vetcoders

skill_name

Template for a new Vibecrafted skill; replace before shipping.

Spawn external specialized AI agents from the user's fleet (Codex, Claude, Gemini). Use this when you need parallel execution, deep isolation, or task-specific cognitive strengths that surpass generic in-thread delegation. Trigger: "vc-agents", "/vc-agents", "delegate to agents", "spawn".

aicx

An Intention Retrieval Engine for Agents' sessions. aicx (formerly ai-contexters) is a sophisticated parser tool that recovers and keeps the central history of agents' sessions in both human- and agent-readable format. Additionally it provides ad-hoc mode to recover agent output that is too large to read or is unreadable. Works on any Claude Code, OpenAI Codex, Gemini JSON, JSONL-format file regardless of extension (.jsonl, .txt, .output). Generates output path automatically — no -o flag needed.

vc-audit

READ-ONLY falsification of a completed plan or multi-task implementation. Builds a per-task requirements matrix, then proves or refuses each claim against code + tests evidence. Default verdict is UNVERIFIED — PASS is earned, never assumed. Runs whenever a written plan claims completion, regardless of upstream — workflow, implement, marbles, human work, or a mix. Trigger phrases: "audit", "vc-audit", "task-by-task audit", "verify implementation plan", "spec falsification", "post-marbles audit", "did this plan actually land", "weryfikuj implementację", "audyt planu", "co naprawdę wylądowało", "falsyfikacja completion".

vc-decorate

Late-stage visual finishing and experience coherence skill. Detects the user's existing design language, audits system consistency, distinguishes identity from drift, upgrades weak patterns, and proposes tasteful polish that works WITHIN the user's system. Never imposes the agent's taste. Never decorates chaos. First make the system coherent. Then make it feel premium. Trigger phrases: "decorate", "make it look good", "add polish", "smaczki", "micro-interactions", "udekoruj", "dopracuj wizualnie", "curb appeal", "premium pass", "finish the experience", "make it feel intentional", "coherence audit", "design system cleanup", "interactive demo", "animate", "add hover effects", "make it feel nice", "visual polish".

vc-delegate

Native operator-side delegation doctrine for small bounded native cuts. Use this when the operator agent must decide whether work should stay in-process through native subagents or be escalated upward into vc-agents. Trigger phrases: "implement with agents", "delegate to subagents", "zaimplementuj", "run agents", "parallel tasks", "delegate safely", "native agents", "Task tool agents", "implement plan", "uruchom agentów", "subagenty natywne", "bezpieczne agenty", "implement without externals", "no osascript".

vc-dou

Definition of Undone audit skill. Runs a systematic gap analysis across the ENTIRE product surface — not just code. Crawls public URLs, audits repo governance, verifies install paths, checks SEO/discoverability, audits representation surfaces for non-web products, and measures the gap between internal capability and external visibility. Trigger phrases: "definition of undone", "dou audit", "co jest niedokończone", "what's undone", "product surface audit", "completion audit", "plague check", "hydration check", "are we shippable", "czy jesteśmy gotowi", "gap analysis", "co brakuje do launchu", "readiness audit", "packaging gap".

vc-followup

AUDIT-FIRST post-implementation trajectory check. Evaluates whether the work is heading in the right direction, what gaps remain, what drift was introduced, and what the next highest-leverage move should be. Reads the work in motion — code, runtime behavior, architecture, integration — without requiring a single artifact like a PR or commit range as its frame. Product-surface completeness (packaging, install, discoverability) belongs to `vc-dou`, not here. Sibling to `vc-review` (per-implementation diff perception) and `vc-audit` (per-plan spec falsification) in the AUDIT-FIRST perception layer of the pipeline. Trigger phrases: "follow-up check", "followup audit", "czy sa jeszcze luki", "readiness before hands-on", "audit this implementation", "po implementacji", "gaps after agents", "co zostało do zrobienia", "post-implementation review", "czy to idzie dobrze", "czy ten kierunek ma sens", "what still feels off".

