borghei/Claude-Skills / personal-productivity/reflect
reflect
VerifiedTurn reflection into decisions by scoring predictions against outcomes and tracking whether commitments held. Use when running a quarterly review, scoring forecast calibration, or reflection keeps producing notes instead of change.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 15.5 KB |
| assets/prediction-log-template.md | markdown | 3.3 KB |
| assets/quarterly-reflection-template.md | markdown | 3.6 KB |
| assets/sample_commitments.json | config | 1.3 KB |
| assets/sample_commitments_quarterly.json | config | 2.5 KB |
| assets/sample_predictions.json | config | 4.7 KB |
| references/calibration-and-forecasting.md | markdown | 18.7 KB |
| references/reflection-cadences.md | markdown | 17.5 KB |
| scripts/calibration_core.py | script | 7.4 KB |
| scripts/calibration_scorer.py | script | 9.0 KB |
| scripts/reflection_prompt_generator.py | script | 11.0 KB |
references/reflection-cadences.md
View on GitHub# Reflection Cadences
What to ask at each timescale, and how to make reflection end in changed
behaviour rather than a paragraph of feelings. Different cadences answer
genuinely different questions; running one cadence's prompts at another's
interval is why most reflection practices quietly stop.
---
## 1. The boundary: reflection vs the weekly operating review
These are two different practices and conflating them damages both.
| | Weekly operating review | Reflection |
|---|---|---|
| **Question** | What happened, and what do I do next week? | Was my judgement any good, and what should I change? |
| **Horizon** | The coming week | Across weeks, quarters, years |
| **Output** | Priorities and a cleared inbox | A behaviour change or a scored belief |
| **Failure mode** | Falling behind | Self-flattery |
| **Cadence** | Weekly, fixed | Daily / weekly / quarterly, layered |
The weekly operating review is a **throughput** practice: it keeps the system
current. Reflection is a **learning** practice: it tests whether the judgement
driving the system is any good.
If your weekly review already covers wins, blockers, and next-week priorities,
do not duplicate that here. The reflection layer adds one thing on top: testing
last cycle's commitments and predictions against what actually happened, and
extracting a change. Run them back to back — operating review first, reflection
second, using the operating review's output as reflection's raw material.
---
## 2. Daily — 5 minutes
**Purpose:** catch today's evidence before memory rewrites it.
Memory degrades fastest in the first 24 hours, and it degrades in a specific
direction: toward a coherent story in which you were less surprised than you
were. Five minutes at the end of the day preserves the raw material.
**Three prompts, no more:**
1. What did I predict about today that turned out wrong?
2. Where did my time actually go, versus where I said it would go?
3. What is the one thing I would do differently if today repeated?
**The rule that keeps it alive:** name one concrete change for tomorrow. If
nothing changes, the entry was journaling — pleasant, occasionally clarifying,
but not a learning practice.
**[RECOMMENDED] Daily reflection is optional.** It is the first cadence to drop
when time is short and the least costly to lose, because weekly reflection
catches most of the same signal at a lower total cost. Do not let a broken daily
streak become the reason you abandon the weekly and quarterly layers, which are
where the value actually lives.
---
## 3. Weekly — 20 minutes
**Purpose:** test last week's commitments against outcomes, and extract one
pattern worth acting on.
This is the load-bearing cadence. If you run only one, run this one.
### Prompts
1. Which commitments did I keep, and which did I quietly drop? Why those
specifically?
2. What did I believe on Monday that I no longer believe?
3. Which piece of work took materially longer than I estimated, and what did I
fail to account for?
4. What did I avoid this week, and what does the avoidance tell me?
5. What surprised me — and had I predicted it, would I have acted differently?
Prompt 1 is the one that does the work. **Which commitments you drop is a more
honest signal of your priorities than any stated list**, because dropping
happens under pressure and without deliberation. A pattern of dropped
commitments in one area usually means either the area does not matter as much as
you claim, or something structural is blocking it.
