TWG CLI

Agent skills

TWG CLI ships installable skills — context files that teach your coding agent how to work with Atlassian products. Install them once, then just describe what you want in plain language.

Install skills

The TWG installer installs skills automatically. It always writes the canonical skill bundle to ~/.agents/skills, which is used by Codex, Cursor, Gemini CLI, GitHub Copilot, Rovo Dev, and other agents that support the universal .agents/skills layout.

If it detects an agent that needs its own skill directory, it also copies the bundle there. For example, Claude Code receives a copy under ~/.claude/skills when Claude Code is already installed.

To repair or refresh skills manually, run:

twg skills install --global

Manual installs can also target a specific additional agent:

twg skills install --global --agent claude

Claude

Claude Code stores skills in .claude/skills/. If Claude Code is detected during TWG install, the TWG skill is placed there automatically. After that, open Claude Code in your project and start asking:

"What are my open Jira issues?" "Summarize the work my team did this week." "Create a Confluence page summarizing our Q2 goals."

Codex

OpenAI Codex reads skills from the universal .agents/skills/ directory, which is always included in the install. No extra step is needed. Then ask Codex:

"Show me the current sprint and any blockers." "Find PRs I authored this month." "What goals are at risk for my team?"

Gemini

Gemini CLI also reads from .agents/skills/ automatically. Then:

"Who do I collaborate with most?" "Get full context on PROJ-123 — linked PRs, docs, and dependencies." "Show my reporting chain."

Other agents

The installer also supports Cursor, GitHub Copilot, Rovo Dev, Augment, Kiro CLI, OpenCode, and any agent that supports a skills or context directory. The universal .agents/skills/ location works out of the box for most agents. Additional-agent copies are installed only when that agent is detected locally or when you request it with --agent.

What gets installed

Skills are placed under .agents/skills/ and detected or requested agent-specific locations as focused sibling skill directories:

Layout Purpose
.agents/skills/twg/SKILL.md Root TWG operating contract, discovery, output handling, and shared safety
.agents/skills/twg-jira/ and .agents/skills/twg-confluence/ Product semantics and write safety
.agents/skills/twg-<workflow>/ Focused workflows such as status rollups, context discovery, engineering work, and operational health
.agents/skills/<skill>/references/ Detailed guidance loaded on demand

Your agent reads these automatically. No special prompt prefix is needed.

Skill and command discovery

Agents already receive the installed top-level skill metadata. When the detailed guidance inside a product or workflow skill is unclear, they can search the unloaded reference catalog locally:

twg help discover-skills "snapshot token editing"
twg help discover-skills "JQL sprint prioritization" --skill twg-jira

Skill discovery returns one primary match and at most two close alternatives. Each match identifies the owning skill and exact reference to load. Search uses curated reference titles and descriptions rather than arbitrary reference prose. After loading that guidance, agents use command help instead of repeating discovery.

Agents should use twg help directly for command discovery:

twg help
twg help confluence content get
twg help jira workitem
twg help describe "jira"
twg help describe "jira workitem get"

Use twg help <terms> before guessing command names, aliases, arguments, or options. Use twg help describe <path> before unfamiliar commands where exact arguments, choices, defaults, or examples matter. Namespace paths return compact YAML routing maps; exact executable command paths return JSON contracts by default. Use rendered --help only when a human-readable command page is specifically needed.

Next steps