菜单
Record or reinforce an Engram lesson — the self-learning loop. Use when the user corrects a mistake, when an action failed in a generalizable way, when the user confirms an approach as right, or when you just applied a previously recalled lesson.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/behl1anmol/Engram --skill engram-lessons命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
| name | engram-lessons |
| description | Record or reinforce an Engram lesson — the self-learning loop. Use when the user corrects a mistake, when an action failed in a generalizable way, when the user confirms an approach as right, or when you just applied a previously recalled lesson. |
Lessons make you measurably better with this user over time, in any agent. CLI path is in the session-start conventions block.
Record when: the user corrected you; your action failed and a general check would have prevented it; the user confirmed "always do it this way".
Quality bar — apply before writing: would this lesson change behavior in a future session? One-off trivia (flaky network, typo) is not a lesson.
Body must follow this shape:
**Mistake:** <what went wrong, concrete>
**Why it happened:** <the assumption or missing check>
**How to apply:** <the behavior-changing rule for next time>
echo "<body>" | ENGRAM_AGENT=claude-code python3 "<engram_py>" add \
--type lesson --name kebab-case-slug \
--description "Discriminating one-liner recall can match" --tags ...
For a confirmed-good approach, same shape with Mistake: replaced by Approach:.
When a recalled lesson actually changed what you did this session:
python3 "<engram_py>" lesson applied <name>
This bumps times_applied (ranks it higher) and renews its expiry. Unapplied lessons lapse naturally — that is the decay model, don't fight it.
基于 SOC 职业分类
