Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/mryll/skills/study-sessionnpx skills add mryll/skills --skill study-sessiongit clone --depth 1 https://github.com/mryll/skillsWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00117 | $0.01670 |
| Opus 5 | $0.00059 | $0.00835 |
| Sonnet 5 | $0.00023 | $0.00334 |
| Haiku 4.5 | $0.00012 | $0.00167 |
Grade A, and why
study-session scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 75 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Study Session — recover and measure the theory of a subsystem
A session over real code in the repo. The user reads and explains; you probe and grade. The goal is Naur's "theory": the user can defend the subsystem's design, invariants, and failure modes without help. Two modes:
- Deep (45-60 min): learn or repair one subsystem/unit. Default for
not-assessed,practiced, orcode-changedtargets. - Checkpoint (10-15 min): measure retention of a due unit — one cold reconstruction + one transfer scenario. Default when the row's next-due date expired.
Core principle: cold evidence is graded separately from learning. What the user produces BEFORE your corrections is what gets measured; what they learn after counts as learning, never as certification.
Finding the comprehension map
The map's schema (columns, status vocabularies, event format, spacing rule) lives in the user's knowledge wiki — this skill defers to it. To locate the map:
- Resolve the wiki root: follow the
~/.claude/work-wikisymlink if present; otherwise ask once and offer to create that symlink. - Glob
**/projects/*-comprension.md(or the wiki's declared comprehension-map convention) withkind: comprehension-mapfrontmatter; match the current repo/workspace (git remote or basename) against each map's repo/paths column — one project may span several repos. - Exactly one match → use it. None → derive the path from the project name per the wiki's convention, confirm with the user, create from the wiki's schema template. Several → ask.
No wiki or schema at all → offer a minimal inline map: | Subsystem | Repo/paths | Current evidence | Scope | Last passed | Next due | Gaps | with outcome (not-assessed | practiced | partial | defensible) and validity (current | evidence-expired | code-changed | code-currentness-unknown) vocabularies.
Target selection (risk-first)
Propose ONE target with a one-line reason, in this priority order — the user can override:
- Relevant code changed since the row's last evidence SHA (
code-changed). - An open gap implicated by a recent MR or incident.
- High-blast-radius unit overdue.
- Any other overdue unit.
- Untested low-risk units.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 75 lines · 117 tokens per session scan A 839b8c6b1afd
study-session is a skill published in the GitHub repository mryll/skills (3 stars, last pushed 13d ago), licensed MIT. It adds 117 tokens to every session and 1,670 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
lesson
Produce or continue a deep-dive learning-track lesson (a focused mini-course on a hard subject the project needs), or review the user's track exercise work. Use when the user asks for the next lesson, help with a track exercise, or to start a new track.
cangjie-skill
Distill a book, long-video transcript, podcast, course, or interview into a coherent set of executable skills. Use when the user asks to "拆书" / "蒸馏一本书" / "把 XX 书做成 skill" / "把这个视频/播客/课程蒸馏成 skill" / "turn a book or video into skills" — i.e. wants the frameworks, principles, and methodologies in long-form content…
naval-almanack
当用户询问《纳瓦尔宝典》这本书本身(书名/作者/章节/整书概览/术语)，或涉及本书未独立成 Skill 的主题时调用: 幸福训练/欲望管理、接受与内耗、冥想与心猴、地位游戏与攀比、选朋友与圈子、打工vs创业(委托代理)、 诚实沟通、身份与自我形象、期望与愤怒、生命意义、该不该信玄学(理性验证)、长期复利与声誉、判断力训练。 以下意图优先使用同包独立 Skill: wealth-structure(致富/股权期权)、productize-yourself(职业定位/副业方向)、…
reading-metaskill
当用户想养成阅读习惯、问「读什么书/怎么读」「如何学习新领域/怎么入门某学科」时调用。 核心理念: 阅读是终极元技能; 读你所爱直到爱上阅读, 没有读完义务; 读原著与经典优先; 以教促学; 每天1-2小时即可进入极少数人行列。 不适用于: 具体某本书的书评、考试备考资料选择。 Triggers: 阅读/读书/怎么学习/入门/原著/书单/reading/how to learn.
principal-agent
当用户选择职业/组织/合作方式、困惑「为什么大公司磨洋工/小公司拼命」「该不该自己干」时调用。 核心理念: 委托人(主人)会把事做好, 代理人会为自己利益优化; 收益与创造价值绑得越紧, 越像委托人; 别让媒体洗脑你需要代理人。 不适用于: 具体薪酬谈判数字、组织架构设计细节。 Triggers: 激励/代理/主人/打工 vs 创业/利益绑定/principal/agent/incentive.
rational-buddhism
当用户问某主张/玄学/灵修该不该信、想建立自己的验证标准、或寻找科学与精神的结合点时调用。 核心理念: 尝试一切, 亲自测试, 保持怀疑, 保留有用的, 舍弃没用的; 不可证伪的主张不进入基本真理; 冥想/接纳等内在技术可保留。 不适用于: 需要信仰安慰的临终/哀伤场景(尊重个体选择)。 Triggers: 该不该信/玄学/灵修/冥想/可证伪/验证/理性/rational buddhism/verify/belief.