### Commitment accountability
Every unresolved commitment needs a decision, not a note:
| Status | The decision required |
|---|---|
| Missed | Was the estimate wrong, the priority wrong, or the commitment never real? The fix differs for each |
| Partial | What stopped it, and is the remainder still worth finishing? |
| Overdue | Re-commit with a date you believe, or drop it |
| **Carried 3+ cycles** | Carrying it again is a decision to never do it. Make it explicit: date, delegate, or kill |
The **three-cycle rule** is the single most valuable mechanic in weekly
reflection. An item that has survived three reviews without progress is not
waiting for time; it is waiting for a decision you keep declining to make. The
cost of carrying it is not the task — it is the small tax on attention paid
every week, plus the erosion of the list's credibility.
### Keep rate
| Keep rate | Reading |
|---|---|
| > 85% | Suspiciously high. You are probably under-committing — the commitments are safe rather than the execution strong |
| 60-85% | Healthy. Ambitious enough to sometimes fail, and mostly holds |
| < 60% | Committing to more than you deliver. **Cut the number of commitments before trying to improve execution** |
The instinct at a low keep rate is to try harder. That reliably fails, because
the cause is almost always volume rather than effort. Halve the commitments and
the rate usually recovers on its own.
### Closing requirement
**State one behaviour change, phrased as a rule rather than an intention.**
| Intention (fails) | Rule (works) |
|---|---|
| "Be better about deep work" | "No meetings before 11:00 on Tuesdays and Thursdays" |
| "Estimate more carefully" | "Any estimate over 3 days gets a written list of what could go wrong" |
| "Say no more often" | "Any new recurring meeting requires dropping an existing one" |
Rules are testable next week. Intentions are not, which is why they survive
indefinitely without ever being kept.
---
## 4. Quarterly — 90 minutes
**Purpose:** score judgement across a meaningful span and change something
structural.
A quarter is the shortest span over which slow effects become visible: whether a
role fits, whether a habit stuck, whether a bet paid off. It is also the
shortest span with enough resolved predictions to score calibration.
### Sequence
| Minutes | Step |
|---|---|
| 0-20 | Score the prediction log — `calibration_scorer.py` |
| 20-35 | Read the quarter's weekly reflections in one sitting, listing recurring themes |
| 35-55 | Work the quarterly prompts |
| 55-75 | Decide: one commitment killed, one method changed |
| 75-90 | Write next quarter's predictions with explicit confidence levels |
Reading the weekly reflections in a batch is what makes them worth writing.
Individually they are unremarkable; twelve in a row show patterns invisible at
weekly resolution — the same blocker recurring, energy tracking a specific kind
of work, a commitment re-made and re-dropped five times.
### Prompts
1. Which of my quarter-start predictions were wrong, and were they wrong in a
consistent direction?
2. What did I commit to that I never started? What does the pattern across those
say about what I actually value?
3. Which recurring commitment has produced no return and should be killed?
4. Where was I overconfident, and what was the cost in decisions made on that
confidence?
5. What did I learn that changed how I work, versus what I merely read about?
6. If I were advising someone else with this quarter's record, what would I tell
them to stop doing?
Prompt 6 does disproportionate work. The advice you would give someone else is
consistently clearer and more decisive than the advice you give yourself,
because it is free of the sunk costs and self-image concerns attached to your
own record.
### The kill list
**Every quarterly reflection must kill at least one thing** — a recurring
meeting, a project, a metric nobody acts on, a habit that has not paid off.
This is not arbitrary. Commitments accumulate monotonically unless something
removes them: each was added for a reason, and no individual one feels worth
removing. Without a forced removal, four quarters of additions produce a
calendar and a commitment list with no room for anything new. Forcing exactly
one removal per quarter is a low-cost mechanism that prevents this without
requiring a crisis.
