trellis-meta

Explains local Trellis architecture and customization entry points for AI agents.

Updated Jun 8, 2026
One-click install
npx skills add https://github.com/SilentFlower/flower-trellis --skill trellis-meta-silentflower
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: trellis-meta
Source: https://github.com/SilentFlower/flower-trellis/tree/main/.agents/skills/trellis-meta
Command: npx skills add https://github.com/SilentFlower/flower-trellis --skill trellis-meta-silentflower

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? After running trellis init, a project contains many generated files across .trellis/ and platform directories (.claude/, .codex/, .agents/, etc.). This Skill teaches an AI how that local Trellis architecture fits together and where to safely make customizations without breaking managed files. ## Core Features & Use Cases - Architecture Map: Explains the workflow, task, spec, workspace memory, context injection, and multi-agent channel runtime layers, plus how platform hooks, agents, skills, and commands connect to them. - Ownership-Aware Editing: Distinguishes project-local files from upstream Trellis templates and Flower/Skill-Garden managed overlays using .flower/state.json, .trellis/.template-hashes.json, and patch markers, so edits go to the durable source. - Customization Routing: Reference guides route requests like changing workflow phases, task lifecycle hooks, agent behavior, context loading, spec structure, or adding project-local skills to the correct file. - Use Case: A user asks the AI to make the check agent run extra project commands; the Skill directs the AI to the platform's trellis-check agent file and .trellis/spec/ instead of editing bundled templates that trellis update would overwrite. ## Quick Start Ask the AI to explain how this project's Trellis workflow and agents are wired, then request a specific customization such as changing the planning phase behavior.

Frequently Asked Questions about trellis-meta

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I customize the Trellis workflow in my project?▼

Edit `.trellis/workflow.md`, which is the runtime contract for phases, routing, and workflow-state blocks. First check whether the target section is project-local or managed by Skill-Garden patches, since managed sections must be changed through their owning Patch source.

Where should I put team-specific coding conventions for Trellis?▼

Put long-term conventions in `.trellis/spec/<layer>/` with an index.md entry point, or create a project-local skill with a unique name. Never store project rules inside bundled skills like trellis-meta, because `trellis update` overwrites them.

Can I edit bundled Trellis skills like trellis-meta directly?▼

Direct edits to bundled skills survive only until the next `trellis update`, which refreshes them from upstream templates. Instead create a differently named project-local skill, contribute the change upstream, or adjust the owning Flower/Skill-Garden patch if one manages the file.

Why does my AI agent not read the current Trellis task context?▼

Check whether the platform uses hook push or agent pull mode, then verify the session-start and inject-subagent-context hooks are registered and that `task.py current --source` resolves an active task. Also confirm the task's implement.jsonl or check.jsonl references the right spec files.

What is the difference between platform agent files and .trellis/agents/ definitions?▼

Platform files like `.claude/agents/trellis-implement.md` control direct sub-agent dispatch from the main session, while `.trellis/agents/{check,implement}.md` are platform-agnostic role cards used by the `trellis channel` multi-agent runtime. Editing one does not affect the other.