Work Context

Work Context

Gathers the user's recent activity across all connected tools and renders an interactive Canvas organized by theme. Optimized for speed via local-first reads and parallel MCP queries.

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.

Allowed operations:

• Jira: searchJiraIssuesUsingJql, getJiraIssue, getAccessibleAtlassianResources, atlassianUserInfo, lookupJiraAccountId
• Confluence: searchConfluenceUsingCql, getConfluencePage
• Slack: search_messages, get_channel_history, get_thread, list_joined_channels, get_channel_id_by_name, whoami
• Google: calendar_get_events, drive_search, drive_get_content
• Local: filesystem reads only (Second Brain wiki)
• NEVER call tools that create, edit, transition, post, or send anything.

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. All errors and skipped sources are reported in the Canvas output.

The only acceptable user prompts are in:

• Step 1 — confirming the time window
• Step 7 — presenting the Canvas

Prerequisites

• Atlassian cloud ID: <YOUR_ATLASSIAN_CLOUD_ID>
• GitHub username: jaquevan
• User identity: read Me.md in the workspace root for name and email

Workflow

Step 1: Parse scope

Determine the time window from the user's prompt.

Store as START_DATE and END_DATE in ISO 8601 format (YYYY-MM-DDTHH:mm:ssZ). Also store START_JQL in YYYY-MM-DD format for JQL queries.

Step 2: Read Second Brain (local, instant)

Read the local wiki at ~/second-brain/ to find recently updated topics.

1. Read ~/second-brain/wiki/log.md — scan for entries within the time window. Extract topic names, dates, and source descriptions.
2. Read ~/second-brain/wiki/index.md — get the full topic catalog for cross-referencing later.
3. For any topics with log entries in the window, read the first 15 lines of each page to get the title and summary.

This is the fastest source — filesystem reads complete instantly. Start Tier 2 queries immediately after initiating these reads.

Also scan agent transcripts for recent work not captured elsewhere:

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).

1. List transcript directories modified within the time window:

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

2. For each matching directory, read the first 3-5 lines of the JSONL to extract the user's initial query and the assistant's approach.

3. Extract: what was worked on, skills built/modified, investigations, decisions made. Skip short Q&A or exploratory chats.

Cap at 10 transcripts. This captures undocumented work done in Cursor agent conversations that hasn't reached Jira, GitHub, or Slack yet.

Step 3: Fire parallel MCP queries (all at once)

Launch all six of these queries in a single parallel batch. Do not wait for one to finish before starting another.

3a. Jira — assigned and updated tickets

Server: user-atlassian

searchJiraIssuesUsingJql(
  cloudId = "<YOUR_ATLASSIAN_CLOUD_ID>",
  jql = "assignee = currentUser() AND updated >= '{START_JQL}' ORDER BY updated DESC",
  fields = ["summary", "status", "issuetype", "updated", "priority", "project",
            "customfield_10020"],
  maxResults = 50
)

Also search for issues where the user is the reporter:

searchJiraIssuesUsingJql(
  cloudId = "<YOUR_ATLASSIAN_CLOUD_ID>",
  jql = "reporter = currentUser() AND created >= '{START_JQL}' ORDER BY created DESC",
  fields = ["summary", "status", "issuetype", "created", "project"],
  maxResults = 25
)

Categorize results:

• Completed — statusCategory = Done, resolved in the window
• In Progress — statusCategory = In Progress
• Created — issues the user created in the window

3b. GitHub — PRs, issues, commits

Use the gh CLI (shell commands). Fire these in parallel:

gh search prs --author="jaquevan" --created="{START_JQL}..{END_JQL}" \
  --json title,url,state,createdAt,repository --limit 50

gh search prs --reviewed-by="jaquevan" --created="{START_JQL}..{END_JQL}" \
  --json title,url,state,createdAt,repository --limit 25

gh search issues --author="jaquevan" --created="{START_JQL}..{END_JQL}" \
  --json title,url,state,createdAt,repository --limit 25

gh api "search/commits?q=author:jaquevan+committer-date:{START_JQL}..{END_JQL}&per_page=50&sort=committer-date" \
  --jq '.items[] | {message: .commit.message, date: .commit.committer.date, repo: .repository.full_name, url: .html_url}'

3c. Google Calendar — meetings attended

Server: user-google-workspace

calendar_get_events(
  time_min = "{START_DATE}",
  time_max = "{END_DATE}",
  max_results = 50
)

Extract: event titles, dates, attendee counts. Filter out declined events.

