AGENT SKILL

study-roadmap-builder

Build folder-structured quarterly study/learning roadmaps.

License 许可
MIT
最近更新
2026-11-15
Stars
★ 0
Forks
0

01 前言 Front Matter

name
study-roadmap-builder
tags
[ "productivity", "learning", "roadmap" ]
author
hermes-agent
license
MIT
version
1.0
description
Build folder-structured quarterly study/learning roadmaps.
when-to-use
Use when the user wants to build a folder-structured quarterly study/learning roadmap with specific sub-fields per quarter, or to teach a section from an existing roadmap.
argument-hint
<build | teach> [quarter | field | section]

02 资源文件 7

03 SKILL.md 正文

Study Roadmap Builder

Build study plans the user actually keeps and follows. For this user: Indonesian, lifelong learner, wants long-term compounding, commits ~2 h/day with one field per day (no context switching), and thinks in quarterly rotations of specific sub-fields.

Global rules

  • Do not ask permission for read-only next steps. "Want me to open the reference file?" / "Should I read the next section?" — just do it. Asking for a trivial read signals you haven't read the material. The user has corrected this explicitly and gets irritated by it.
  • Specific sub-fields, not broad domains. Never plan "math" or "biology" — plan a concrete slice (e.g. "linear algebra", "molecular cell biology", "classical mechanics"). The user explicitly rejected whole-discipline fields as too large.
  • Foundational → unlocks next. Each quarter's fields are prerequisites that open the next quarter's fields. State the unlock relationship for every field ("F1 linear algebra unlocks ML in Q2").
  • Flat section files, no sub-folder. Each section of a field's source roadmap is its OWN file inside the field folder: NN-section-slug.md. No per-section sub-folder, no plan.md inside a sub-folder. The user explicitly wants this flat layout.
  • Quarterly rotation, ~7 specific fields per quarter. The user commits ~2 h/day but rotates one field per day across ~7 fields, so each field actually receives one 2-hour session per week (~12 sessions / quarter). Use that number to size how many section-subtopics a field can truly cover in the quarter. ~26 h is the theoretical maximum; realistic per-field throughput is lower because the field only runs on its rotation day.
  • No context switching within a day — the user's hard rule; the plan must make one field per day explicit, never two fields on the same day.
  • Always end with a concrete start-now action, not just the file tree.
  • Honest hour math. Don't promise mastery of a whole discipline in 26 h; promise mastery of one specific slice, which is the point of the design.
  • Don't over-scope a field to its full source roadmap. Expand every section of the public tree so nothing is lost, then cut the live curriculum to what the quarter's rotation cap can actually cover. The user pushes back on sections they cannot finish ("I only have 3 months"). List the remainder as a named Q2 backlog, never imply the whole tree fits.
  • Don't name the user's files after the source site. No roadmap.sh in the tree; files are <Field>FieldDoc.md, NN-section-name.md, and optionally a field-named reference file. Name sections by topic ("S01 Vectors and Spaces"), and if a reference tree file is kept, call it by the field name only (e.g. linear-algebra-reference.md), never the source domain. The user corrected this explicitly.
  • Keep the content self-contained, not a verbatim dump. Expand each section's subtopics and concepts into the section file's own checklist. Do not rely on a separate roadmap-sh-*.md full-tree file; if you save one, it is optional reference, and it must not be named after the source site.
  • Trim to the actually-reachable scope, and be honest about it. A field studied one day per week in a 7-field rotation gets 1 two-hour session a week, so ~12 sessions in a quarter. That is the real cap, not "26 h per field." List only the sections that fit; the rest is a named Q2 backlog, not a promise.
  • Every field has an exit test — a real benchmark, not "I feel done." No pass = stay on that field. End-of-quarter, fields are "done" when their exit test passes.
  • Keep a no-forgetting layer. 2 h/week spaced review (Anki) of everything learned so earlier quarters don't decay. Monthly 1-h meta-review: what's decaying → add to Anki; what's strong → mark done.
  • Use free resources. Khan Academy, 3Blue1Brown, CrashCourse, MIT OCW.

Routing

Step Reference Purpose
Build procedure references/procedure.md — 7 steps: clarify → pick 7 specific sub-fields → verify hour budget → fetch public roadmap full tree → scaffold folder structure → update README.md → give start-now action The full build workflow with per-step guidance.
Deliverable structure references/deliverable-structure.md — the flat-per-field filesystem layout, <Field>FieldDoc.md vs NN-section-name.md, the 2-hour 3-part session template (Input → Solve → Write) How to scaffold the filesystem; do not restore the old timed 4-part format.
Prose style references/prose-style.md — the user's writing standard for this repo (no trailing parenthetical in headers, no emoji, lead with the point, no invented claims, plain active voice) Match on first draft, not on correction. Repo AGENTS.md carries the full rules — read it before writing docs in that repo.
Interactive teaching references/interactive-teaching.md — the full hard rules of the teaching mode (session start, turn shape, scoping questions, term definitions, language and register, long-session degeneration) Building the roadmap is step one; teaching a section is a distinct mode. Running it like a chat assistant is the failure mode this section exists to prevent.
Video sourcing references/sourcing-videos.md — the verified curl pipeline (DuckDuckGo lite → oEmbed verification → duration scrape) and the fetch-tool calls that do not work on YouTube URLs For a session's input block.
Public roadmap API references/public-roadmap-api.md — how to pull the content from a public roadmap FETCH ITS FULL TREE, then TRIM to the quarter's reachable scope.
Practice problems references/practice-problems.md Problem sets and worked examples for the build/teach modes.
Reference example (Q1) F1 linear algebra · F2 basic statistics · F3 molecular cell biology · F4 genetics · F5 classical mechanics · F6 technical English · F7 personal finance Each ~26 h; unlocks ML/data, biochem, E&M, technical reading, and economics in Q2.

04 同仓库更多 Skills