Standup Writer

Standup Writer

Reuses work-context Step 3 data gathering and tomorrow-calendar-accepted event filtering to auto-draft a standup update. Output is paste-ready text in chat — not a Canvas — because standups are fast and paste-and-go.

Skill Composition

What this skill adds: standup cadence detection, yesterday/today/blockers formatting, work-channel filtering, blocker detection, and agent transcript scanning. It skips Drive, Confluence, and Second Brain — those sources are not standup-relevant.

Read-Only Constraint

This skill MUST NEVER write, create, update, or modify data in any external system. All MCP tool calls must be read-only operations. See work-context for the full allowed operations list.

Autonomy Principle

This skill runs fully autonomously after Step 1. No user prompts during data gathering (Steps 2-5). If a query fails, log the error silently and continue with remaining sources.

The only acceptable user prompts are in:

• Step 1 — confirming the standup scope (if ambiguous)
• Step 6 — presenting the draft

Prerequisites

Same as work-context, minus Drive, Confluence, and Second Brain. The key tools are:

Workflow

Step 1: Detect standup cadence

Determine when the last standup happened to set the lookback window.

Server: user-google-workspace

calendar_get_events(
  query = "standup",
  time_min = "<3 days ago>",
  time_max = "<now>",
  max_results = 10
)

Search for the recurring "standup - future sustainability & E2E" event. Find the most recent past occurrence = LAST_STANDUP.

Store as SINCE_DATE in both ISO 8601 and YYYY-MM-DD format.

Step 2: Fire parallel data queries

Launch all five of these in a single parallel batch, reusing work-context Step 3 patterns with the narrower standup window.

2a. Jira — status changes since last standup

Server: user-atlassian

Reuse work-context Step 3a with modified JQL:

searchJiraIssuesUsingJql(
  cloudId = "<YOUR_ATLASSIAN_CLOUD_ID>",
  jql = "assignee = currentUser() AND status changed DURING ('{SINCE_DATE}', now()) ORDER BY updated DESC",
  fields = ["summary", "status", "issuetype", "updated", "project"],
  maxResults = 25
)

Second query for currently in-progress work:

searchJiraIssuesUsingJql(
  cloudId = "<YOUR_ATLASSIAN_CLOUD_ID>",
  jql = "assignee = currentUser() AND statusCategory = 'In Progress' ORDER BY priority DESC",
  fields = ["summary", "status", "issuetype", "priority", "project"],
  maxResults = 15
)

Third query for blocked tickets:

searchJiraIssuesUsingJql(
  cloudId = "<YOUR_ATLASSIAN_CLOUD_ID>",
  jql = "assignee = currentUser() AND (status = 'Blocked' OR status = 'Impediment') ORDER BY priority DESC",
  fields = ["summary", "status", "project"],
  maxResults = 10
)

2b. GitHub — PRs and commits since last standup

Reuse work-context Step 3b with SINCE_DATE:

gh search prs --author="jaquevan" --created="{SINCE_DATE}..{TODAY}" \
  --json title,url,state,createdAt,repository --limit 25

gh search prs --reviewed-by="jaquevan" --created="{SINCE_DATE}..{TODAY}" \
  --json title,url,state,createdAt,repository --limit 10

gh api "search/commits?q=author:jaquevan+committer-date:{SINCE_DATE}..{TODAY}&per_page=25&sort=committer-date" \
  --jq '.items[] | {message: .commit.message, date: .commit.committer.date, repo: .repository.full_name}'

2c. Calendar — meetings attended + today's schedule

Server: user-google-workspace

Two queries:

Since last standup (for "yesterday" section):

calendar_get_events(
  time_min = "{SINCE_DATE}",
  time_max = "<now>",
  max_results = 25
)

Today forward (for "today" section):

calendar_get_events(
  time_min = "<now>",
  time_max = "<end of today>",
  max_results = 15
)

Apply tomorrow-calendar-accepted filtering to both:

• If attendee entry has "self": true, include only if "responseStatus": "accepted"
• If no attendees array, include if user is the organizer

2d. Slack — work channel activity only

Server: <YOUR_SLACK_MCP_SERVER>

Reuse work-context Step 3d, but scoped to work channels only:

search_messages(
  query = "from:me after:{SINCE_DATE_SLACK}",
  sort = "timestamp",
  limit = 50
)

Work channel allowlist — only include messages from these channels:

• #wg-agent-eval-harness
• #wg-rhai-first-uxd
• #team-uxd-rhai
• #team-openshift-ai-dashboard

Filter out messages from DMs and social/watercooler channels.

