shousper/github-work-summary

GitHub Work Summary

Overview

Generates a comprehensive work summary from a user's GitHub activity (PRs, issues) for a specified time period and organization, using the gh CLI for data collection and parallel subagents for detail fetching.

When to Use

• User asks "what did I do last week?" or similar
• User needs a weekly/monthly work summary or standup notes
• User wants to review contributions across an organization
• User needs to compile accomplishments for a performance review or report

Not for: Repository-level changelogs, release notes, or team-wide activity reports.

Process

Copy this checklist and track progress:

Work Summary Progress:
- [ ] Step 1: Clarify parameters (period, org)
- [ ] Step 2: Run data collection script
- [ ] Step 3: Launch parallel subagents for detail fetching
- [ ] Step 4: Analyze contributions
- [ ] Step 5: Generate categorized summary

Step 1: Clarify Parameters

Determine two inputs from your human partner's request:

| Parameter | Default | Examples | |-----------|---------|---------| | Period | "last week" | "this week", "last month", "last 14 days", "2026-01-01 to 2026-01-31" | | Organization | Ask user | "ethpandaops", "*" for all orgs |

If either is ambiguous, ask before proceeding.

Step 2: Run Data Collection Script

Run the categorization script:

python3 skills/github-work-summary/scripts/collect_activity.py "<period>" "<org>" "<today_date>"

The script outputs JSON with all matching PRs and issues. Capture this output for Step 3.

If the script fails: Verify gh CLI is installed and authenticated (gh auth status). The user must be logged in.

Step 3: Fetch Details with Parallel Subagents

CRITICAL: Launch ALL detail-fetching agents in a SINGLE message using multiple Task tool calls for maximum speed.

For each PR in the JSON output, create one subagent:

gh pr view <number> --repo <owner/repo> --json body,comments,author,commits

For each issue, create one subagent:

gh issue view <number> --repo <owner/repo> --json body,comments,author

Wait for all agents to complete before proceeding.

Step 4: Analyze Contributions

For each item, determine your human partner's actual contribution:

| Activity Type | Include? | What to Extract | |--------------|----------|----------------| | PR authored by user | Yes | PR description, what was built/fixed/changed | | Issue authored by user | Yes | Issue description, what was reported/documented | | PR with user's commits | Yes | Commit descriptions, code contributions | | PR user only reviewed | No | Skip entirely | | PR/issue user only commented on | No | Skip entirely |

Focus on what your human partner did, not what happened around them.

Step 5: Generate Summary

Organize by theme, not chronologically. Use this structure:

# Work Summary: <start_date> to <end_date>

## Major Initiatives
- [theme name]: Built/Implemented/Designed X (PR #123, #124)

## Feature Development
- Built [feature] for [purpose] (PR #N - merged/open)

## Bug Fixes
- Fixed [issue] that caused [impact] (PR #N - merged)

## Infrastructure / DevOps
- [entry]

## Open Work
- [in-progress items with current status]

Writing style:

• Active voice, first person: "Built X", "Fixed Y", "Refactored Z"
• Include PR/issue numbers as links
• Note status: merged, open, closed
• Group related PRs under a single initiative
• Highlight key technical outcomes

Data Collection Script

The data collection script lives at skills/github-work-summary/scripts/collect_activity.py.

Date Calculation Reference

| Period | Meaning | |--------|---------| | "last week" | Previous completed Monday-Sunday (NOT last 7 days) | | "this week" | Current Monday through today | | "last month" | First through last day of previous month | | "last N days" | N days ago through today | | "YYYY-MM-DD to YYYY-MM-DD" | Explicit date range |

All dates use ISO 8601 format. Use "*" as org to include all organizations.

Common Mistakes

| Mistake | Fix | |---------|-----| | Searching by --created instead of --updated | Always use --updated to capture work on existing PRs | | Including items where user only reviewed/commented | Only include items where user authored or committed code | | Fetching details sequentially | Launch ALL subagents in a single message for parallel execution | | Missing items due to low limit | Always use --limit 1000 | | Confusing "last week" with "last 7 days" | "Last week" = most recent completed Mon-Sun | | Writing passive summaries ("PR was merged") | Write active, first-person: "Built X", "Fixed Y" |