AAR Loop skill
A skill that reviews an agent session with four after-action questions and saves lessons to LESSONS.md so mistakes stop repeating.
- 106 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The skill edits your skill and rule files itself
- Lessons in `LESSONS.md` enter the context of every later session
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/aar-loopDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The skill edits your skill and rule files itself
- Lessons in `LESSONS.md` enter the context of every later session
This is third-party code. Review the repository files before installing.
What it does
AAR Loop brings the after-action review method used by the US Army to agent work. The `/aar-loop` command or a request to run an AAR starts a review by four questions: what was supposed to happen, what actually happened, why they differ, what to do the same or differently next time. The skill picks findings worth fixing and applies the fix straight into your skills or rules files so the correction is already in place. Lessons are appended to a persistent `LESSONS.md` file that the next session reads at start. It is a manual version of Reflexion: you trigger reflection once per session and the file serves as the memory buffer. A helper script `scripts/append_lesson.py` handles writing a lesson.
Who it is for. Developers who work with an agent a lot and are tired of seeing the same slips in every new session.
Good fit when
- After a deployment or a long task you need to record what went wrong.
- The agent repeats a mistake that was already fixed in a previous session.
- Project rules should grow after each piece of work.
- A team wants a shared lessons file loaded at session start.
Not a fit when
- Automatic reflection without your involvement is needed: the run is manual.
- Project rule files must not be changed by the agent without separate review.
- The task is one-off and will not repeat.
Example request
/aar-loop for the deployment session, save the lessons and fix the skill that led to the mistake.Limitations
The skill only works in the session it reviews and remembers what is in context: details lost to compaction cannot be reviewed. Edits go straight into your skills and rules, so review changes in Git before committing. The `LESSONS.md` file grows and needs thinning from time to time, or it will take much context. The method is documented in English. The repository was last updated on 26 July 2026 and uses the MIT licence.
How to disable. Delete the `aar-loop` folder from the agent skills directory. The `LESSONS.md` file and applied edits stay and must be removed by hand.
FAQ
Which four questions does the review ask?
What was supposed to happen, what happened, why there is a difference and what to do the same or differently next time.
Where are lessons saved?
To the persistent `LESSONS.md` file, which the next session loads at start.
Does the skill change my rules?
Yes. For found problems it applies corrections directly into your skills or rule files, so review the edits.
How to start a review?
With the `/aar-loop` command or a phrase like "run an aar" after finishing a task.
How does it differ from Reflexion?
It is a manual version: you trigger reflection once per session and memory sits in `LESSONS.md` with no automatic loop.
Discussion
No questions or reviews yet.
Related
SkillsSelf-Improving Agent
self-improving-agent
Skill for OpenClaw: the agent logs errors, corrections and missing features to .learnings files and promotes verified lessons into workspace memory
Self-learning
self-learning-skills
A meta-skill that saves a hard-won working procedure as a new skill or rule, so the next session starts already knowing it
The xskill CLI gathers sanitized session trajectories from Claude Code, Codex and OpenCode and grows a shared skill library.
The agents-meet-rl skill for Claude Code: a handbook on RL training of AI agents, failure diagnosis and framework choice.
A CLI that reads local Claude Code, Codex and Hermes sessions, finds repeated failures and stages fixes for manual review
Скилл ведет читателя по научной статье в шесть шагов, присылая нужные рисунки: от мотивации до разбора примера по коду.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| AAR Loop (this card) | no | 106 | ||
| claude-reflect | no | 1,947 | catches your corrections in dialogue and, once confirmed, saves them to CLAUDE.md and AGENTS.md; here a whole-session review by four questions | |
| Claude Self-Reflect | no | not required | 230 | local conversation memory with search over past sessions; here a short lessons file and rule fixes |
README in short
The README explains the method origin in the US Army and relates it to Reflexion, and installation comes down to giving the repository address to the agent.