decide
/decide
Zapisuje decyzje i pilnuje ich w procesie slice-driven: tworzy rekordy decyzji z poziomem ugruntowania (dowód albo założenie robocze) oraz rozstrzyga, gdy ktoś kwestionuje zamkniętą decyzję.
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 22.09.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Zapisuje decyzje i pilnuje ich w procesie slice-driven: tworzy rekordy decyzji z poziomem ugruntowania (dowód albo założenie robocze) oraz rozstrzyga, gdy ktoś kwestionuje zamkniętą decyzję.
Wyzwalacze z opisu
Argumenty
/decide <optional decision statement, or the decision being doubted>
Oryginalny opis (EN)
Record and gatekeep decisions in the slice-driven workflow: write decision records with a grounding tier (grounded with evidence link, or provisional assumption), and arbitrate when a settled decision is challenged. Trigger when the user says "/decide", "record this decision", "let's lock this in", "I'm doubting decision X", "should we revisit…", or when another slice-driven skill (work, harvest, build, intent) makes or challenges a decision. Invoked with no argument, it takes on the next provisional decision and works it through a Socratic dialogue. Do NOT trigger for read-only lookups ("what
Jak działa
- Record format
- Recording
- The standing-assumption pass (no argument)
- The reopening gate
Podgląd SKILL.md
---
name: decide
description: >
Record and gatekeep decisions in the slice-driven workflow: write decision records with a
grounding tier (grounded with evidence link, or provisional assumption), and arbitrate when
a settled decision is challenged. Trigger when the user says "/decide", "record this
decision", "let's lock this in", "I'm doubting decision X", "should we revisit…", or when
another slice-driven skill (work, harvest, build, intent) makes or challenges a decision.
Invoked with no argument, it takes on the next provisional decision and works it through
a Socratic dialogue.
Do NOT trigger for read-only lookups ("what did we decide about X?") — answer those from
DECISIONS.md directly.
argument-hint: <optional decision statement, or the decision being doubted>
allowed-tools: Read, Glob, Grep, Edit, Write, Bash
---
# Decide — Evidence-Tiered Decision Records
> For the full workflow this skill belongs to, see [workflow-overview.md](../_shared/workflow-overview.md).
You maintain the project's decision log and enforce its two-tier discipline. The log exists to kill re-litigation: settled ground stays settled unless evidence moves it.
The user's argument, if any: `$ARGUMENTS` — either a fresh decision to record (start at Recording, using it as the draft one-sentence form) or a reference to an existing decision being doubted (start at the reopening gate). Tell the two apart by checking `DECISIONS.md`. If empty, run the **standing-assumption pass** below.
## Record format
Decisions live in `decisions/NNNN-slug.md` (zero-padded ascending), indexed one-line-each in `DECISIONS.md` at the repo root:
```markdown
---
id: 12
slug: admin-api-unauthenticated
title: Admin API is unauthenticated
status: accepted # accepted | superseded
grounding: provisional # grounded | provisional
evidence: null # grounded: slice branch / commit / benchmark ref
superseded_by: null
decided: 2026-08-02
---
## Decision
One or two sentences, imperative present tense.
## Context
Why this came up; the alternatives that were live.
## Grounds
grounded: what the evidence showed, with the concrete observation (numbers, transcript, branch).
provisional: the assumption being made and what would invalidate it.
## Consequences
What this binds; which work items / spec sections rely on it.
```
Index line format in `DECISIONS.md`: `NNNN-slug.md — [grounding/status] one-line decision`.
## Recording
When called with a fresh decision: confirm the one-sentence form with the user, classify the tier honestly — **grounded requires a concrete evidence link** (slice branch, commit, benchmark output); "we discussed it thoroughly" is not evidence, that's provisional — write the record, add the index line. When a slice or build supersedes a decision, write the new record, flip the old one to `superseded` with `superseded_by`, and update both index lines. Never edit a superseded record's content — the trail is the point.
## The standing-assumption pass (no argument)
Called with no argument, you work the decision log's oldest standing assumption. Run `../_shared/scripts/board.sh` (relative to this skill's directory) and read its **OPEN DECISIONS** section. Take the lowest-numbered `provisional` record; show the list and say which one you're taking so the user can redirect. Ignore the `undecided: … — awaiting probe N` lines: those are already routed to a slice, and re-deciding them in dialogue is exactly the debate the probe exists to replace. If there are no provisional records, say the log holds no standing assumptions and stop — never invent a decision to review, and never widen the pass into an audit of the log.
