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 |
assets/prediction-log-template.md
View on GitHub# Prediction Log
Append-only. **Never revise a prediction after the outcome is known** — the log's
entire value is that it records what you believed before you knew, which is the
one thing memory cannot be trusted to preserve.
---
## Writing a scoreable prediction
A prediction is scoreable when a disinterested third party could look at the
world on the resolution date and declare it true or false without argument.
**Claim + resolution date + confidence number.**
| Not scoreable | Scoreable |
|---|---|
| "The migration will go well" | "The migration completes with no rollback and under 30 min downtime, by June 30" — 70% |
| "We'll hire someone soon" | "We sign a staff engineer offer by August 31" — 55% |
| "Churn should improve" | "Monthly churn is below 3.0% in July" — 60% |
**Conventions:** 50-95% only (below 50%, restate as the negation). Round to 5%.
Avoid 100% and 0% — nothing about the future deserves them. 50% is a legitimate
answer, and a log with no 50% entries usually means you are only recording
claims you already feel safe about.
**Volume:** 15-30 per quarter. Below 20 resolved, the calibration estimate is
too noisy to act on; above 40, logging becomes a chore and the habit dies.
---
## Entries
```json
{
"predictions": [
{
"id": "P1",
"statement": "The migration ships before end of Q2",
"domain": "delivery",
"confidence": 0.70,
"made_on": "2026-07-21",
"resolve_by": "2026-09-30",
"outcome": null
}
]
}
```
| Field | Required | Notes |
|---|---|---|
| `statement` | Yes | Must be resolvable by a third party |
| `confidence` | Yes | 0-1 or 0-100; both accepted |
| `outcome` | Yes at resolution | `true` / `false`; `null` or `"unresolved"` while pending |
| `domain` | Recommended | Drives the per-domain breakdown — where the actionable signal lives |
| `made_on`, `resolve_by` | Recommended | `resolve_by` is what stops predictions rotting unresolved |
---
## Table format
| ID | Statement | Domain | Conf. | Made | Resolve by | Outcome |
|---|---|---|---|---|---|---|
| | | | | | | |
| | | | | | | |
| | | | | | | |
---
## Weekly (5 min): resolve what is due
- [ ] Every prediction past its `resolve_by` marked true or false
- [ ] Disputed resolutions noted — a dispute means the statement was underspecified; write the next one tighter
- [ ] New predictions added for decisions made this week
An unresolved prediction teaches nothing. Resolution is the cheap step people
skip, and skipping it is what makes prediction logs quietly useless.
---
## Quarterly (30 min): score
```bash
python3 scripts/calibration_scorer.py --input predictions.json --buckets 5
```
| Metric | This quarter | Target |
|---|---|---|
| Resolved count | ____ | ≥ 20 |
| Brier score | ____ | < 0.20 |
| Skill score | ____ | > 0.15 |
| Overall gap | ____ | within ±10% |
| Worst domain | ____ | — |
**One correction for next quarter:** _______________________________________
Apply exactly one and re-measure over a full quarter. Changing several things
at once means you learn nothing about which of them worked.
If overconfident, the crude correction outperforms the clever one: subtract the
measured gap from every stated confidence for a quarter. Expect the worst domain
to be your own delivery dates — that bias is close to universal, and finding it
is the point rather than a failure.
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 |