Career retro

Career Retro

This is the payoff moment for a year of logging. Every other skill in the reflection system deposits into the memory — stakeholder reflections, wins, self-reflections, pulses. This skill withdraws the accumulated interest: it reads all of it at once and answers the question no single entry can: who have I become, and is it who I meant to become?

Its most important property is that it writes its own output and reads the previous one. The first retro is a snapshot. The second is a story — because it can say what changed. The value grows with every cycle, which is exactly the moat plain Claude can't replicate.

Distinguish the ask first

• "I need to hand in my self-eval / build my promo packet" → that's an audience-facing artifact; hand off to wins-curate (perf-review / promo-packet mode) or report-promo-case. This skill can feed them, but it is not the submission document.
• "Let me reflect on how I handled last week" → a single entry; that's self-reflect.
• "Where am I in my career, honestly, across everything" → this skill.

If the user wants both, run the retro first (it produces the honest internal picture), then offer to spin the relevant slice into the external artifact.

Loading — read broadly, cite specifically

Set the period first (default: since the last retro, or 12 months if none). Then read everything under $BETTERSENSE_WORK_REFLECTIONS_HOME (default ~/bettersense-work-reflections/) that falls in the window:

1. The previous retro — self/retros/retro-*.md, most recent. This is the baseline for every delta. If none exists, say so: this first retro sets the baseline for all future ones.
2. profile.md — the stated role, level, and strategic focus. The retro's sharpest question is often the gap between stated focus and where the evidence shows the time actually went.
3. wins.md — the impact record, already dated and structured.
4. self/reflections.md — the user's own read on themselves over the period; trajectory of mood, energy, fulfillment.
5. All stakeholder files (<category>/*.md via stakeholders.json) — how the user's relationships and management evolved. Growth as a leader lives here more than in wins.
6. pulses/ and team-charters/ — product and team arcs the user drove.
7. ADRs / decision logs — decision-log stores these in the code repo, not here; ask the user to point at docs/decisions/ if they want decisions folded in. Don't fabricate their absence into a weakness.

Citation discipline (inherited from the synthesize skills): every claim about growth or a gap cites specific dated entries. "You got better at hard conversations" is worthless; "in March you logged avoiding the feedback conversation with Alex for three weeks; by September you logged delivering same-week feedback to two reports (wins 2026-09-04, 2026-09-19)" is the product. Never assert a trajectory you can't cite.

The retro structure

Write it to self/retros/retro-YYYY-MM-DD.md. Structure:

# Career Retro — [period, e.g. 2026 H1] — generated YYYY-MM-DD
baseline: [previous retro date, or "first retro"]

## The one-line arc
[If you had to describe this period's growth in a sentence, cited. The thing you'd want a sponsor to know.]

## Deltas since last retro
[The heart of the document. For each dimension, what CHANGED, with citations on both ends.]
| Dimension | Then | Now | Evidence |
|---|---|---|---|
| Scope / impact | ... | ... | wins 2026-... |
| Leadership / people | ... | ... | stakeholder entries ... |
| Technical / craft | ... | ... | ... |
| Judgment | ... | ... | ... |
| Visibility / influence | ... | ... | ... |
[First retro: replace "Then" with the baseline you're establishing now.]

## Evidence highlights
[3–6 strongest cited moments of the period — the ones that would anchor a promo case.]

## Gaps and honest weak spots
[Cited. Where the evidence is thin, where a pattern repeats, where stated focus and actual work diverged.
This is the section that earns the user's trust — no retro is all growth.]

## Trajectory read
[Is the arc pointing where the profile says it should? Toward the next level, or plateauing? Name it.]

## Focus for next period
[2–4 concrete, evidence-driven priorities. These become next retro's things-to-measure-against.]

Operating principles

• Longitudinal, not point-in-time. The question is always "compared to when," and the answer must be cited on both sides. A retro that reads like a single self-reflection has failed.
• Honest gaps are the credibility. A retro that's all upward-and-to-the-right reads as marketing. Surface the plateau, the repeated pattern, the divergence between what the profile claims and what the log shows.
• Never invent evidence. Same rule as stakeholder-synthesize and pulse-synthesize — if the memory is thin in an area, say the memory is thin (and note it as a logging gap to close before the next retro), don't manufacture a narrative.
• Persist, always. Writing the dated file is not optional bookkeeping — it is the feature. The delta only exists because last time's file exists.

Composition

• Reads the output of: wins-log, stakeholder-reflect, self-reflect, product-pulse, team-charter
• Feeds into: wins-curate (turn the retro's evidence highlights into a promo packet / self-eval), report-promo-case (if the user manages people, the same lens on a report)
• Pairs with: weekly (the weekly ritual keeps the memory fed; the retro is what that feeding was for)
• If a gap surfaces that's really a direction question, hand to the-career-coach