---
## 5. Making reflection produce decisions
Six mechanisms, ranked by how much they change behaviour.
| Mechanism | Effect |
|---|---|
| **Written predictions with confidence** | The strongest. Converts opinion into a scoreable record memory cannot edit |
| **Commitment carry-count** | Makes chronic avoidance visible as a number rather than a feeling |
| **Rules, not intentions** | Testable next cycle; an intention can survive years unkept |
| **Forced kill per quarter** | Prevents monotonic accumulation of commitments |
| **Batch reading of past reflections** | Surfaces patterns invisible at single-entry resolution |
| **The advice-to-another framing** | Bypasses sunk cost and self-image |
### What makes reflection fail
| Failure | Why it happens | Fix |
|---|---|---|
| It becomes journaling | Writing feels productive; deciding is uncomfortable | Require a named change to close every session |
| Prompts are too broad | "How did the week go?" has no wrong answer | Use specific, falsifiable prompts |
| No record of prior beliefs | Nothing to check against | Log predictions with confidence numbers |
| Too long, so it gets skipped | An hour weekly is unsustainable | 5 min daily, 20 weekly, 90 quarterly |
| Only run when things go badly | Reflection becomes associated with failure | Fixed cadence regardless of how the period went |
| Self-flattery | Unaided memory edits toward a coherent, flattering story | Score against written records, not recollection |
| No structural change ever results | Every conclusion is "try harder" | Force a kill and a method change quarterly |
---
## 6. Starting from nothing
Do not start all three cadences at once. The full stack is roughly 3 hours per
quarter plus 20 minutes a week, which is sustainable — but only after the weekly
layer is established.
1. **Weeks 1-4:** weekly reflection only, 20 minutes, fixed slot after your
operating review. Nothing else.
2. **Week 2 onward:** start logging predictions with confidence numbers as
decisions arise. Aim for 15-30 in the first quarter. Resolve them weekly.
3. **End of quarter 1:** run the first quarterly reflection and score the
predictions. Expect the score to be humbling — overconfidence on your own
delivery dates is close to universal, and finding it is the point.
4. **Quarter 2:** add daily reflection only if the weekly layer has held for a
full quarter. It is the optional layer.
The most common failure is starting with daily reflection because it seems
smallest, missing three days, and abandoning the whole practice. Start with
weekly — it carries most of the value and survives a missed week without
collapsing.
---
## 7. The annual layer
Optional, and only worth running after four quarterly reflections exist to read.
Its value comes entirely from having a year of scored records; without them it
degenerates into the kind of aspirational goal-setting that produces nothing.
**Time:** 3 hours. **Inputs:** four quarterly reflections, the full prediction
log, the year's kill list.
### Questions a year answers that a quarter cannot
| Question | Why it needs a year |
|---|---|
| Did my calibration actually improve? | 60-100 predictions — the first sample large enough to distinguish improvement from noise |
| Which commitments recurred every quarter without ever completing? | Chronic avoidance is invisible at quarterly resolution; across four quarters it is undeniable |
| Did the things I killed stay dead? | Killed commitments frequently reappear in a new form within two quarters |
| Is the work still the work I want? | Role fit changes slowly and is unobservable at shorter horizons |
| What did I get right that I dismissed at the time? | Correct contrarian calls are only visible in hindsight, and they are the best evidence about where to trust yourself |
The last question is the one people skip and the most valuable. Reflection
skews toward cataloguing errors, which produces a systematically pessimistic
self-assessment. **Where you were right against consensus is direct evidence of
where your judgement outperforms** — and that tells you which decisions to weight
your own view on, which is far more actionable than another list of mistakes.
### Annual output
Exactly three things:
1. **One thing to stop** — larger than a quarterly kill; a project, a
responsibility, a standing obligation.
2. **One thing to start** — with a quarterly milestone, or it will not happen.
3. **One belief to test** — written as a prediction with a date and a confidence
number, so next year's reflection can score it.
---
## 8. Two worked reflections
### Weekly (20 minutes)
Commitments: 3 kept, 2 missed, 1 partial, 4 open. Keep rate 50%.
**Prompt 1 — which did I drop, and why those?** Both misses were things
requiring a difficult conversation: declining a standing meeting, and writing a
performance review. Neither was time-constrained; both were avoidance.
**Prompt 3 — what took longer than estimated?** The headcount plan, estimated at
two hours, took five. The gap was entirely in gathering inputs from other
people, which was not in the estimate at all.
**Prompt 4 — what did I avoid?** The performance review, for the third week.
**Accountability:** the review has now carried three cycles. Under the
three-cycle rule this is no longer a task, it is a decision being declined.
