Guides customization of local Trellis workflow, hooks, agents, and platform files.
npx skills add https://github.com/exhyy/face-crop --skill trellis-meta-exhyy
Please help me install this Agent Skill. Skill: trellis-meta Source: https://github.com/exhyy/face-crop/tree/main/.claude/skills/trellis-meta Command: npx skills add https://github.com/exhyy/face-crop --skill trellis-meta-exhyy
trellis init, teams often need to adapt the generated Trellis setup to their own workflow, but it is unclear which local files control phases, hooks, agents, skills, and context injection. This Skill teaches an AI assistant the local Trellis architecture so it can safely modify .trellis/ and platform directories without touching upstream source code. ## Core Features & Use Cases - Architecture Guidance: Explains the three-layer local Trellis model (workflow, persistence, platform integration) and which files are editable versus managed. - Customization Routing: Maps user requests like "change the workflow" or "the agent did not read specs" to the exact reference document and file to edit. - Platform Coverage: Documents file locations and integration modes for Claude Code, Cursor, Codex, OpenCode, Kiro, Gemini CLI, GitHub Copilot, and more. - Use Case: A user says "make the check agent run our lint command before finishing." The AI reads this Skill, opens the change-agents reference, and edits the platform's trellis-check agent file plus .trellis/spec/ accordingly. ## Quick Start Ask the AI to customize your local Trellis setup, for example: change the workflow so tasks can be skipped for one-line fixes in this project.High-intent search queries and answers about installing and using this skill.
Edit `.trellis/workflow.md`, which is the local workflow source of truth defining phases, skill routing, and workflow-state prompt blocks. After editing, check that platform skills, commands, and agents still describe the same flow.
Edit the platform agent file such as `.claude/agents/trellis-implement.md` or the Codex TOML equivalent. Keep role boundaries intact, specify read order starting from the active task PRD and JSONL context, and define explicit write boundaries.
Trellis generates files for Claude Code, Cursor, OpenCode, Codex, Kiro, Gemini CLI, Qoder, CodeBuddy, GitHub Copilot, Factory Droid, and Pi Agent, plus main-session workflow platforms like Kilo, Antigravity, and Windsurf. Actual directories depend on which `trellis init` flags were run.
First verify the active task with `task.py current --source` and validate the task's implement.jsonl or check.jsonl. Then determine whether the platform uses hook push (edit the inject-subagent-context hook) or agent pull (edit the agent file's read steps).
No. Customization targets are local project files: `.trellis/` and platform directories. Only work on upstream Trellis source when explicitly contributing a PR, forking, or changing npm package contents.
Put long-term engineering conventions in `.trellis/spec/` with an index.md per layer, or create a project-local skill for project-specific flows. Do not store team-private rules inside the public trellis-meta skill.
EcoPasteHubGuides AI agents in understanding and customizing local Trellis architecture within user projects.
LonelyHerbivoreArchive completed Trellis tasks and record session journals with work commit hashes.
yizhitangtongxueRecords work sessions and archives completed tasks in Trellis project journals.
21essGuide customization of local .trellis architecture and platform directories.