vc-hydrate

Packaging and go-to-market hydration skill. Takes DoU audit findings and executes the non-code work that bridges the gap between "it works" and "someone can buy this." Generates marketplace listings, SEO fixes, distribution artifacts, onboarding flows, landing page content, and representation surfaces for products that do not naturally have a public web UI. Trigger phrases: "hydrate", "package for market", "prepare for launch", "przygotuj do launchu", "fix the packaging gap", "marketplace listing", "nawodnij", "make it shippable", "go-to-market", "distribution", "SEO fix", "landing page", "onboarding", "completion sprint".

vc-implement

End-to-end implementation skill for when the user is done talking and needs the thing built. VC-ship WRITE stage: full delivery with autonomous decision making. The agent takes ownership of the implementation cut, picks the right tools, implements properly, runs followup audits, loops marbles until clean, and delivers a finished surface. No phase theatre; no permission-seeking on obvious moves. The user says what; the agent figures out how. Trigger phrases: "implement", "vc-implement", "implement this e2e", "build this properly", "ship the feature", "zaimplementuj to", "full implementation", "od pomyslu do realizacji", "caly feature", "before tomorrow". Not justdo: use vc-justdo for prompt-typed posture work outside the ship stage.

vc-init

Technical due diligence at the start of every repo session — the entrypoint to all repo work, like reading CLAUDE.md. Before touching anything, see the asset as it IS: materialize the Loctree context atlas and READ IT TO THE END, recover intent (AICX), verify ground truth (git/security), and grade the risk. Non-pipeline; runs every session. Trigger: "init", "initialize", "bootstrap", "daj kontekst", "zainicjuj", "przygotuj agenta", "start fresh with context".

vc-intents

Operator-side intention-to-runtime truth audit. Use when the team needs to know which planned implementations actually landed in code, which are only partially present, which never materialized, and what the highest remaining truth is. This skill pulls intentions from aicx, reduces them to a bounded implementation checklist, then verifies each item against the live repo. Trigger phrases: "intents", "co z planu siedzi", "which planned items exist", "what from the plan is in code", "check intent coverage", "planned vs code", "highest truth", "checklist from intents".

vc-justdo

Standalone Just Do posture skill + launcher — not an alias of vc-implement. No ceremony, no best-of-n: take the task and deliver. Task type is defined by the PROMPT (implement, review, audit, research, fix, recon — anything), not by this skill. Carries the vc-ownership posture. Non-pipeline: not a VC-ship cadence phase (unlike vc-implement). Daily rescue when the founder is tired and still needs the work done — orient, act, prove. Trigger phrases: "just do", "just do it", "vc-justdo", "weź i zrób", "zrób to", "ogarnij to", "take the task", "no questions just do it", "zrób review/audyt/research <X>", "bez gadania zrób".

vc-marbles

WRITE step that floods every crack with deliberate over-correction. Single workers see one round, one truth-forcing cut, one commit; the skill at swarm level produces an intentional excess of fixes — marbles in every hole — which `vc-polarize` then strips back to one axis. Use when implementation already exists but the codebase still lies: overgenerated surfaces, drift between runtime paths, false certainty from one-shot agent output, or a product that "works" while remaining fragile. Each worker invocation is isolated and blind to prior marble history. Trigger phrases: "marbles", "kulki", "stabilize", "stabilizacja", "loop until done", "reduce chaos", "fortify the foundation", "adultification", "rzuć kulki", "wypełnij pęknięcia".

vc-operator

Autonomous orchestration posture for conducting a fleet through a planned multi-wave dispatch chain. Use when the agent is not building one slice but reading a plan, building a wave atlas, dispatching peer agents, awaiting durable artifacts, verifying reports and gates, issuing recovery dispatches on stalls, and stopping at the operator button for actions not already permitted by the written plan or current session. Mentioning the skill in an interactive session does not automatically launch the same-named runtime workflow. Trigger phrases: "operator mode", "vc-operator", "Agent-Operator", "tryb operatora", "prowadz fleet", "konduktorze", "orkiestracja", "dispatch the plan", "fire the wave", "dirygentura", "multi-dispatch", "orchestrate this plan", "stop at the button".

