pandoscope/skills

to-spec

Turn the current conversation into a spec published to the project tracker — no interview, just synthesis of what was already discussed.

Voir la source
Document Skill original

Rendu depuis le dépôt source en conservant titres, exemples, code, tableaux, liens et images.

To Spec

Produce spec (aka PRD) from current conversation + codebase understanding. Do NOT interview user — synthesize what you already know.

Reproducibility invariant

Spec — and tickets later derived from it — must be self-contained: replayed by any implementer, human or agent, builds roughly same application. Every outcome-shaping decision lives in spec itself, or durable docs it references — ADRs (writing-adrs), glossary terms (docs/glossary/). Never conversation context, tribal knowledge, implementer discretion.

Litmus test: two independent implementers could build meaningfully different things → spec underspecified. Add decision to spec, or record as ADR/glossary term + reference.

Grilling gate

Spec is load-bearing. No grilling session (docs/glossary/grilling-session.md) in context → stop, ask user: "No grilling session found — really skip?" Proceed only on explicit confirmation; note skip in Further Notes.

Process

  1. Explore repo if not already done. Use glossary vocabulary (uvx disambiguate <term>) throughout; respect ADRs in touched area.
  1. Sketch seams for testing the feature. Prefer existing seams; new seams at highest point possible. Fewer seams better — ideal is one.

Check seams with user.

  1. Write spec per template — as short as possible, caveman mode preferred (caveman skill), precision and understandability must not suffer. Apply litmus test to every section, publish per consumer repo's tracker conventions (AGENTS.md). Apply ready-for-agent label — no further triage.

Spec template:

markdown
## Problem Statement

Problem, from user's perspective.

## Solution

Solution, from user's perspective.

## User Stories

LONG numbered list: "As an <actor>, I want <feature>, so that <benefit>".
Example: "As a mobile bank customer, I want to see balance on my accounts, so that I can make better informed decisions about my spending."
Extremely extensive — cover all aspects.

## Implementation Decisions

Decisions made: modules built/modified, interfaces, technical clarifications, architecture, schema changes, API contracts, specific interactions.

Hard-to-reverse or surprising decisions → also record as ADR (`writing-adrs`) + reference. Decision only in conversation, not listed here → fails litmus test — write it down.

No file paths/code snippets — stale fast. Exception: prototype snippet encoding a decision more precisely than prose (state machine, reducer, schema, type shape) → inline within relevant decision, note prototype origin, trim to decision-rich parts.

## Testing Decisions

What makes a good test (external behavior only, not implementation details), which modules tested, prior art in codebase.

## Out of Scope

What this spec excludes.

## Further Notes

Anything else.
du même dépôt

Autres Skills

Tous les Skills
pandoscope
Communauté

caveman

Terse communication that stays easy to parse. Drops filler, pleasantries and hedging while keeping full technical accuracy. Chat keeps natural language, because humans read it; files, docs, comments and commit messages save tokens through cut fluff and flat sentence structure, never through nested clauses. Use when user says "caveman mode", "talk like caveman", "use caveman", "less tokens", "be brief", or invokes /caveman.

installations
6
GitHub Stars
1
Mis à jour
23 sept.
pandoscope
Communauté

documenting-decisions

Routes human attention to decisions that matter in agent-generated code. Active during planning, implementing, fixing. Defines when and how to place DECISION markers in code comments. Also applies when reviewing a diff/PR on request.

installations
6
GitHub Stars
1
Mis à jour
23 sept.
pandoscope
Communauté

domain-modeling

Build and sharpen a project's domain model. Use when the user wants to pin down domain terminology or a ubiquitous language, record an architectural decision, or when another skill needs to maintain the domain model.

installations
6
GitHub Stars
1
Mis à jour
23 sept.
pandoscope
Communauté

grilling

Grill the user about a plan, decision, or idea via scored multiple-choice questions. Use when the user wants to stress-test their thinking or uses any 'grill' trigger phrase.

installations
6
GitHub Stars
1
Mis à jour
23 sept.