deep-investigate
Deep Investigate
The problem this skill solves: when asked to investigate something non-trivial, a single agent tends to write a short, shallow markdown summary — a few bullets per topic, no file paths, no code excerpts, no edge cases. The investigation fits in one context window, so anything that doesn't fit gets compressed out. The result reads like a table of contents, not a reference document.
This skill fixes that by splitting the work. One scout pass produces an outline grounded in real exploration. Then each section is researched by its own subagent, with its own fresh context window, its own depth budget, and a clearly bounded scope. The outputs get stitched into a directory of long, concrete section files plus an index.
When to use
Trigger when the user explicitly asks for depth on a non-trivial topic. Examples that should trigger:
- "Do a deep investigation of how authentication works in this codebase"
- "Write me a detailed report on the event loop in Node"
- "I want a thorough writeup of the payment flow"
- "Really dig into why the build is slow and document everything"
- "Research transformers in depth and give me something I can refer back to"
Do not trigger for quick questions ("how does X work?", "what does this function do?", "find the file that handles Y"). Those get a direct answer, not a skill invocation. If the topic is small enough to fit in a few paragraphs, don't use this skill.