Skip to content
Back to skills

Wayblade

ASecurity

Manually-invoked delivery-contract skill. A single continuous Claude Code context owns an entire delivery end-to-end — clarify, plan, implement, debug, validate, docs, commit/push, PR — delegating only for the mandatory Ruinor review and for specialists on explicit request. Use when one session should carry a whole change without full multi-agent orchestration.

  • 3 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added November 11, 2026
ai-agentsgogit

Works with

  • claude code

Security analysis

A100/100

Scanned November 11, 2026

npx -y skills add alkofu/ai-tpk --skill wayblade --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Wayblade?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Wayblade
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/alkofu-wayblade/badge)](https://www.skillsdirectory.com/skills/alkofu-wayblade)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: wayblade
description: Manually-invoked delivery-contract skill. A single continuous Claude Code context owns an entire delivery end-to-end — clarify, plan, implement, debug, validate, docs, commit/push, PR — delegating only for the mandatory Ruinor review and for specialists on explicit request. Use when one session should carry a whole change without full multi-agent orchestration.
argument-hint: "[--self-review] [review flags] <contract>"
disable-model-invocation: true
---

# Wayblade — Single-Context Delivery Contract

<!--
Behavioral note: `disable-model-invocation: true` removes this skill's `description`
from Claude's context entirely, so Wayblade never fires from model auto-invocation.
As of Claude Code v2.1.196 that flag also prevents the skill from firing via
scheduled tasks. Both consequences are intended for a manual-only skill — do not
mistake them for a defect. `user-invocable` is deliberately left at its default
(true) so Wayblade stays visible in the `/` menu; do not set it to `false`.
-->

Wayblade is invoked manually (`/wayblade`) when a user wants **one** continuous Claude Code context to own a whole delivery contract, rather than the full Dungeon Master multi-agent orchestration.

## Ownership + DM boundary

A single continuous root context owns the full delivery contract end-to-end: **clarify → plan → implement → debug → validate → docs → commit/push → PR**. This root context does the work itself with its full tool set — it does not decompose the contract across a planner/implementer/reviewer team. The only mandatory delegation is the Ruinor review gate (below); specialists are delegated only on explicit request.

**Topology-drift escape valve (routing by topology, not task size):** if mid-stream the work turns out to be *multiple independent streams*, or it *needs formal planner/implementer separation*, or it *needs the resumable audit lifecycle*, **STOP and recommend starting a fresh Dungeon Master session.** This is a routing decision about the shape of the work, not its size — a large-but-single-stream contract stays in Wayblade; a small-but-genuinely-multi-stream one does not.

## Clarify + plan

Understand the task before planning it. First invoke the `clarify` skill and follow it — short code scan, restated task, every clarifying question in one batch — so the user answers once and the rest of the contract runs without check-ins about intent. Then, when the contract is multi-step or has more than one viable approach, invoke the `planning` skill: it starts from the Task Brief, asks only on a real fork between approaches, and asks for plan approval only when the work is complex. Keep the brief and plan in the conversation; do not write them to `~/.ai-tpk/plans/` (that is DM's plan-file lifecycle). For a trivial, unambiguous contract, skip both and go straight to implementation.

## Implement + debug

For the implement and debug steps, invoke the `implementation-discipline` skill and follow it — minimal diff, pattern matching, incremental verification, and its three-attempt escalation limit (escalate to the user). Do not reproduce its contents here.

When a failure has no known cause (an open-ended "why is this broken?" rather than a failing test with an obvious fix), invoke the `investigation` skill first and agree on the root cause before changing code.

## Docs

For the docs step, invoke the `docs-writing` skill: update user-facing docs the change made stale or incomplete, in the same PR.

## Constitution handling

Wayblade reads the target repo's `.claude/constitution.md` directly (root session, real cwd) — there is no heavyweight preflight. Constitution injection happens at exactly **one** point: the Ruinor delegation handoff.

## Ruinor handoff (mandatory gate)

Mirror the Dungeon Master's handoff. The Ruinor delegation prompt contains:

- A `## Project Constitution` block at the **top** of the prompt, formatted per the shared block format in `claude/references/constitution-block-format.md` (wrap the verbatim contents of the target repo's `.claude/constitution.md`).
- The **worktree-absolute** file paths of the changed files.
- Changed-files / commit-range context so Ruinor can locate the work; **Ruinor runs its own `git diff`** and is **NOT** given the Worktree Context Block.

**Conditional constitution injection:** include the `## Project Constitution` block **only when** the target repo's `.claude/constitution.md` exists. Wayblade is installed globally and invoked in arbitrary repos, most of which will NOT contain that file (it is ai-tpk-specific). When the file is absent, **skip constitution injection silently — no warning, no error, and no empty, placeholder, or hallucinated block** — exactly mirroring DM's conditional/no-op behavior. Everything else in the Ruinor handoff is unchanged.

## Gate 1 — Review (do not skip)

**Before declaring the contract done, you MUST call Ruinor.** This gate is mandatory by default and is waivable **only** via the `--self-review` flag. When `--self-review` is passed, do not silently self-approve: present the raw diff and collect a verdict, acting purely as a conduit for the review outcome. Absent `--self-review`, a contract is never "done" until Ruinor has reviewed it.

## Gate 2 — Validate before PR (do not skip)

**Before opening any PR, you MUST run the `validate-before-pr` skill.** No PR is opened until its lint and format checks pass.

Both gates above are written as deliberately unmissable prose for compaction/salience resilience — so they survive long contexts and are not dropped when the conversation is compacted. (The compaction-resilience rationale is partly empirical, not an over-engineered guarantee.)

## Delivery

Invoke the existing skills in order — do not reproduce their contents: use `commit-message-guide` for commit messages, run `validate-before-pr` as the pre-PR gate, then use `open-pull-request` to open the PR.

## Specialists

Specialists (Riskmancer / Windwarden / Knotcutter / Truthhammer) run on **explicit request only** — Wayblade does not reproduce DM's specialist-triggering matrix.

## Worktrees + capability-gap note

Use native `claude -w` worktrees. Wayblade intentionally does **NOT** reproduce DM's `.worktrees/` sidecar subroutine, plan-file lifecycle, resumability, or audit trail. These are deliberate trade-offs for a lean single-context path — when a contract genuinely needs them, use the topology-drift escape valve and recommend a fresh DM session.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

View all in ai-agents →

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.