fwd-skills

Setup wizard for HeadingFWD's optional Claude Code conventions. Runs unattended — zero prompts, always project-local scope: installs every convention in one batch — smartlint Stop-hook, lessons memory file, gitignore entries for fwd runtime artefacts, clear-context prompt on plan accept, and no-attribution (disables the Claude commit/PR byline). Copies bundled payload files into .claude/hooks/ or .claude/lessons/, merges JSON snippets into .claude/settings.local.json, injects an instructions section into CLAUDE.md (lessons), or appends a marker-bracketed block to .gitignore. Idempotent and modular — each feature lives in scripts/<feature>/. Prints a fixed-format Dutch summary from the bundled OUTPUT.md template, never free-form prose. Runs on Haiku. Use only when the user invokes /fwd:setup explicitly.

Voer een stappenplan van /fwd:steps-plan uit — attended, precies één stap per beurt. Per stap; falende test eerst (rood), minimale implementatie langs de Lazy Ladder (groen), volledige gate, vers oordeel door een read-only reviewer-subagent, en een stap-rapport van ±15 regels met "Stap N/M"-teller — daarna stopt de beurt en beslist de gebruiker (ok = commit & door, auto = autonoom afmaken, m = meer detail, stop = pauze, vrije tekst = correctie/vraag/planwijziging). Elke 4 goedgekeurde stappen een tussenbalans door twee doubt-subagents. Use when the user runs /fwd:steps-run <slug>, zegt "volgende stap", "ga door met het stappenplan", of "hervat <slug>". Zonder argument: lijst alle stappenplannen. Niet voor onbeheerd werk — dat is fwd:mission-run.

Plan een klus als een reeks kleine, toetsbare stappen voor attended uitvoering met /fwd:steps-run — de lichte tegenhanger van fwd:mission-plan. Scope het doel kort, pin een Definition of Done mét concreet eindbeeld (input→output-voorbeeld of ASCII-mockup), spreek de testplekken (seams) af, en lever een genummerde stappenlijst waarin elke stap één aantoonbaar gedrag is met een machinaal toetsbaar klaar-criterium. Use when the user zegt "plan dit in stappen", "maak een stappenplan", "steps-plan", of invokes /fwd:steps-plan. Niet voor onbeheerd/overnight werk — dat is fwd:mission-plan.

Plan a multi-agent "mission" — scope a software goal through conversation, write a PRD plus a validation contract (what "done" means, before any code), decompose into features and milestones, then create the mission branch and persist the plan on it. The interactive half of the fwd:mission-* orchestration layer (modelled on Factory.ai Missions); execution is handled afterwards by /fwd:mission-run. Use when the user wants to plan a larger feature as an orchestrated mission, says "plan a mission", "start a mission", "scope this as a mission", or invokes /fwd:mission-plan.

Execute a planned mission — the resident orchestrator of the fwd:mission-* layer (a Claude Code take on Factory.ai Missions). Reads the mission's state.json, drives features one at a time by spawning a fresh coder subagent each, runs adversarial validators (Scrutiny + User-Testing) at milestone boundaries, and records a checkpoint after every unit so the mission resumes from any worktree or clone. Runs autonomously — never prompts. Use when the user runs /fwd:mission-run <slug>, says "run/execute/resume mission <slug>", or wraps it in /loop for a long multi-day run. Pass `status` as a second argument for a read-only progress report.

Plan een implementatie — verzamel codebase-context, presenteer eerst een DoD-voorstel met numbered bullets (akkoord of corrigeer in plain text, géén AskUserQuestion), stel daarna 0-3 verdiepende keuzes via AskUserQuestion, en presenteer 1 of 3 plannen in visueel distincte boxen met spec-strip + TL;DR + Wijzigingen-tabel. Sluit af met (Recommended)-tag op het beste plan en een verdict-block, en legt na de plan-keuze een licht contract vast in `.claude/plan-contracts/<slug>.md`. `/fwd:plan check [<slug>]` toetst achteraf de diff en de DoD-bewijsregels tegen dat contract. Use when user wants to plan a feature, refactor, or change met meerdere opties op tafel, of invokes /fwd:plan.