vc-ownership

Full-spectrum Vetcoders ownership mode for moments when the user wants Agent to take the wheel and drive a product from A to Z: architecture, coding, runtime debugging, UI polish, packaging, docs, testing, local tooling, agent orchestration, and wow-effect finish. Use whenever the user says things like "take ownership", "you drive", "od a do z", "zrob to cale", "dowiez to", "wow effect", "superprodukcyjny", "manufakturer produktowy", or when the team clearly wants decisive end-to-end execution with minimal back-and-forth. This skill is intentionally pushy: if the user is asking for total delivery, use it even when they do not explicitly name the skill.

vc-partner

Proactive interactive posture for shared steering with the operator. `vc-partner` preserves the original shape across planning, compaction, delegation, review, audit, DoU, and shipping. Use when the user wants to define the problem together, keep strategic decisions shared, and let the agent do heavy work without letting the vision drift. Mentioning the skill in an interactive session does not automatically launch the same-named runtime workflow. Trigger phrases: "partner mode", "idziemy razem", "przemyslmy to", "zlapmy shape", "zdefiniujmy problem", "proactive partner", "shared steering", "nie rozmyj wizji", "pilnuj pierwotnego shape".

vc-polarize

WRITE step that strips back the marbles excess to one truth. Where the swarm of marble workers plastered every crack in deliberate over-application, polarize picks one axis, rejects the competing ones, and aligns runtime, tests, docs, artifacts, and public promises so they all agree. Gated on Loctree `loct prism` bands — `0..4 abort`, `5..8 memo`, `9..12 pass`, `13..15 doctrine`. Emits DoU / release handoff. Trigger phrases: "polarize", "vc-polarize", "wyostrz", "one sharp truth", "code smear", "prism score", "after marbles", "choose one axis", "decisive cut".

vc-prune

Repository curation, not clear-cutting. Map what truly participates in runtime truth versus what is silently parked — then decide revive, archive, or delete. Includes the silencer strip: rip every `#[allow(...)]`, `// nosemgrep`, `eslint-disable`, `@ts-ignore`, `# noqa`, `# type: ignore`, panic-vs-skip pattern, and any other annotation that mutes a quality gate. Run the gates. Listen. Triage with care — `#[allow(dead_code)]` (and equivalents) is often the most valuable smell in a repo: parked work the team forgot about. Surface those as forgotten gems for the operator to decide. This skill is a gem hunter, not a clear-cutter. Trigger phrases: "prune", "strip dead code", "wyczyść mądrze", "strip the silencers", "zdejmij wszystkie ignore", "zobacz co realne", "forgotten gems", "co tam zapomnieliśmy".

vc-prview

Bounded PR, branch, commit-range, or artifact-pack review pipeline: generate prview-rs artifacts then produce a findings-max audit. Use when the user asks to "review PR", "analyze branch", "run prview", "sprawdź PR", "zrób review", "audit PR", "daj findings", "zbadaj branch", "artifact pack", "PR quality check", "merge gate", "findings-max", "deep review", or needs structured diff artifacts with line-level analysis for AI review pipelines.

vc-release

Final outward ship skill. Turns "done in the repo" into "safe, visible, deployable, discoverable, and launchable in the world." Covers release mechanics, deployment topology, reverse-proxy defaults, Semgrep-gated security hygiene, domain and DNS wiring, SEO/indexability, verification challenges, onboarding truth, and post-release smoke checks. Trigger phrases: "release", "ship to market", "publish", "deploy to production", "vc-release", "go live", "launch", "wypuść wersję", "deploy", "release prep", "launch path", "launch checklist", "production checklist".

