Dispatch Haiku sub-agents to gather facts from files, directories, and web lookups.
npx skills add https://github.com/ahostbr/liteharness --skill ls-scout
Please help me install this Agent Skill. Skill: ls-scout Source: https://github.com/ahostbr/liteharness/tree/main/liteharness/catalog/skills/ls-scout Command: npx skills add https://github.com/ahostbr/liteharness --skill ls-scout
ls-scout offloads routine exploration and factual lookups to a cheaper Haiku sub-agent so you can focus on synthesis, decisions, and actions.
Ask Opus to “scout the codebase for how authentication works,” and it will dispatch a Haiku research sub-agent to gather the relevant evidence for you to synthesize and decide.
High-intent search queries and answers about installing and using this skill.
You can automate codebase exploration by dispatching a Haiku research sub-agent to read files and list directories. It proactively gathers structured facts like file paths and function names, returning the evidence for the main agent to synthesize.
Agent orchestration for parallel research involves spawning multiple sub-agents concurrently to handle independent research questions. This allows simultaneous web lookups, file reading, and document summarization to gather facts efficiently.
Yes, you can dispatch a read-only Haiku sub-agent to gather facts before making decisions. It performs scouting tasks like listing directories and reading files, returning the findings to the main agent to synthesize while keeping decision-making centralized.
To scout a codebase for how a mechanism works, you provide a clear research question with sufficient context. A Haiku sub-agent then investigates the files and directories, returning structured findings on the mechanism for you to synthesize.
No, dispatching Haiku sub-agents for codebase exploration does not require explicit requests for each task. The agent proactively initiates read-only scouting, file reading, and document summarization to gather necessary context automatically.
The limitation of a read-only research sub-agent is that it cannot execute code or modify files. It requires clear research questions and sufficient context to function, returning findings for the main agent to synthesize rather than making decisions itself.
danalexilewisSynthesize end-of-chat context and docs into a reviewer-only investigation plan.
PlanxnxGenerate executable implementation plans with exact file paths and code blocks.
jzaq16Coordinate multi-step tasks by orchestrating a swarm of specialized worker agents.
PieroSierraImport markdown content into the raw/ directory of the Second Brain knowledge base.
