Atlassian CLI

Atlassian CLI (acli)

acli is a real, installed command-line tool that talks to the user's actual Jira and Confluence Cloud site. When a request touches Jira or Confluence, reach for acli instead of speculating about what a ticket might say or asking the user to go look it up — you can just look it up.

Every command has its own --help. Flags shift between acli releases, so if something in this skill doesn't match reality, run acli <command> --help and trust that output over this document.

Before doing anything

1. Check auth once per session, don't assume:

   acli jira auth status
   acli confluence auth status
   

   If not authenticated, tell the user and run acli auth login (opens an OAuth browser flow — this needs the user present, don't try to script around it).

2. Only one site is usually configured. If the user mentions a specific Atlassian site and the auth status shows a different one, flag the mismatch rather than silently querying the wrong site. Use acli auth switch / acli jira auth switch / acli confluence auth switch if they have multiple accounts configured.

3. Don't guess IDs. Project keys, board IDs, sprint IDs, space keys, and space/page IDs are all opaque numbers or short codes you can't reliably infer. Look them up first: acli jira project list, acli jira board search, acli confluence space list, etc. The one exception is a work item key the user already gave you (e.g. PROJ-123).

General patterns that apply everywhere

• Add --json when you need to parse the output yourself (e.g. to extract a field or feed it into another command). Skip it when just relaying a quick answer to the user — the default table/text output is often more readable for that.
• Add --web to open something in the browser instead of printing it, when the user wants to look at something rather than have you summarize it.
• Bulk/targeting flags are consistent across jira workitem subcommands: most of assign, edit, transition, clone, and comment accept a target via --key "KEY-1,KEY-2", --jql "...", or --filter <id> — pick whichever the user's request naturally maps to (a JQL query for "all bugs in PROJ assigned to me", explicit keys for "these three tickets").
• Mutating/bulk commands prompt for confirmation unless you pass --yes (or -y). Only pass --yes for actions the user has clearly already approved (e.g. they explicitly asked you to transition/assign/delete something) — for anything destructive or irreversible (delete, archive), state exactly what you're about to do first if there's any ambiguity about scope, especially for JQL/filter-based bulk targeting where the affected set isn't obvious from the command alone.
• JQL is the query language for Jira search (--jql on workitem search, edit, assign, transition, clone, sprint list-workitems). See references/jql.md for a cheat sheet if you're unsure of the syntax.

Jira quick reference

Full command tree and flags: references/jira.md. The most common operations:

A few things worth knowing before you reach for them:

• workitem create supports --from-json/--generate-json for structured/bulk-ish input, and create-bulk exists separately for creating many issues at once from CSV/JSON.
• --assignee "@me" self-assigns; --assignee default resets to the project default.
• Descriptions and comment bodies accept plain text or full Atlassian Document Format (ADF) JSON if the user needs rich formatting.

Confluence quick reference

Full command tree and flags: references/confluence.md. The most common operations:

Important gap: as of the currently installed acli version, confluence page only supports view — there is no page create or page update subcommand. If the user wants to create or edit a Confluence page (not a blog post), don't fabricate a command for it. Tell them acli doesn't support page authoring yet, run acli confluence page --help to confirm whether that's still true (acli gets new subcommands regularly), and fall back to pointing them at the Confluence web UI, or asking if a blog post (acli confluence blog create) would work instead.

Output formats

Most search/list commands support --json and often --csv in addition to the default human-readable table. When the user wants to pipe results somewhere else, save them to a file, or you need to extract specific fields programmatically, use --json and parse it rather than scraping the table output.