Data & Documents Listed

vc-research

Standalone triple-agent research skill. Co-define the problem with the user, write a research plan, then spawn claude + codex + gemini simultaneously on the same questions. Three independent reports come back. Synthesize into one gap-free research document ready for implementation. Use whenever the team needs ground truth before coding: unknown APIs, architecture decisions, library assessment, protocol research, best-practice survey, competitive analysis, or any situation where one agent's perspective is not enough. Trigger phrases: "research this", "zbadaj to", "triple research", "research swarm", "3 agenty research", "gap-free research", "zbadaj przed implementacją", "co mówi dokumentacja", "state of the art", "SoTA research", "porównaj podejścia", "analyze options", "research plan", "plan researchu".

vc-review

READ-ONLY bounded code review. Generates structured artifacts with prview-rs, then runs findings-max analysis with falsification-first discipline. Every finding carries an explicit evidence grade (STRONG / MEDIUM / WEAK / NONE) and either passes or fails an adversarial pass. Stage-aware verdicts prevent mid-stage PRs from being judged as fully-staged. Per-implementation perception step in the pipeline; for per-plan post-marbles falsification use `vc-audit` instead; for trajectory direction checking use `vc-followup`. Trigger phrases: "review PR", "analyze branch", "run prview", "sprawdź PR", "zrób review", "daj findings", "zbadaj branch", "artifact pack", "PR quality check", "merge gate", "findings-max", "deep review".

vc-scaffold

Founder-first main brainstorm + planwriting — the armored lighthouse (pancerna latarnia) that carries a single cut, multiple cuts, or a whole project into the autonomous VC-ship pipeline. The WRITE entry of the read/write cadence: produces a measurable, self-sufficient plan a fleet executes with the operator absent mid-flight. This skill should be used when the user asks to "scaffold", "plan this", "architect this", "break this down", "I have an idea", "design the system", "vc-scaffold", "zaplanuj to", "rozrysuj architekturę", "mam pomysł".

vc-screenscribe

Screenscribe workflow skill for analyzing screencast recordings and for working inside the Screenscribe repo itself. Use this whenever the user mentions Screenscribe, screencast review, app review videos, bug demo recordings, HTML Pro reports, transcript-first artifact extraction, extracting actionable findings from narrated videos, batch video analysis, or wants to debug/build/improve the Screenscribe project or the default https://github.com/vetcoders/Screenscribe repository. Prefer this skill even if the user does not explicitly ask for "Screenscribe" but clearly wants a spoken screen recording turned into structured engineering findings.

vc-skillaunch

Distills a completed user workflow, session interaction or pattern into a reusable agent skill. Use when the user asks to turn their workflow, interaction, or multi-step process into a skill, or when they say "make this a skill", "create a skill from what we just did", "package this workflow" or similar. Do not use for creating skills from scratch without an existing workflow (use a generic skill-creator for that).

vc-workflow

This skill should be used when the user asks to "examine and implement", "research then implement", "zbadaj i zaimplementuj", "workflow", "pipeline", "examine → research → implement", "full workflow", "ERi pipeline", "ERi", "plan and implement", "analyze then build", "structured implementation", "przebadaj repo i zaimplementuj", or describes a task that requires understanding code structure before making changes. Orchestrates a three-phase pipeline: Examine (loctree), Research (Brave Search / web), Implement (subagents). Each phase feeds context to the next.

vibecraftsmanship

Meta-doctrine for human taste, agent force, Loctree/AICX structure, and reality-tested shipping posture.

vc-guard

In-flight enforcer sibling of vc-trust. Inventories existing gates (commit-msg, pre-commit, pre-push, loctree-first, classifier hard-stops, commit-msg-diff advisory, trust-block dispatch refuse) and ships one fail-closed proof path: refuse workflow continuation when the trust journal records block for HEAD. Guard never invents settlement letters or re-judges claims. Trigger: "guard", "vc-guard", "gate inventory", "refuse on trust block".

canary

Ownership catalog of a repo: sense planes via loctree, one agent per dynamic scope, add missing docstrings without logic changes, one supervisor commit, then findings report for discussion. Use when the user asks to "canary", "ownership catalog", "docstring sweep", "skataloguj repo", "repo atlas", or runs /vc-canary / vibecrafted canary.

loctree

Structural and literal repository perception before edits, deletes, refactors, or unfamiliar-code work. Loctree maps scope, dependencies, consumers, exact occurrences, and blast radius; AICX supplies historical intent.

vc-dispatch

Operate external Vibecrafted fleet lines with prompt assembly, await/observe, reports, and recovery.