Stage and commit with conventional message — pre-flight scan blocks risky files (.env, logs, keys, secrets, >1MB)

Black-box self-evaluation for any Claude Code skill. Reads the target SKILL.md, extracts its surface (triggers, CLI flags, input/output formats, documented exit codes, examples), generates ~10 experiments covering happy paths, flag interactions, error paths, and domain invariants, runs each in an isolated workdir under tmp/eval/, and reports pass/fail in a single markdown table. Refuses to run on a dirty working tree. Ends the report with an undo prompt — reply with `x` or `undo` to remove tmp/eval/. Use when the user says "self-evaluate skill X", "shake down skill X", "test skill X end-to-end", "regression-check skill X after my refactor", "does this skill still behave the way SKILL.md claims", or invokes /fwd:skill-eval.

Rewrites a referenced text into plain, readable language — "jip-en-janneketaal". One pass, chat output only; output is always Dutch unless another language is explicitly requested. Invoke when someone says "maak dit leesbaarder", "schrijf dit in jip-en-janneketaal", "rewrite in plain language", "make this readable", or invokes /fwd:jip-janneke. Not fwd:explain (interactive layered walkthrough — explains, does not rewrite). Not fwd:caveman (compresses WITH abbreviations, cuts to fragments — opposite direction). fwd:jip-janneke rewrites the full text once, into complete readable sentences.

Interactief de `.claude/rules/` directory bootstrappen voor een repo. De skill scant de codebase, stelt regelbestanden voor — elk voorzien van minimaal één golden example (een pad naar een echt, exemplarisch bestand) — en schrijft pas weg na expliciete goedkeuring van de gebruiker. Gebruik bij "rules audit", "bootstrap rules", "stel regels op", "maak claude rules", of als je `/fwd:rules-audit` aanroept.

Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when user wants to stress-test a plan, get grilled on their design, or mentions "grill me".

Stress-test a plan by imagining it has already failed, then list concrete failure modes, grade each on Likelihood × Impact, and ICE-rank candidate mitigations for the meaningful ones. Use when user wants to pre-mortem a plan, asks "what could go wrong", wants to harden a design before commitment, or invokes /fwd:premortem.

Compact the current conversation into a handoff document for another agent to pick up.

Interactief een GitHub issue opstellen volgens een vast 5-secties template (probleem, voorbeelden, bevindingen, potentiële oplossing, tests). Skill leest input uit een argument óf de huidige conversatie-context, detecteert bestaande repo-labels, vraagt om een assignee, en laat de gebruiker de complete draft per sectie reviewen voordat `gh issue create` wordt aangeroepen. Use when user wants to create a GitHub issue from a problem they are describing, runs `/fwd:issue-create`, says "maak een issue", or "create issue from context".

Privately work through your assigned GitHub issues overnight — pick the oldest open issue, fix it in an isolated worktree, run tests, commit (no push). Driven by /loop. Strictly read-only on GitHub (no labels, no comments, no PRs) so collaborators can't tell the work was automated; all state local in `.claude/issue-loop/state.json`. Use when user says "work through my GitHub issues overnight", runs `/loop /fwd:issue-fix`, or invokes `/fwd:issue-fix` to fix one issue.

Break down anything heavy — a plan, code file, diff, doc, stack trace, PR, URL, or concept — into a layered walkthrough. Builds a mental model first (problem framed + best-fit form: diagram, analogy, before/after, or causal narrative — plus structure map), then one chunk at a time on demand. Use when the input is too long to skim, when you've come back to something and lost the thread, or when ramping up on unfamiliar material.

Ultra-compressed communication mode. Cuts token usage ~75% by dropping filler, articles, and pleasantries while keeping full technical accuracy. Use when user says "caveman mode", "talk like caveman", "use caveman", "less tokens", "be brief", or invokes /fwd:caveman.