jira-comment

Jira Comment

One job: take comment text the user already has (or just authored) and put it on a Jira ticket so it RENDERS CLEANLY. The Atlassian MCP converts markdown to ADF and mangles common syntax, so the real work is making the text render-safe before posting. This skill does NOT write the analysis — it formats and posts text that already exists.

Inputs

• The ticket key (e.g. PROJ-42). If absent, ask.
• The Jira cloudId — discover at runtime from the project CLAUDE.md's Jira: cloudId=<uuid> key=<KEY> line. NEVER hardcode a cloudId: it is per-site, and a wrong one posts to someone else's Jira. A repo whose CLAUDE.md has no Jira: line has no Jira configured — ask the user for the cloudId rather than guessing.
• The comment body. If the user just authored it in the conversation, use that verbatim as the source.
• Any CC mentions — preserve EXACTLY as given (e.g. User:<accountId>). These are live account-id references; do not reword, reformat, or drop them.

Render-safe rules (apply BEFORE posting — this is the whole point)

The MCP eats these; rewrite the text so they don't appear:

1. No **bold** — it leaks as literal \*\*word\*\*, and a multi-word bold run becomes a ** between every word. Use a plain label line instead (e.g. Mechanism: on its own line), not inline bold.
2. No fenced code blocks — the ``` fence is lost. Indent code/data 4 spaces instead so it stays monospace-ish and visually grouped.
3. Underscores in CONSTANT_NAMES still partly escape (_ reads as italic). It's the one residue you can't fully kill via the API — leave the name as-is, and if it matters, tell the user to retype that token in the Jira web editor.
4. No _italic_, no # headers, no - bullet markdown that relies on rendering. Plain text + indentation + blank-line separation only.
5. Inline literals (Locale.US, method names) may auto-linkify — harmless, mention only if asked.

Keep the prose, the meaning, and the CC line. Only strip the markup that won't survive.