Agents
The JSON output contract and safety flags for coding agents.
Jira CLI is built to be driven by coding agents as well as humans. It works with Claude Code, OpenAI Codex, Cursor, or any coding agent or agent harness that can run shell commands.
The output contract
- Add
-o json(or-o yaml) to any list/get command for machine-readable output. - stdout carries only the requested data. stderr carries informational
lines ("Created ABC-1", the JQL that
issue listbuilt, warnings). Parse stdout; log stderr. - Non-zero exit codes signal failure; API errors preserve the upstream HTTP status.
Safety flags
| Flag | Env | Effect |
|---|---|---|
--read-only | JIRA_READ_ONLY=1 | Blocks every write command (create/edit/delete/transition/...) |
--no-input | JIRA_NO_INPUT=1 | Disables all interactive prompts; fails instead of prompting |
--quiet | JIRA_QUIET=1 | Suppresses informational stderr output |
# Safe, non-interactive read for an agent
JIRA_READ_ONLY=1 JIRA_NO_INPUT=1 jira issue search "project = ABC" -o jsonDuplicate-safe issue creation
jira issue create is safe to rerun by default. Before writing, it looks for
an issue created by the current user in the same project during the last 10
minutes with the exact issue type, summary, and parent. A match is returned
instead of creating another issue, even if the description was reformatted
between agent attempts.
Use --allow-duplicate only when two otherwise identical issues are
intentional:
jira issue create -p ABC --type Task --summary "Repeated title" --allow-duplicateConfiguration via environment
No auth login step is needed. Set credentials directly:
export JIRA_SITE=https://acme.atlassian.net
export JIRA_EMAIL=me@acme.com
export JIRA_TOKEN=xxxx
jira whoami -o jsonClaude Code skill
A ready-to-use skill lives at
jira/SKILL.md,
with command and API references under jira/references/.