brag-sheet

Brag Sheet — Work Impact Writer

Turn engineering work into evidence-backed impact statements for performance reviews, self-reviews, promotion packets, and weekly updates. Uniquely mines Copilot CLI session logs, git history, and PRs to reconstruct forgotten work.

USE FOR: "brag", "log work", "what did I do", "backfill", "performance review", "self-review", "promo packet", "weekly update", "status report", "write impact statement", "what did I ship", "I forgot to log my work", "review prep", "accomplishments" DO NOT USE FOR: project management, sprint planning, time tracking, ticket creation

Quick Start

User wants...	Mode	Output
Log one accomplishment	Capture	1 impact-first entry
"What did I do last week?"	Backfill	Entries grouped by week, mined from git/PRs/sessions
Prep for review or promo	Review Pack	Entries grouped by impact theme + STAR narratives

Agent Behavior Rules

1. DO confirm the time range and scope before scanning sources. Don't assume "last week" — ask.
2. DO check which tools are available (save_to_brag_sheet, git, gh) before choosing a workflow.
3. DO always include all three parts: action → result → evidence. If evidence is missing, write (evidence needed) — never silently omit.
4. DO show drafted entries to the user before saving. Never auto-save without confirmation.
5. DO group related commits into a single entry. Ten commits on the same feature = one entry.
6. DO preserve the user's voice. Reframe for impact, but don't invent accomplishments or inflate scope.
7. DO NOT fabricate metrics, team sizes, or impact numbers. If the user doesn't provide a number, don't invent one.
8. DO NOT write entries for work the user only described verbally without verifying. Ask: "Did this ship? Is there a PR or doc I can reference?"
9. DO NOT skip the backfill scan steps or draft entries before scanning is complete.
10. DO NOT pad weak periods with trivial entries. An honest gap is better than inflated fluff.

Entry Format

Every entry uses impact-first framing with three required parts:

Did [action] → [result/impact] → [evidence]

Do not output an entry unless it includes all three parts. If evidence is missing, ask for it or mark as "(evidence needed)".

Anti-Patterns

❌ Don't	✅ Do instead
"Fixed a bug in auth"	"Fixed token refresh race condition → eliminated 401s affecting 12% of API calls → PR #247"
"Worked on dashboards"	"Built latency dashboard in Grafana → on-call detects P95 spikes in <2min → deployed to prod"
Invent a metric: "saved 40% of eng time"	Ask: "Do you have a rough estimate, or should I keep this qualitative?"
One entry per commit	Group related commits into one entry with highest-impact framing
Passive voice: "The pipeline was improved"	Active voice: "Built CI matrix → caught Windows-only bug before release"
List technologies used	State the outcome: "Migrated 4 services to IaC → deploy time 45min → 8min"
Silently drop weak entries	Mark (evidence needed) and present for user to fill in

Evidence Ladder

Not every entry needs a metric. Use the strongest evidence available:

Strength	Evidence type	Example
? Best	Quantified metric	"Reduced P95 latency from 800ms to 120ms"
? Strong	PR, commit, or doc link	"PR #312, design doc in wiki"
? Good	Observable outcome	"Unblocked Team X", "Resolved Sev2 incident Y"
✅ Acceptable	Qualitative + context	"Reduced toil for on-call rotation — see updated runbook"
⚠️ Weak	Activity only	"Worked on auth" — reframe or mark (evidence needed)

Never invent a metric to fill the gap. Qualitative evidence with context beats fabricated numbers.

Categories

ID	Emoji	Use for
pr	?	Merged PRs, shipped features
bugfix	?	Bug fixes, incident patches
infrastructure	?️	Infra, deployments, migrations
investigation	?	Root cause analysis, debugging
collaboration	?	Reviews, mentoring, design discussions
tooling	?	Dev tools, scripts, automation
oncall	?	Incident response, on-call wins
design	?	Design docs, architecture decisions
documentation	?	Docs, runbooks, guides

How to Help the User

Follow this decision tree:

1. If save_to_brag_sheet tool is available → use extension tools directly (save_to_brag_sheet, review_brag_sheet, generate_work_log). Do not reference or attempt to call these tools unless they are confirmed available.

2. If git or gh CLI is available → backfill from commits and PRs (see Backfill section below)

3. Otherwise → guided interview: "What did you work on?", "Who benefited?", "What's the evidence?"

For each entry, walk through: What (the deliverable) → Why (who benefits) → Evidence (PR, metric, link). Output formatted markdown the user can paste into a review doc.

Backfill Workflow

When the user asks "what did I do last week" or "backfill my history":

Follow these steps in order. Do not draft entries until scanning is complete.

