cogikyo/dotfiles/workflow

Claude Code and agent skill by cogikyo/dotfiles

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Use to write workflow fences for proposals with parallel lanes, conditions, joins, or bounded repair, and launches of two or more children; the TUI renders the card and dependency graph.

Install

npx vskill@latest install cogikyo/dotfiles/workflow

SKILL.md

Workflow fence

Follow COLLAB.md's Delegation rule for proposals, launches, and approved waves. The approved contract sets authority and routing; this skill defines the declarative fence. Use diagram for source-derived architecture diagrams and annotated trees.

Declare the work

Every workflow fence has a title, then a required summary of one to three sentences on what the work achieves and why it matters to the user; never restate the title or describe scope, constraints, shape, or step interactions. The renderer word-wraps the summary; nothing else wraps. Blank lines are ignored; the renderer supplies the graph and styling. It flags invalid declarations inline, including unknown keys or step numbers, duplicate numbers, and missing child models. Start each step at column zero as number owner: headline, with a unique positive integer, normally in order. Write every headline, including the workflow heading, as `Title: context`. Use a Title Case title of 2–4 words and at most 20 display cells; the renderer emphasizes it. start refuses colon-less headlines over 20 cells. Keep checklist items under about 70 characters, with no trailing periods. Write the fence in reply text before workflow({ start: true }); it cannot read a fence that exists only in reasoning.

  • Owners: self for Collab's work, gate only for Collab's own decisions, and you for user actions and approvals, including route changes.
  • Children: use role/scope [lane] provider/model[#variant] as the owner.
  • Roles are build, review, scout, and verify; scope is the agent name after the slash.
  • The optional lane is a kebab-case name for its concern, such as preset-tests; omit it for a one-shot child.
  • Require the model token to match the child's launch; add why: only for a departure from ROUTING.md's routes and guidance.
  • Details: indent each line by two spaces under its step.
  • Use - text for what the step does; builder checklists name the files or behavior they change.
  • Use done: text only for specific acceptance beside a checklist, never as the only detail line; omit generic verdicts.
  • Use after: 1 2 for dependencies; omission defaults to the previous step.
  • Give parallel steps explicit `after:` dependencies, or after: none for independent roots.
  • Joins wait for all predecessors; if: condition steps form an exclusive alternative and are required only when selected.
  • Use repair: 1 max 2 on a gate for bounded repair, or repair: 1 2 max 2 for parallel lanes that each resume only when they have findings.
  • Repair targets must be upstream of their gate.

Gate and review checklists name concrete risks, invariants, and failure modes to look for in evidence, never the inputs they read. Approvals, commits, and Collab bookkeeping are headline-only; omit checklists, done:, and if: for conditions already implied by Collab's process. Repair routing, limits, reruns, and evidence invalidation are process mechanics, not step prose.

Process judgment

Size each workflow to the headroom workflow reports. Judge the actual work: wide, cheap steps can fit where a few deep sequential steps cannot. End each workflow at a commit, compaction into the next workflow, a you gate, or done.

Keep cheap checks inside their owning step, avoid splitting one assignment by file, and check parallel write scopes for overlap. A gate marks only Collab's routing, acceptance, or repair-versus-return decision; routine report receipt needs none. Name each repair's owner and invalidated evidence in Collab's repair brief, not the card; record its limit through repair: … max K. Usually resume the same lane with findings as a delta, then repeat the affected checks. Tie dependency guarantees to source locations and the exact version in use. Denied permission, exhausted repair, and missing approval return a blocker unless an approved permitted route remains.

Security

  • Security scanPassed, 90 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

Repository stars
122
Repository forks
6
7-day trend score
29

Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/cogikyo/dotfiles/workflow/badge)](https://verified-skill.com/skills/cogikyo/dotfiles/workflow)

Something wrong with this skill or its page? Tell us.