jira-daily-update - SKILL.md Agent Skill

name: jira-daily-update description: >- Post a daily work summary as a Jira comment by collecting GitHub commits/PRs and Cursor chat history for the day. Use when the user asks to update a Jira ticket with today's work, log daily progress on a ticket, post a Jira comment with what was done, or runs /jira-daily-update.

Jira Daily Update

Collect today's GitHub activity and Cursor chat history, then post a concise comment on a Jira ticket summarizing what was accomplished.

Usage

/jira-daily-update <JIRA-KEY> <owner/repo> [branch]

JIRA-KEY — required (e.g. RHOAIENG-12345)
owner/repo — required GitHub repository (e.g. openshift/kubeflow)
branch — optional; defaults to the repo's default branch

If the user omits arguments, ask for them before proceeding.

Step-by-Step

1. Resolve identities

Call GitHub MCP get_me to get the authenticated username.
Call Jira MCP getAccessibleAtlassianResources to get the cloudId.
Call Jira MCP getJiraIssue with cloudId and issueIdOrKey (the JIRA-KEY) to confirm the ticket exists. Use responseContentFormat: "markdown". Extract the issue summary for context.

2. Collect GitHub activity (today only)

Compute today's date as YYYY-MM-DD.

Commits:

Call GitHub MCP list_commits with owner, repo, sha (branch if given), and author set to the GitHub username from step 1.
Filter to commits where commit.author.date starts with today's date.
Keep: commit message (first line) and SHA (short, first 7 chars).
If no commits match today's date exactly, record zero commits — do NOT include commits from previous days.

Pull Requests:

Call GitHub MCP search_pull_requests with query: repo:<owner>/<repo> author:<username> updated:>=<YYYY-MM-DD>
Keep: PR title, number, state (open/closed/merged), URL.

3. Collect Cursor chat history (today only)

List ~/.cursor/projects/**/agent-transcripts/ using ls -lt and keep only directories whose modification timestamp matches today's date exactly.
For each matching transcript file, read JSONL lines and extract user messages:
- Prefer text inside <user_query>...</user_query> tags.
- Fall back to the first 200 chars of user message text.
A message only counts as today's activity if the transcript file was modified today — never pull messages from transcripts modified on earlier dates.
Distill into a short list of what the user worked on / asked the agent to do.
Deduplicate similar entries. Aim for 3-5 distinct activity bullets.
Ignore meta/skill invocations (e.g. /jira-daily-update, /update-internship-tasks, single-word messages, or messages that are only tool scaffolding).

4. Gate check — verify real activity exists TODAY

Before composing any comment, confirm that at least one of the following is true:

There is at least one GitHub commit authored today.
There is at least one PR updated today.
There is at least one transcript file modified today containing substantive user messages (coding questions, feature work, debugging, design decisions).

If none of these are true → STOP. Do NOT post a Jira comment. Tell the user:

"No activity found for today (). No Jira comment was posted."

CRITICAL: Never recycle or restate work from previous days as if it happened today. If the only evidence is old commits or old transcripts, that is zero activity for today — stop and report it.

5. Compose the Jira comment

Merge all sources into a concise progress update. Use this template:

**Progress update — <YYYY-MM-DD> (AI generated)**

- <bullet 1: what was done, with PR/commit link if applicable>
- <bullet 2>
- <bullet 3>
- ...

Rules:

3-5 bullets max. Each bullet is one sentence.
Lead with the most impactful work.
Inline GitHub links where relevant: [PR #123](https://github.com/owner/repo/pull/123).
Combine related commits + chat activity into a single bullet when they describe the same work.
Do NOT list raw commit SHAs unless there's no PR to link.
Do NOT include filler like "continued working on…" — be specific.
Only include work that happened today. Never summarise prior days' work.
If no activity found for a source, omit it silently.

6. Post to Jira

Call Jira MCP addCommentToJiraIssue with:
- cloudId from step 1
- issueIdOrKey — the JIRA-KEY
- commentBody — the composed comment
- contentFormat: "markdown"
Confirm success and show the user the posted comment text.

7. Show output in chat

Always display the final comment in chat so the user can review what was posted. Format it clearly with a heading like:

✅ Comment posted to <JIRA-KEY>:
<the comment body>

Failure Handling

If Jira MCP is unavailable or auth fails → output the comment in chat and instruct the user to paste it manually.
If GitHub MCP is unavailable → skip GitHub data, build comment from chat history only.
If no transcripts found for today → skip chat data, build comment from GitHub only.
If both sources are empty for today → inform the user: "No activity found for today."
If the Jira ticket key is invalid → report the error and stop.

Example

User runs: /jira-daily-update RHOAIENG-53408 openshift/kubeflow release-1.9

Posted comment:

**Progress update — 2026-04-01**

- Rebased SDK PR #368 on latest upstream and resolved merge conflicts
- Added unit tests for the new TrainingClient checkpoint API
- Investigated nvcc build failure blocking RHOAIENG-56094 ([PR #705](https://github.com/openshift/kubeflow/pull/705))