dotai

Personal agentic AI configuration — agent definitions, shared instruction files, prompts, and skills for OpenCode and GitHub Copilot.

Structure

agents-config/
  opencode/     ← OpenCode agent files
  copilot/      ← GitHub Copilot agent files
instructions/   ← shared instruction markdown files, appended to agents via prompt_append
prompts/        ← reusable prompt snippets
skills/         ← reusable skills

~/.config/opencode/agents and ~/.copilot/agents are symlinks into this repo — edits here take effect immediately with no sync step.

Agents

Agent	Available in	Description
commit	OpenCode, Copilot	Commits staged changes with a conventional commit message (≤50 chars) and pushes
create-pr	OpenCode, Copilot	Creates a GitHub PR from the current branch, filling in a template if one exists
review	OpenCode, Copilot	Reviews a GitHub PR given a URL; asks for confirmation before posting
cm-ticket	OpenCode, Copilot	Creates a CM Jira ticket from one or more GitHub PR URLs
release-plan	OpenCode, Copilot	Creates a Confluence release plan page from a Jira CM ticket
sync-docs	OpenCode, Copilot	Syncs README, copilot-instructions, and agent files to match current project state

Setup

OpenCode

ln -s ~/ws/dotai/agents-config/opencode ~/.config/opencode/agents

GitHub Copilot

ln -s ~/ws/dotai/agents-config/copilot ~/.copilot/agents

Adding a new agent

1. Create agents-config/opencode/<name>.md with YAML frontmatter + body (see below)
2. If it needs shared instructions, add them to instructions/ and reference via prompt_append
3. Include a fallback read line in the body (see Conventions)
4. Mirror to agents-config/copilot/<name>.md if Copilot support is needed

OpenCode agent file format

---
name: my-agent
description: One-line summary shown in agent picker
mode: subagent
temperature: 0
prompt_append: "file:///Users/ashumon/ws/dotai/instructions/<file>.md"
permissions:
  bash: allow | ask | deny
  edit: allow | ask | deny
  webfetch: allow | ask | deny
  task: ask | deny
---

Agent body instructions here...

Conventions

prompt_append paths must be absolute

Use file:///Users/ashumon/ws/dotai/instructions/<filename> — not ~ or relative paths. OpenCode does not expand ~ in prompt_append.

Always include a fallback read line

prompt_append can silently fail. Every agent that references an instruction file must include this at the top of its body:

Read ~/ws/dotai/instructions/<file>.md with the Read tool before doing anything else, and follow it exactly.

Confirmation before external writes

Agents that write to Jira, Confluence, or GitHub must show full output and ask for explicit confirmation before submitting.

Permission defaults

• bash: ask for agents that run shell commands but shouldn't do so freely
• edit: deny for read-only agents (review, cm-ticket, release-plan)
• webfetch: deny unless the agent explicitly needs to fetch URLs

Instruction file style

Files in instructions/ are plain markdown. Use ## for section headings. No horizontal dividers between sections.

External integrations

• GitHub: all PR operations use the gh CLI