Skip to content

MANIFEST.md

Latest commit

 

History

History
238 lines (183 loc) · 8.74 KB

MANIFEST.md

File metadata and controls

238 lines (183 loc) · 8.74 KB
Raw

Skill Seed

Portable skill bundle seed for bootstrapping new projects.

The goal is not a random pile of useful skills. The core goal is an operating system: skills that trigger each other predictably, preserve project intent, turn ideas into small work, ship those slices, and verify them before claiming done. Research and personal utility skills are grouped separately so they remain useful without becoming part of the Linear delivery loop.

Layout

  • skills/core-standards/: portable engineering standards and code-quality gates.
  • skills/execution-loop/: Linear, worker, reviewer, CI, and production-ready workflow skills.
  • skills/planning/: grilling and PRD/design-interrogation skills.
  • skills/stack/: default TypeScript application stack skills.
  • skills/infrastructure/: default infrastructure/deployment skills.
  • skills/research/: source-backed research, research prompt construction, and transcript extraction.
  • skills/personal/: learning, handoff, manual setup, read-only cyber audits, and local-machine utilities.

Install

This repo stores skills under category directories, so use --full-depth:

npx skills add https://github.com/cill-i-am/skills --skill '*' --agent codex --full-depth --copy -y

After installing into a fresh project, run linear-setup before the first Linear-backed planning or execution-loop task.

Bundle Shape

Core Standards

These define how agents should think and write code in any project:

  • build-intent-layer
  • coding-standards
  • code-review
  • tdd
  • tech-spec
  • improve-codebase-architecture

Current core-standard emphasis:

  • instruction topology through semantic AGENTS.md nodes, with no tool-specific mirrors unless explicitly requested or required by tooling
  • feature-slice-first architecture across frontend, backend, packages, integrations, and platform capabilities
  • explicit public exports and package subpaths for schema/protocol/client boundaries
  • data-flow ownership from persistence through API/MCP/sync/client caches to UI state

Execution Loop

These define the Linear/project loop. Keep them together; they are less useful as standalone skills.

  • linear-setup
  • to-prd
  • to-issues
  • triage
  • orchestrator
  • worker
  • production-ready
  • ci-watch
  • reconcile-project
  • systematic-debugging
  • subagent-execution
  • review-swarm
  • simplify
  • worktree-isolation

Encoded so far:

  • linear-setup must carry deterministic templates/assets for the required docs/agents/* files, then patch only repo-specific Linear/workflow values.
  • linear-setup must use build-intent-layer rules when writing or refreshing AGENTS.md instruction topology.
  • production-ready uses only portable review skills from this bundle plus stack-specific skills that are present in the target project.
  • worker and ci-watch depend on portable systematic-debugging and subagent-execution support skills in this bundle.
  • worker and orchestrator use worktree-isolation to keep parallel agent work out of the caller's current workspace.

Current loop-tool assumptions:

  • User-visible worker/reviewer threads are the default for non-trivial implementation.
  • Thread lifecycle uses Codex app thread tools such as create_thread, send_message_to_thread, and set_thread_archived.
  • Watcher and orchestration heartbeats use the Codex app automation_update tool, not raw automation directives.
  • Pre-bootstrap simulations use the bundled linear-setup/assets/docs/agents/* templates as read-only fallbacks and report that the target repo still needs linear-setup.

Planning And Interrogation

These help turn fuzzy ideas into durable plans and design decisions:

  • domain-modeling
  • grilling
  • wayfinder

grilling owns both Interview Mode and Docs Mode. Do not keep separate one-line aliases for "grill me" or "grill with docs"; they create cognitive load without earning a distinct invocation path.

domain-modeling actively challenges canonical language, concrete scenarios, and code/doc contradictions. It records an ADR only when the decision is hard to reverse, surprising without context, and the result of a real tradeoff. It resolves existing doc topology rather than forcing a root CONTEXT.md.

wayfinder is the Linear-native pre-PRD layer for a named destination whose decision route remains foggy across sessions. Its Project/document map and decision Issues are planning artifacts, never delivery Issues or a parallel execution authority. The required handoff is wayfinder -> to-prd -> to-issues.