3d. Slack — messages sent

Server: <YOUR_SLACK_MCP_SERVER>

search_messages(
  query = "from:me after:{START_JQL}",
  sort = "timestamp",
  limit = 100
)

Group results by channel. Note which channels had the most activity and which threads generated discussion.

3e. Google Drive — recently modified docs

Server: user-google-workspace

drive_search(
  query = "modifiedTime > '{START_DATE}'",
  max_results = 25
)

Extract: document names, types, last modified dates, and web links.

3f. Confluence — pages contributed to

Server: user-atlassian

searchConfluenceUsingCql(
  cloudId = "<YOUR_ATLASSIAN_CLOUD_ID>",
  cql = "contributor = currentUser() AND lastmodified >= now('-7d') AND type = page",
  limit = 25
)

Adjust the now('-Nd') value to match the user's time window.

Step 4: Optional detail follow-up (capped)

After the parallel sweep completes, fetch details for at most 5 high-signal items across all sources. This prevents runaway context gathering.

Candidates for follow-up:

Skip this step entirely if the user asked for "today" (short window, detail is unnecessary).

Step 5: Deduplicate and theme

Cross-reference results to avoid double-counting the same work:

1. Jira key matching — scan PR titles, Slack messages, and Drive doc names for Jira issue keys (e.g. <YOUR_JIRA_PROJECT>-123). Group items that reference the same ticket.
2. URL matching — detect shared links across Slack messages, Drive docs, and Confluence pages.
3. Keyword clustering — group remaining items by project or topic keywords.

Build themes — each theme is a cluster of related activity across sources. A theme has:

• A short descriptive name (e.g. "MaaS onboarding flow redesign")
• The sources that contributed to it (icons/labels)
• The individual items within it

Items that don't cluster into a theme remain as standalone entries under their source section.

Step 6: Build Canvas

Read the Canvas skill and follow its instructions to create a .canvas.tsx file. The canvas location is: /Users/<user>/.cursor/projects/<workspace>/canvases/work-context.canvas.tsx

Embed all gathered data inline in the canvas file.

Canvas sections:

1. Header — "Work Context" title, date range, generated timestamp
2. Stats row — one stat per source with counts:
   • Jira: tickets touched
   • GitHub: PRs opened / reviewed, commits
   • Calendar: meetings attended
   • Slack: messages sent, channels active in
   • Drive: docs modified
   • Confluence: pages contributed to
   • Second Brain: topics updated
   • Agent transcripts: conversations with substantive work Omit any source that returned zero results or was skipped.
3. Themes section — cards for each cross-source theme, showing the theme name, contributing sources, and key items. This is the primary content.
4. Per-source detail — for items that didn't cluster into themes, show them under their source heading. Use a mix of card styles and open sections for visual variety.
5. Coverage footer — small text listing which sources were included and which were skipped (with reason).

Canvas rules:

• Import only from cursor/canvas. No npm packages, no fetch calls.
• Use useHostTheme() tokens for all colors. No hardcoded hex.
• No emojis, no gradients, no box-shadows.
• Default-export the top-level component.
• Omit sections with no data — never render empty states.

Step 7: Summarize in chat

Alongside the canvas, provide a brief text summary:

• Top 3–5 themes with one-line descriptions
• Key stats (total tickets, PRs, meetings, etc.)
• A link to the canvas for the full breakdown
• Which sources were skipped, if any

Fallback Behavior

Always report which sources were unavailable:

Sources not available for this summary: {list}. The canvas reflects only the sources that were reachable.

Common Mistakes