Step 1: Scan available sources

Check what's available, then mine each source:

git --version 2>/dev/null         # for commit mining
gh --version 2>/dev/null          # for PR mining
ls ~/.copilot/session-state/ 2>/dev/null  # Copilot session logs

Git commits — recent commits by the user in the current repo:

git log --author="$(git config user.email)" --since="2 weeks ago" \
  --pretty=format:'%h|%ad|%s' --date=short --no-merges

PR history — merged PRs across repos:

gh pr list --author @me --state merged --limit 20 \
  --json number,title,repository,mergedAt

Copilot session history (unique to this skill):

• Path: ~/.copilot/session-state/<session-id>/workspace.yaml
• Read fields: summary, cwd, repository, branch
• Skip sessions without a summary field
• Note: this directory may not exist on all machines

If none of these sources are available, fall back to the guided interview.

Step 2: Group related work

Cluster related signals into one entry:

• Same PR + its commits → 1 entry
• Multiple commits on the same file/feature within 3 days → 1 entry
• Copilot sessions referencing the same repo + branch → merge into PR entry if one exists

Step 3: Draft entries

Write impact-first entries for each group. Assign categories.

Step 4: Present and refine

Show all drafted entries to the user. Adjust based on feedback.

Step 5: Output

Format as markdown grouped by week:

## Week of 2025-04-14

### ? PRs & Features
- **Migrated auth service to managed identity** → eliminated 3 secret rotation incidents/quarter → PR #312

### ?️ Infrastructure
- **Built CI pipeline for copilot-brag-sheet** → 107 tests across 3 OSes × 3 Node versions → shipped v1.0.0

Performance Review Prep

When the user is preparing for a performance review (Connect, annual review, etc.):

Structure

1. Gather — collect entries from the work log (or backfill using the workflow above)
2. Select — pick the top 3–5 highest-impact items
3. Rewrite each item with three parts:
   • What I did — the specific action
   • Why it mattered — who benefited, what changed
   • Proof — PR number, metric delta, dashboard link, customer outcome
4. Organize by impact theme (not chronologically):
   • Delivering results / operational excellence
   • Customer / team impact
   • Collaboration / mentoring / leadership
   • Growth / learning
5. Ask for gaps — if evidence is missing, prompt the user: "What metric changed?", "Who was unblocked?", "What's the PR or incident ID?"

Strong vs weak entries

✅ Strong	❌ Weak
Outcome-first, quantified	Activity list ("worked on X")
Tied to customer/team impact	No beneficiary mentioned
Includes evidence (PR, metric)	No measurable result
Shows ownership or leadership	Pure task completion

Narrative format

For longer narrative sections, use STAR: Situation → Task → Action → Result.

For Microsoft employees using the Connect preset, frame entries around Core Priorities: delivering results, customer obsession, teamwork, and growth mindset.

Output Contract

Before finishing, ensure:

1. Every entry has action → result → evidence (mark (evidence needed) if missing)
2. No fabricated metrics — only user-provided or source-verified data
3. Entries shown to user before saving
4. Time range explicitly stated
5. Output is pasteable markdown with categories assigned

Gotchas

No recent commits in the current repo

The user may work across multiple repos. Before concluding there's nothing to backfill:

1. Ask if they want to scan a different repo or branch
2. Check gh pr list --author @me --state merged for cross-repo PRs
3. Fall back to the guided interview — not all impactful work leaves git traces (design docs, incident response, mentoring)

Review period doesn't match git history

Performance reviews often cover 6–12 months. Explicitly set the date range:

git log --author="$(git config user.name)" --since="2024-07-01" --until="2025-01-01" --oneline

PR history (gh pr list --state merged) is more reliable for long time ranges than commit logs.

User can't quantify impact

Not every entry needs a number. See the Evidence Ladder above. Acceptable evidence includes PR links, "unblocked Team X", or qualitative outcomes with context. Never invent a metric to fill the gap.

Copilot session directory doesn't exist

~/.copilot/session-state/ only exists if the user has run Copilot CLI sessions. Don't error — silently skip and note: "No Copilot session history found; scanning git and PRs only."

"brag" might mean something else

The user might say "brag about this feature to my team" (a launch announcement, not a work entry). Confirm intent if ambiguous.

Pair programming or co-authored commits

If multiple authors appear on the same commits, ask: "Should I credit this as your work, shared work, or skip it?"

Automatic Session Tracking (Optional)

For automatic background tracking of every Copilot CLI session (files edited, PRs created, git actions), install the copilot-brag-sheet extension. It adds save_to_brag_sheet, review_brag_sheet, and generate_work_log tools to every session.