Self-improving for Codex skill
Skill setting up Markdown file memory for Codex via AGENTS.md: profile, rules, lessons, errors and capability requests.
- 132 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The skill creates and edits local memory files; AGENTS.md changes only with permission.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/self-improving-for-codexDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skill creates and edits local memory files; AGENTS.md changes only with permission.
This is third-party code. Review the repository files before installing.
What it does
The skill builds long-term memory for Codex from five files with distinct roles. `PROFILE.md` holds stable preferences, `ACTIVE.md` holds rules worth reading before every task, `LEARNINGS.md` collects lessons, `ERRORS.md` records environment failures and fixes, and `FEATURE_REQUESTS.md` tracks missing capabilities. The agent first audits the existing `AGENTS.md`, memory directory and related automations and keeps what is already configured. It then prepares an exact snippet for `AGENTS.md`, rules for promoting entries into `ACTIVE.md` and `PROFILE.md` and, optionally, a nightly job to clean duplicates. The skill must not change `AGENTS.md` without explicit permission.
Who it is for. Codex users who want the agent to remember their preferences and accumulated experience across sessions without background services.
Good fit when
- The agent repeats the same mistake in every new session.
- Stable preferences for work style and communication need a home.
- You want to check what is already set up in the global `AGENTS.md` and memory folder.
- Notes have piled up with duplicates and need regular cleanup.
Not a fit when
- You need memory for OpenClaw with `SOUL` and `HEARTBEAT.md`; the skill does not cover them.
- You want `AGENTS.md` edited automatically with no human review.
- You plan to keep notes in a public repository alongside personal data.
Example request
Audit my current `AGENTS.md` and memory directory, keep everything existing and propose a safe set of changes without editing files.Limitations
The skill targets Codex and relies on `AGENTS.md`. It runs nothing by itself and makes no automatic edits: results depend on how the agent follows instructions. Memory files can hold personal preferences, paths and project experience, so the real `~/.codex/memories` must not be committed to a public repository. Passwords, tokens, keys and whole `.env` files must never go into memory. The README is in Chinese.
How to disable. Delete the `~/.codex/skills/self-improving-for-codex` folder and the snippet added to `AGENTS.md`. Remove the memory files in `~/.codex/memories/` separately if no longer needed.
FAQ
Which files does the skill create?
Five: `PROFILE.md`, `ACTIVE.md`, `LEARNINGS.md`, `ERRORS.md` and `FEATURE_REQUESTS.md`. The recommended place is `~/.codex/memories/`.
Does it edit AGENTS.md itself?
No. By default it gives an exact snippet, and the file may change only with direct user permission.
Is a background service needed?
No. The repository holds only Markdown instructions and skill metadata, with no services or telemetry.
What is the nightly job?
Optional automation that removes duplicates, merges similar notes and prunes stale ones. It works only on collected memory and does not replay the whole conversation history.
How does it differ from the OpenClaw variant?
It uses the native Codex `AGENTS.md`, and OpenClaw mechanisms such as `SOUL` and `HEARTBEAT.md` are not needed.
Discussion
No questions or reviews yet.
Related
SkillsA base AGENTS.md and 12 skills: delegation to Codex, Grok, Devin and Antigravity, planning, copywriting, scoped changes.
A Codex skill collection: phased planning, parallel subagents, documentation access, a browser, 51 hooks and ready agent roles.
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
Self-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
CodexPotter loops Codex with a clean context each round until the code matches the goal, as a replacement for the built-in /goal.
Skills keep project memory in a docs/ai folder: short size-capped files, a one-page handoff and a verifier for the install
README in short
The repository contains `SKILL.md`, Codex metadata and three reference files: the `AGENTS.md` snippet, memory templates and nightly cleanup rules.