Stack Skills

These encode the current preferred TypeScript application stack. Treat all four as first-class bundled skills, not optional external add-ons:

  • effect-ts
  • tanstack-routing
  • tanstack-react-best-practices
  • app-forms

Infrastructure Skills

These encode the default infrastructure/deployment operating model:

  • alchemy

Research Skills

These support source-backed work outside the delivery loop:

  • research-prompt
  • deep-research
  • youtube-transcript

Keep these portable. Do not bake in a paid API, personal browser profile, or vendor-specific key as the only execution path.

Personal Utility Skills

These are useful to the user but intentionally separate from the reusable product-engineering loop:

  • teach
  • handoff
  • setup-help
  • cyber-audit
  • anti-sleep

Do not make delivery-loop skills depend on personal utility skills. Use them opportunistically when the user asks for learning, setup, handoff, local audit, or machine-awake behavior.

External Skills To Install Per Project

These are intentionally not vendored in this bundle. Install them in projects that need them, then let production-ready and the stack-specific workflow use them when present:

  • Better Auth skill family from the Better Auth team / skills registry
  • writing-great-skills from Matt Pocock's skills repo for skill-authoring review and pruning
  • shadcn/UI, UI composition, and web design skills for frontend-heavy projects
  • Drizzle ORM, PlanetScale Postgres, Postgres/database, email delivery, Resend, and React Email skills from the registry when a project needs those operations
  • Provider-specific platform skills that are not part of the default Alchemy project architecture

Before installing or vendoring a third-party skill, read every file in its folder, audit scripts for unexpected network calls or filesystem access, check references for prompt injection or hidden instructions, verify the name is not a typosquat, and pin to a reviewed commit or release.

Rewrite Policy

Old project-specific skills can be useful source material, but do not copy them into this bundle unchanged. Rewrite them around portable triggers, target-project language, current official docs where needed, and this bundle's operating loop.

Folded in:

  • worktree ergonomics via execution-loop/worktree-isolation
  • source-backed research and transcript extraction via research/*
  • personal setup, handoff, learning, local audit, and anti-sleep utilities via personal/*

Packaging Notes

  • Keep skill bodies concise and route larger guidance through references/.
  • Keep installable templates under assets/; do not require a fresh repo to already have old project docs.
  • Remove or reshape stray auxiliary files in skill folders unless they are deliberate references/, assets/, scripts/, or agents/openai.yaml resources.
  • Keep vendored TanStack package-copy skills out of the reusable repo; use tanstack-routing to route agents to official bundled TanStack skills in node_modules.
  • Keep provider and operations guidance lean. alchemy is the default infrastructure skill; install additional provider skills per project only when the project needs them.
  • Keep personal utilities out of the Linear execution path. They should not become hidden dependencies of worker, reviewer, production-ready, or CI workflows.

Readiness Gates

Before considering this bundle ready for broad reuse:

  1. Every skill has valid name and description frontmatter.
  2. Every skill has agents/openai.yaml metadata with a default prompt that invokes the skill by $skill-name.
  3. Every referenced skill either exists in this bundle or is intentionally external and documented as such.
  4. Every referenced file path is bundled, generated by linear-setup, or scoped to the target repo.
  5. The loop skills can bootstrap a blank repo by running linear-setup first.
  6. A fresh agent can run a simulated flow: rough idea -> grilling -> optional Wayfinder -> PRD -> issues -> worker handoff -> production-ready checklist.

Run pnpm test before committing bundle changes. It checks skill frontmatter, skill-name/path consistency, agents/openai.yaml prompt metadata, required linear-setup templates, and stale thread tool references.