trellis-research

Searches codebases and external docs, persisting findings to task research directories.

Updated Sep 5, 2026
One-click install
npx skills add https://github.com/jiozhaoyue/ST-BgLoader --skill trellis-research-jiozhaoyue
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: trellis-research
Source: https://github.com/jiozhaoyue/ST-BgLoader/tree/main/.kimi-code/skills/trellis-research
Command: npx skills add https://github.com/jiozhaoyue/ST-BgLoader --skill trellis-research-jiozhaoyue

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Research findings shared only in chat replies get lost when conversations are compacted or sessions end. This Skill ensures every codebase search and external documentation lookup is persisted as a markdown file under the active task's research/ directory, so findings survive across sessions. ## Core Features & Use Cases - Internal Code Search: Locates files, components, and code patterns with specific file paths and line numbers. - External Research: Gathers library documentation, API references, and best practices. - Mandatory Persistence: Writes each research topic to {TASK_DIR}/research/<topic>.md in a structured format with findings, code patterns, and caveats. - Strict Write Boundaries: May only write inside the active task's research/ directory; code, specs, and config files are off-limits. - Use Case: A main agent dispatches this sub-agent to investigate how authentication is implemented; the agent searches the codebase, writes research/auth-flow.md, and replies with only the file path and a one-line summary. ## Quick Start Dispatch the trellis-research agent with an active task path and a research question, and it will write its findings to that task's research directory.

Frequently Asked Questions about trellis-research

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

FAQPage Schema
How do I persist research findings across AI coding sessions?▼

Dispatch the trellis-research agent with an active task path. It writes each research topic to {TASK_DIR}/research/<topic>.md in a structured markdown format, so findings survive conversation compaction and remain readable in later sessions.

How does the research agent know which task directory to write to?▼

It runs python ./.trellis/scripts/task.py current --source to resolve the active task path, or uses an 'Active task: <path>' line from the dispatch prompt. If no active task exists, it asks the user rather than guessing.

Can the trellis-research agent modify source code files?▼

No. It may only write markdown files under the active task's research/ directory. Code files, spec files, platform configs, and other task directories are forbidden; code changes should be delegated to the trellis-implement agent instead.

What format do research output files follow?▼

Each file includes the query, scope, date, a table of files found, code patterns with file:line citations, external references, related specs, and a caveats section for anything incomplete or not found.

Why does the agent reply with file paths instead of full research content?▼

The files are the contract between agents. Returning only paths and one-line summaries keeps the main agent's context small while guaranteeing the full findings remain available on disk for any future session.