Additionally, scan for blocker language in messages from any channel:

search_messages(
  query = "from:me blocked OR waiting on OR can't proceed after:{SINCE_DATE_SLACK}",
  limit = 10
)

2e. Agent transcripts — undocumented work

Search recent Cursor agent conversations for work done since the last standup that may not appear in Jira, GitHub, or Slack.

Transcript location: ~/.cursor/projects/<YOUR_WORKSPACE_PROJECT_ID>/agent-transcripts/

Each chat is a directory containing a <uuid>.jsonl file (one JSON object per line with role and message fields).

Search strategy:

1. List transcript directories modified since SINCE_DATE:

   find ~/.cursor/projects/<YOUR_WORKSPACE_PROJECT_ID>/agent-transcripts/ \
     -maxdepth 1 -type d -newermt "{SINCE_DATE}" | sort
   

2. For each matching directory, read the first 3 lines of the JSONL file to extract the user's initial query and the assistant's summary. This reveals the topic without reading the full conversation.

What to extract:

• Skills built or modified
• Files created or changed
• Investigations or debugging work
• Anything that represents progress not captured by other sources

Cap: Read at most 5 transcripts. Only include entries that represent substantive work (skip short Q&A or exploratory chats).

Step 3: Categorize results

Sort all gathered data into three buckets:

Yesterday / Since last standup:

• Jira tickets with status transitions (e.g., "Transitioned -2658 to In Progress")
• GitHub PRs opened, merged, or reviewed
• Substantive Slack messages in work channels (shared artifacts, decisions, announcements)
• Meetings attended (accepted events only)
• Agent transcript work (skills built, investigations, undocumented progress)

Today:

• Jira tickets currently in progress (the "continue" items)
• Today's upcoming accepted meetings with times
• Any planned work mentioned in recent Slack messages

Blockers:

• Jira tickets with "Blocked" or "Impediment" status
• Slack messages containing blocker language
• If nothing found, output "None"

Step 4: Filter for substance

Remove noise from each section:

• Drop trivial Slack messages — emoji-only, single-word replies, social chat. Only keep messages that mention work artifacts (ticket keys, PR links, tools, project names) or convey decisions/updates.
• Drop personal calendar events — lunch blocks, personal reminders. Keep meetings with other attendees only.
• Deduplicate — if a PR closes a Jira ticket, mention once (prefer the Jira framing since that's what the standup audience tracks).

Step 5: Format output

Generate the standup update as concise paste-ready text. Each bullet should be one line, action-oriented, and specific.

Format:

**Yesterday / Since last standup:**
- Transitioned <YOUR_JIRA_PROJECT>-2658 to In Progress
- Opened PR #123 on agent-eval-harness (download toolbar)
- Attended RH AI-first UX meeting
- Shared eval report updates in #wg-rhai-first-uxd

**Today:**
- Continue: <YOUR_JIRA_PROJECT>-2658 prototype evaluation framework
- 1:00 PM standup - future sustainability & E2E
- 1:30 PM Evan / <YOUR_MANAGER> 1:1

**Blockers:**
- None

Writing rules:

• Start each bullet with a verb (Transitioned, Opened, Attended, Shared, Continue)
• Include ticket keys, PR numbers, and channel names for specificity
• Keep bullets to one line each — no sub-bullets, no paragraphs
• Include meeting times in the "Today" section
• "Continue: " prefix for in-progress items carrying over
• If no items exist for a section, show "- (no activity)" for yesterday and "- None" for blockers

Step 6: Present in chat

Output the formatted standup text directly in chat. Do not create a Canvas — standups are paste-and-go.

After the standup text, add a small note:

Lookback window: {SINCE_DATE} to now. Sources: {included list}. {skipped sources, if any}.

Fallback Behavior

Always report which sources were unavailable.

Common Mistakes