Worklog

Scan everything produced on the target date across GitHub (and Jira/Confluence when the Atlassian MCP is connected), then produce a per-project summary, ask about untracked work, and suggest how many hours to log.

Required flow

Always run in this order. Never skip a step:

1. Normalize the target date (ask if not given).
2. Collect data from the integrations (sections 1–3).
3. "Other work" checkpoint (section 4): ask the user and wait for the answer.
4. Estimate hours (section 5).
5. Write the final report (Output section) — only after steps 3 and 4.

Target date

Accept any format (2026-06-02, 06/02, "today", "yesterday") and normalize to YYYY-MM-DD. The window is the local day 00:00:00–23:59:59. GitHub timestamps are UTC: convert them to the local timezone (date +%Z) before filtering.

gh api user --jq .login

1. GitHub (gh CLI)

Main sweep via the Events API (commits, PRs, reviews, comments; includes private events when authenticated as yourself):

DAY=2026-06-02
gh api --paginate /users/<login>/events --jq ".[] | select(.created_at | startswith(\"$DAY\"))"

Also query the day before and after in UTC when the local offset pushes events across midnight.

Complement and dedupe (by URL/sha) with targeted searches:

gh search prs     --author=@me      --created=$DAY     --json title,url,repository,state
gh search prs     --author=@me      --updated=$DAY     --json title,url,repository,state
gh search prs     --reviewed-by=@me --updated=$DAY     --json title,url,repository
gh search commits --author=@me      --author-date=$DAY --json repository,sha,commit

Group by repository.

2. Jira (Atlassian MCP, optional)

Skip silently if the Atlassian tools are not available. Use getAccessibleAtlassianResources for the cloudId, then searchJiraIssuesUsingJql (JQL dates use yyyy/MM/dd):

(assignee = currentUser() OR status CHANGED BY currentUser() AFTER "2026/06/02" BEFORE "2026/06/03")
AND updated >= "2026/06/02" AND updated <= "2026/06/02 23:59"
ORDER BY updated DESC

For each candidate, getJiraIssue with expand=changelog and fields=comment, and keep only status changes and comments made by me on the target date. Group by project key.

3. Confluence (Atlassian MCP, optional)

searchConfluenceUsingCql (CQL dates use yyyy-MM-dd):

(creator = currentUser() OR contributor = currentUser()) AND type = page
AND lastModified >= "2026-06-02" AND lastModified < "2026-06-03"

Distinguish created from edited by comparing the creation date with the target date. Group by space.

4. "Other work" checkpoint (mandatory)

Integrations only see the digital trail. Before writing the report, ask with AskUserQuestion (both questions in one call):

1. Meetings — "Any meetings or calls on <date>?" Options: "No meetings" + "Other" (which, with whom, how long).
2. Untracked work — "Anything the integrations can't see?" E.g. environment setup, research, pairing, helping a teammate, local docs, devops. Options: "Nothing else" + "Other" (describe, with approximate time).

Rules:

• Never write the final report before the user answers.
• If the answer is "nothing", omit the section — but never skip the question.
• Never invent items; log only what the user reports, with the stated duration.

5. Hours

Suggest the value to log, showing the reasoning:

1. Active window: first to last activity timestamp of the day (local time), across all sources.
2. Breaks: subtract gaps longer than ~90 min with no events.
3. Other work: add meetings and untracked work that fall outside the active window; don't double count what falls inside.
4. Weighting: commits/PRs/reviews/tickets/pages signal focused work. Use them to adjust the range, not to inflate it.
5. Result: a range plus a suggestion rounded to 0.5h (e.g. range ≈ 6–7h → suggestion: 6.5h). Say clearly it is an estimate.

With several projects in the day, split the suggestion proportionally to each project's load.

Output

Write the report in the user's language. Omit empty sections and sources.

Each bullet states what was done and the substance (which problem, which solution, which module), not just counts:

• Bad: "3 commits on branch X"
• Good: "fix/PROJ-512-gateway-validation [PROJ-512] — 3 commits (api): payload validation fix, null edge-case test, error handler refactor; PR #142 opened."

Tickets always with key and short title; status as From → To. Merge repos, Jira projects and Confluence spaces that are the same project.

Sections, separated by ---:

1. Header — full date with weekday.
2. Summary — 2–4 sentences with the thread of the day.
3. Activities by project — subsections per source (GitHub / Jira / Confluence / Other).
4. Hours — reasoning, range and highlighted suggestion.
5. Ready to paste — compact version per project, no links, ready for the timesheet tool.