Then read the record in full and run a **Socratic dialogue** to reach shared understanding of the assumption — not to talk it into being grounded. Open by restating, in your own words, the assumption and what its Grounds say would invalidate it, then ask **one question at a time** and let the answer choose the next:
- What has happened since it was recorded — code shipped, a slice harvested, a benchmark run — that bears on it?
- What is currently leaning on it (which work items, which spec sections)? If nothing is, it may not be worth carrying.
- What would we observe if it were wrong, and would we have noticed by now?
- Is being wrong here still cheap, or has the accumulating dependency made it expensive?
The dialogue closes on exactly one of these, and you name which:
- **Sharpen** — the assumption stands but the record is vague: tighten the Decision sentence and, above all, make the invalidation condition in Grounds concrete. Stays `provisional`.
- **Ground** — only if the dialogue surfaced a *concrete* evidence link (a landed commit, a harvested slice, benchmark output) that the record didn't cite. Flip `grounding: grounded`, fill `evidence:`, write what the observation showed, update the index line. Agreement in conversation is never evidence; if there's no artifact to link, this outcome is unavailable.
- **Supersede** — the assumption is now known wrong: write the replacement record, flip this one to `superseded` with `superseded_by`, update both index lines, leave its content untouched.
- **Probe** — grounding needs running code and being wrong is expensive: spawn a probe item scoped to exactly this decision and route to `/slice`.
- **Retire** — nothing depends on it any more and it binds nothing: say so, and record that in Consequences rather than deleting the record.
Keep it short. One decision per invocation; close by noting how many provisional records remain, so a repeat invocation drains the backlog naturally.
## The reopening gate
When a decision is challenged (by the user's doubt or by a skill hitting a conflict):
- **provisional** → doubt is legitimate by definition. Don't debate. Offer: (a) if being wrong is cheap, revise the record directly and move on; (b) if expensive, spawn a probe item scoped to exactly this decision and route to `/slice` — the answer comes back as evidence.
- **grounded** → show the record's Grounds first; most doubt dies on contact with the evidence that settled it. It reopens only if the challenger brings *new* evidence or a materially changed context — in which case route to a slice or supersede directly. Absent that, the decision stands; say so plainly.
Never run unprompted review passes over the decision log, and never let another skill re-ask a question a grounded decision answers.
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · mtrense/skills | 100 | MIT |
| vegardjorg/ravenseat | 100 | MIT |
| generous-corp/pulp | 100 | MIT |
| wrg32786/aigent-os | 100 | MIT |
| alirezarezvani/claude-skills | 100 | MIT |
| AO-Cyber-Systems/devflow-claude | 100 | MIT |
| DvJ4Y/shipwright | 100 | MIT |
| blockrunai/blockrun-mcp | 100 | MIT |
| Abilityai/cornelius | 100 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki 000404994960 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100slice
Przeprowadza eksploracyjną próbę: najmniejszy pionowy wycinek działającego kodu, który rozstrzyga jedną decyzję. - 100harvest
Zamyka otwarty wycinek w pracy sterowanej wycinkami: decyzja o losie (zatwierdź, popraw, odrzuć), zamiana obserwacji w uzasadnione decyzje i wpisy do INTENT.md, QUESTIONS.md oraz sekcji Results zadania. - 86log-decision
Zapisuje decyzję albo przyjęte ryzyko w dzienniku decyzji, gdy właściciel lub zespół coś ustali. - 100slice-closure
Rozstrzyga, czy bieżący wycinek pracy jest gotowy do zamknięcia, oddziela ukończenie wycinka od ukończenia zadania i zapisuje wynik w notatkach i TASK_STATE.md. - 100decision-log
Zapisuje, zastępuje i kwestionuje decyzje lidera w dzienniku decyzji hubu, gdy wykraczają poza jedno zadanie. - 100auditing-decision-record-currency
Cyklicznie audytuje zapisane w repozytorium rekordy decyzji (ADR, RFC): spójność statusów, decyzje odwrócone w kodzie, spełnione warunki rewizji, brak sprawdzalnych warunków i technologie po końcu wsparcia.