uraxii/dotai/bulk-reader

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Delegate large reads to a small, fast subagent. Use for questions about files >350 lines, questions across 3+ files, or large diff summaries.

Install

npx vskill@latest install uraxii/dotai/bulk-reader

SKILL.md

The same SKILL.md was published earlier as spotify/portal-ai-plugins/bulk-reader.

Keep the large file out of the main agent's context. Give a read-only subagent absolute file paths and a specific question, not the file contents. Ask for a concise answer with path:line citations, at most five bullets unless the question needs more. Do not ask it to return the whole file or large excerpts.

  • Claude Code: use the Agent tool with subagent_type: "Explore",

model: "haiku", and run_in_background: false.

  • Copilot CLI: use the task tool with name, agent_type: "explore", a short

description, prompt, and mode: "sync". The built-in explore agent uses read tools. Set model: "gpt-5.4-mini" when the task schema exposes it and the model is available; otherwise keep explore's fast default model. Do not invent unsupported task parameters.

  • Codex: use an available subagent tool with a read-only exploration role and

the smallest available model. Tool names and model controls vary; inspect the exposed schema. If there is no subagent mechanism, use targeted searches and chunked reads yourself.

Include these instructions in the subagent's prompt:

Answer <specific question> using <absolute paths>. Read only; do not edit or
delegate again. Return a concise answer with file paths and line numbers,
not the full contents. Large reads hit the same shunt hooks in your context.
Use targeted searches or chunks of at most SHUNT_MIN_LINES lines (default
350): Read with offset and limit, view with inclusive view_range [start, end],
or sed -n 'START,ENDp' FILE. For all/every/total questions, cover the whole
file with chunks or a complete targeted search before answering. Never
infer completeness from the first chunk. A deny reason suggesting
delegation applies to the parent; you are already the reader, so use chunks.

Ask follow-up questions through the subagent with the same paths. Before editing, re-read only the exact cited lines with a targeted read in the main agent. For exact content rather than a summary, read in chunks of at most SHUNT_MIN_LINES lines until you have the required content. No Portal CLI is needed for reads.

Security

  • Security scanPassed, 90 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

7-day trend score
5

Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/uraxii/dotai/bulk-reader/badge)](https://verified-skill.com/skills/uraxii/dotai/bulk-reader)

Something wrong with this skill or its page? Tell us.