| name | jira-comment |
| description | Write concise, precise JIRA comments in English that a junior developer can act on without asking follow-up questions, and that do not read like AI-generated text. Covers status updates, root-cause analysis, handover/how-to instructions, and review or decision rationale. Drafts the comment, self-checks it against known AI writing tells, shows it for confirmation, then posts it via the Atlassian MCP. Use when asked to: comment on a JIRA ticket, "write a JIRA comment", "add a comment to IF-123 / ITZVULN-456 / PROJ-789", "document this in JIRA", "update the ticket", "explain this in the ticket for a junior", or "post my findings to the ticket". Triggers on: "jira comment", "kommentar", "comment on <TICKET-KEY>", "update ticket", "document in the ticket".
|
| version | 1.0.0 |
| author | Lukas Benjamin Roth <lukasbenjamin.roth@strato.de> |
| license | proprietary |
| user-invocable | true |
| metadata | {"category":"workflow","tags":["jira","writing","communication","atlassian"],"complexity":"medium"} |
jira-comment
Produce one JIRA comment. English, short, concrete, readable by someone who joined the
team two weeks ago. It must not carry the fingerprints of AI-generated text.
Workflow
1. Identify target and type
Get the ticket key from the user's message. If none is given, ask for it.
Pick exactly one comment type. If the user did not say, infer it from what they just did:
| Type | Use when |
|---|
status | Reporting progress, a blocker, or a handoff of state |
analysis | Explaining a finding, root cause, or assessment |
handover | Telling someone how to continue, reproduce, or verify |
decision | Recording a review outcome or why an approach was picked/dropped |
Shapes and worked examples: references/comment-types.md. Read it before drafting.
2. Gather facts
Read the ticket before writing: jira_get_issue with comment_limit: 20. This prevents
repeating something already in the thread and shows the vocabulary the ticket already uses.
Pull the concrete details from the session: commands run, their output, file paths, commit
SHAs, PR URLs, image tags, versions, CI run links, error messages.
Never invent a fact to make a sentence complete. If a number, cause, or timeline is unknown,
either leave it out or say plainly that it is still open.
3. Draft
Apply the rules below and in references/ai-tells.md. Target 3-12 lines.
4. Self-check
Write the draft to a scratch file and run the checker. It lives next to this file, so
resolve it through the plugin root rather than the current directory:
SLOPCHECK="${CLAUDE_PLUGIN_ROOT:-$HOME/.claude/plugins/marketplaces/claude-private-skills/plugins/tools}/skills/jira-comment/scripts/slopcheck.py"
python3 "$SLOPCHECK" /tmp/draft.md
It flags banned vocabulary, AI transitions, contrast reframes, em dashes, tricolons, hedge
openers, resolution closers, and uniform sentence length. Rewrite whatever it flags and run
it again. Do not explain the flags to the user; just fix them.
