are unaffected.

It also adds an agentsmd target to the target repository's Makefile (creating the file if needed, skipping if the target already exists). Run it to pull the latest instructions later:

make agentsmd

Skills

The skills/ directory holds reusable Agent Skills shared across agents. These are not touched by install.sh / make agentsmd; they are installed separately with the skills CLI, which creates one canonical copy and symlinks it into each agent (Claude Code, Codex, Cursor, …). OMP discovers skills from those agent directories automatically.

• orca-orchestrate — delegate work to real agents running inside Orca: launch them in Orca worktrees and terminals through the orca CLI (not Claude's built-in subagents), monitor them, and manage their lifecycle. Routes to Orca's version-matched orca-cli and orchestration skills and adds the launch/monitor/cleanup playbook and the JSON-parsing gotchas on top.
• herdr-orchestrate — orchestrate a fleet of coding subagents inside herdr: pick agents by usage, give each a worktree/pane, launch them so herdr sees their state (no false idle), supervise, review, and deliver deployable PRs.
• init-linear-tasks — bind a repo to a Linear project and set up the Linear task workflow: create the committed .agents/tools/linear-tasks.md and the ## Tasks section of PROJECT.md holding the project name, project id, project label, default team key and workspace id. Everything is committed and machine-agnostic, so a checkout on another machine needs no setup beyond Orca being connected to the same Linear workspace.
• linear-task-done — close out an issue once work is finished: read the PRs Linear's own GitHub/GitLab integration linked, post a completion comment for this round of work, and move the issue to In Review. In Review is the agent's terminal status; Done and Canceled stay the user's call.
• init-obsidian-tasks — superseded by init-linear-tasks; kept for repositories still on the vault workflow. Set up the Obsidian (TaskNotes) task-management workflow in a repo: create the committed .agents/tools/obsidian-tasks.md and ## Tasks section of PROJECT.md, plus a gitignored .agents/tools/obsidian-tasks.local.md holding the machine-specific vault path. The vault path stays out of git, so the same repo works across machines without conflicts; on a new machine, re-running the skill just regenerates the local file.
• obsidian-task-done — superseded by linear-task-done; kept for repositories still on the vault workflow. Close out a task once work is finished: move it to review, append a time entry, write a body summary, and add a Changelog line on the project page. Resolves the vault and project from the files init-obsidian-tasks created, so it stays machine-agnostic.
• set-tools-color-schemes — methodology to sync Ghostty / herdr / yazi / micro to a VS Code or Cursor theme via per-tool configs (not “inherit terminal”): shared palette, semantic ANSI, hex chrome, and micro’s MICRO_TRUECOLOR requirement.

Install all skills from this repo globally (available in every project):

npx skills add dani-polani/agents-init -g

Install a specific skill, or target specific agents:

# one skill
npx skills add dani-polani/agents-init -g --skill herdr-orchestrate

# choose agents explicitly (default: all detected)
npx skills add dani-polani/agents-init -g --skill herdr-orchestrate -a claude-code -a codex -a cursor

To update later, re-run the same npx skills add command — it refreshes the canonical copy from this repo.