Wins Log

Capture a single "win at work" in a structured form so it can be remixed later. The output is one entry, prepended (newest first) to ~/bettersense-work-reflections/wins.md.

This is the front-of-funnel skill for the wins bundle (wins-log, wins-due, wins-curate). It works in two modes: fresh (the user just did something and wants to capture it) and retroactive (the user is logging something that happened recently — last week, last month).

Data location

~/bettersense-work-reflections/wins.md (configurable via $BETTERSENSE_WORK_REFLECTIONS_HOME). Same root as the stakeholder reflection skills.

First-run setup

If ~/bettersense-work-reflections/wins.md doesn't exist:

1. If ~/bettersense-work-reflections/ itself doesn't exist, hand off to stakeholder-register first — that skill owns the root directory setup and the privacy warning. Resume here once it's done.
2. Create wins.md with this frontmatter:

   ---
   subject: wins
   since: <today's date>
   ---

   # Wins at Work

   <!-- Newest entries are prepended below. Each entry begins with `## YYYY-MM-DD — <headline>`. -->

Per-entry structure

Every entry uses this exact shape. The structure is what makes the entries remixable later — fight to fill every section.

## 2026-05-01 — Shipped contract-summarization to GA

**Type:** delivery, judgment
**Scope:** Q2 multi-team launch
**Collaborators:** Priya (eng lead), Trae (design), legal review team
**Solo vs. shared credit:** Led PM side end-to-end; eng lead and design owned execution; legal team validated the safety story

What happened:
The legal team was spending ~4 hours per deal manually summarizing contracts. We
had committed to GA by end of Q2 against eng leadership skepticism that we'd hit
the eval bar in time.

What I did:
- Pushed back on the original scope (8 contract types) and got us to a focused 3
  in week 2 — that bought 6 weeks of timeline.
- Designed the eval rubric with the legal team directly rather than guessing,
  which surfaced the "non-standard termination clause" failure mode we would
  have shipped past.
- Made the call to delay GA by 1 week when the eval dropped under 0.85, against
  pressure to ship on the original date.

Impact:
- Time-to-summary: 4 hr → 90 sec for the legal team (verified by their team lead)
- 12 paying customers using the feature within 2 weeks of GA
- 0 P0 incidents in first 30 days
- Eval score on launch: 0.91 (held steady at 0.89 in production over month 1)

Evidence:
- Q2 OKR doc, "Contract Summarization GA" section
- Legal team time-study email (April 28)
- #legal-summarization Slack channel
- Postmortem ADR-007 (the eval-driven delay decision)

The capture flow

When invoked, walk the user through these — but quickly. The whole session should take 5–10 minutes for a fresh win, longer for a complex retroactive one.

1. Headline + date

One short title, present tense or past tense, no buzzwords. "Shipped contract-summarization to GA" — not "Successfully delivered industry-leading AI capability."

2. Type

One or more from this list, comma-separated:

• delivery — shipped something
• judgment — made a hard call under uncertainty (especially a no, a delay, a scope cut)
• leadership — set direction, owned outcome, brought people along
• mentorship — grew someone else
• recovery — saved something that was going sideways
• craft — technical / PM excellence (a sharp spec, a clean prototype, a great eval rubric)
• learning — picked up a new capability
• range — worked outside your normal lane
• culture — changed how the team works (a ritual, a norm, a vocabulary)

The Type field is the single most important structural choice — it drives what wins-curate can do later. Don't let the user list everything; push for the 1–3 that actually fit.

3. Scope

One of: sprint, quarter, multi-quarter, ongoing. Helps calibrate the win's weight at curation time.

4. Collaborators + credit framing

Two questions, both required:

• "Who else was involved?" — name them with their role.
• "What was specifically your contribution vs. theirs?" — force an honest framing. "Led the PM side; eng owned execution" is calibrated. "I shipped this" when 5 people worked on it is inflation.

5. What happened (situation)

What was at stake, what was hard, what was unusual. 2–4 sentences. The sentence that matters most: what made this a real win and not just routine work?

6. What I did (action)

The user's specific contribution. Push past team verbs to "I" verbs. "We decided" → "I argued for X and got buy-in from Y." Distinguish what would have happened without the user from what did happen with them.

7. Impact

Quantified when at all possible. Push hard here — this is where most logged wins are weakest.

If the user offers vague impact ("improved the system"), probe:

• For whom? Which users / customers / teams?
• By how much? What numbers, even rough ones?
• Over what time? A spike vs. a sustained shift.

If the user says "I can't quantify it":

• Is there an N you can name? "Helped 3 engineers" beats "helped engineers."
• Is there a downstream effect that's measurable? "Mentored Priya" → "Priya shipped her first solo feature 6 weeks after we started pairing."
• Is there a counterfactual? "Would have been a P0 if we'd shipped" — even that's a measurable risk you reduced.

Some wins genuinely resist quantification (e.g. team culture changes). For those, anchor with specific witnessed examples rather than abstractions.

8. Evidence

Where someone could verify this. Doesn't have to be polished — informal works. "#legal-summarization Slack channel, week of April 15" is fine. "Email from Sarah dated April 28" is fine.

If the user can't name any evidence, gently flag: a win without evidence is hard to defend later when curation matters. They don't have to fix it now, but the entry should note Evidence: none captured at the time so it's honest.

Writing the entry

Prepend the new entry to wins.md (newest first), under the frontmatter and intro. Use today's date.

Confirm the entry with the user before writing — show them the structured version and ask if it lands.

Composing with other skills

When the win came from work that touches another skill, surface the cross-link:

• Feedback-driven wins ("I gave Priya direct feedback that landed and she changed the behavior") → mention feedback-frameworks in the entry. Good fodder for the mentorship type.
• Judgment wins with a real decision behind them → suggest invoking decision-log to capture the underlying ADR. Cross-reference both.
• Wins about a stakeholder relationship ("I built trust with the new VP by doing X") → suggest also logging this in the relevant stakeholder file via stakeholder-reflect. The two systems compose.

Don't force these. The capture is the primary product; cross-links are bonuses.

Operating principles

• Push past vague. Specifics compound; vagueness rots. The skill's value is in being slightly more demanding than the user would be alone.
• Calibrate solo vs. shared credit. The brag doc is more credible — and protects the user's reputation when others read the artifact later — when it's calibrated, not inflated.
• Honor "I had a quiet week." If the user genuinely doesn't have a win this session, don't manufacture one. The forced-positivity school of hype-doc maintenance produces fictions.
• The user's voice. Don't sand off candor. The wins file is private; cleaned-up phrasing is the curate skill's job, not log.
• Don't over-edit retroactive entries. If the user is logging something from 3 weeks ago and the details are fuzzy, capture what they remember, mark it (reconstructed from memory), and move on. Some signal beats no signal.

Anti-patterns to flag

• "I shipped X" when 5 people shipped X. Push for the credit framing.
• Adjectives instead of numbers. "Massively improved" is noise; "reduced P95 from 2.4s to 800ms" is a win.
• Wins that are really tasks. Completing a routine ticket is not a win unless something genuinely hard happened. Distinguish achievement from output.
• Resume buzzwords. "Spearheaded a strategic initiative" is signaling, not capture. Force plain language.
• No evidence + no impact. An entry with neither is a story, not a win. Push for at least one.
• Inflated type tags. Tagging a routine ship as leadership + judgment + craft + range to load up on dimensions hurts curation. One or two honest tags beats five aspirational ones.