Decision taken: scheduled for Tuesday 09:00 in a protected block.
**Closing rule:** *"Any commitment that requires a difficult conversation gets
scheduled into a specific block when I make it, not added to a list."*
Note the shape of the result. The pattern was not "I am busy" — it was that
avoidance clusters around interpersonal difficulty, and the rule addresses that
specific mechanism rather than effort in general.
### Quarterly (90 minutes)
**Calibration:** Brier 0.294, skill score -0.19, overconfident by 16%, worst
domain delivery (+32%).
**Weekly themes across twelve reviews:** the same two projects appear as
"blocked on someone else" in nine of them. Deep-work volume tracks inversely
with weeks containing three or more days of travel.
**Prompt 2 — committed but never started?** A customer advisory board,
re-committed five quarters running. The pattern says clearly that it is
something believed to be a good idea rather than something wanted.
**Prompt 6 — advice to someone else with this record?** "Stop predicting your
own delivery dates above 80%, and either start the advisory board this month or
admit you are never going to."
**Decisions:**
- *Killed:* the customer advisory board. Five quarters is answer enough.
- *Method changed:* delivery-date confidence capped at 75%, and every estimate
over three days gets an outside-view check against the last comparable items.
- *Chronic commitments:* two given hard dates, one delegated, one killed.
**Next quarter:** 22 predictions logged, including six delivery dates at capped
confidence.
---
## 9. Fitting reflection into an existing week
The practical question is where these sit relative to work you already do.
| Practice | When | Relationship |
|---|---|---|
| Weekly operating review | Friday afternoon or Sunday evening | Runs **first**; produces the raw material |
| Weekly reflection | Immediately after, same sitting | Uses the operating review's output; adds the learning layer |
| Prediction resolution | Inside the weekly reflection, 5 min | Mark what resolved; add new predictions from the week's decisions |
| Daily reflection | End of the working day | Optional layer |
| Quarterly reflection | Last Friday of the quarter, blocked in advance | Needs a 90-minute protected block, booked a month ahead |
**[RECOMMENDED] Run weekly reflection in the same sitting as the operating
review, not on a different day.** A separate session requires a separate
trigger, a separate defended slot, and separate motivation — three failure
points. Appended to an existing habit, it inherits all three.
### Total cost
| Cadence | Per instance | Annual |
|---|---|---|
| Daily (optional) | 5 min | ~20 h |
| Weekly | 20 min | ~17 h |
| Quarterly | 90 min | 6 h |
| Annual (optional) | 3 h | 3 h |
| **Core (weekly + quarterly)** | — | **~23 h/year** |
Twenty-three hours a year is about half a percent of working time. The question
is not whether that is affordable — it is whether it changes anything, which is
precisely why the closing requirement of every session is a named change. A
practice that produces no changes should be stopped, and the log makes that
judgement possible rather than a matter of feeling.
---
## 10. When to stop or change the practice
Reflection is not exempt from its own quarterly kill list.
| Signal | Diagnosis | Action |
|---|---|---|
| Three consecutive sessions with no named change | Prompts have gone stale | Replace the prompt set; keep the cadence |
| Sessions consistently run over time | Cadence mismatched to the questions | Move quarterly questions out of the weekly session |
| Dread before the session | It has become an accounting of failures | Add "what did I get right that I dismissed?" |
| Calibration flat across a year | Not applying corrections, or applying several at once | One correction, held for a full quarter |
| Reflection duplicates the operating review | The boundary has collapsed | Re-read §1; reflection tests judgement, the operating review tracks throughput |
| Keep rate above 90% for three quarters | Commitments too safe to be informative | Commit to something that might fail |
**The one signal that means stop:** if a year of reflection has produced no
behaviour change you can name, the practice is not working for you and the
honest response is to stop rather than to continue performing it. That
conclusion is itself a legitimate output — and it is only available because the
sessions were recorded in a form that can be checked, which is the argument for
the whole apparatus.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| receiving-code-reviewobra/superpowersUse when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementation | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirements | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| using-git-worktreesobra/superpowersUse when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallback | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| verification-before-completionobra/superpowersUse when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |