Astra Skill Optimizer plugin
Codex plugins: a Skill audit for GPT-6 Astra and Jarvis, an assistant that restores a project's goal, stage and blocker
- 125 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Reads project files and Codex logs; edits run only after an explicit request and stay within its scope.
Install
- Works in
Manual install
codex plugin marketplace add "$PWD"Run in the clone directory after git clone, followed by the two codex plugin add calls from the README.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Reads project files and Codex logs; edits run only after an explicit request and stay within its scope.
This is third-party code. Review the repository files before installing.
What it does
Astra Skill Optimizer consists of two Codex plugins. The first, gpt6-astra-skill-optimizer, takes one project and its related skills, finds outdated constraints and proposes the smallest verifiable change. The second, requirement-ledger, supplies the Jarvis workflow: a project manager that restores the goal, current stage, evidence freshness, blocker and one next step from an ordinary phrase. Jarvis separates verified facts, inferences and unknowns. It picks one of six scenes: project setup, progress review, requirement change, blocker diagnosis, version acceptance and handoff. A quick audit changes no files, and a fix runs only after a clear request.
Who it is for. Codex users whose long projects lose decisions inside chat and whose older skills stop working when model behaviour changes.
Good fit when
- A project runs for weeks in Codex and you must quickly recall its goal, stage and blocker.
- A client changed a requirement and you need to see which assumptions are now invalid.
- Before a release you want an acceptance review with separate statuses: passed, failed, skipped, unknown.
- Project skills were written for an earlier model and you want them checked for outdated constraints.
Not a fit when
- Work runs in Claude Code: the plugin is built only for Codex and uses no Claude hooks.
- You want ready scheduled report automation: daily and weekly reports run on demand.
- You need a full home directory scan: the tool works on one selected project.
Example request
Hi Jarvis, take over this project: recover its goal, current stage, blocker and one next step. Change nothing.Limitations
The plugin is a thin skills layer over a Python core, and complete GPT-6 Astra adaptation, by the author's own words, still needs feedback from real projects. The legacy Python package and CLI remain as compatibility entry points during the rename. A quick audit in Codex does not pass the source-bound verification chain, and a reproducible result needs the second level with a window, timezone and exact files. Installation runs from a cloned repository. The latest commit is from September 2026.
How to disable. Remove both plugins with `codex plugin remove` or drop the local requirement-ledger-local marketplace from Codex settings. The cloned folder can be deleted by hand.
FAQ
What is the difference between the two plugins?
The first checks project skills for outdated constraints and proposes one change. The second drives the Jarvis project report: goal, stage, blocker and next step.
Does a quick audit change files?
No. An audit and project takeover only read. A fix runs after an explicit request with a bounded scope.
What if Codex does not pick the skill itself?
Retry once explicitly, for example `$requirement-ledger-workflow Hi Jarvis, take over this selected project.`
Which scenes does Jarvis know?
Six: project setup, progress review, requirement change, blocker diagnosis, version acceptance and handoff. It picks one from an ordinary phrase.
Does it work in Claude Code?
No. The author states it is not a Claude Code port and version 1 depends on no Claude hook or configuration.
Discussion
No questions or reviews yet.
Related
PluginsA personal library of 25 skills: posters and storyboards, Google-style code review, lab report reading and scenario simulations
Eight Claude Code skills for Obsidian: Granola notes, NotebookLM imports, TaskNotes tasks, recall of past sessions, chat export
SkillRoster scans skills across all your agents, finds duplicates and broken links and builds a per-agent roster with undo.
AGT Skill Pack gives an agent 24 free skills: review by another model, idea screening, UI checks and decision tracking
18 portable skills for work where AI is weakest: messy voice input, message tone, small trend tests and safe shipping.
A desktop app and skm CLI: one copy of each skill in a central hub, symlinked into every AI tool
README in short
Two review levels are described: a quick Codex audit with no source binding and an evidence-bound review with a reproducible result. The second suits results that get handed off.