This repository provides a Codex skill that installs and uses a pinned copy of AUTO-07p inside a project without modifying global shell configuration.
The intended interface is conversation with Codex. Most users should not need to run Python helpers, source AUTO environment files, or interpret raw AUTO output by hand.
AUTO-07p itself is not distributed by this repository. The installer retrieves
it from its official repository and records the exact source commit and local
toolchain in .auto/install-manifest.json.
It gives Codex a repeatable workflow to:
- inspect the local compiler and Python environment;
- install a pinned AUTO-07p source revision inside the current project;
- verify the installation with a real continuation calculation;
- run equilibrium and periodic-orbit continuation; and
- render branch data as PNG, SVG, tidy CSV, and JSON.
You do not need to clone this repository to use the skill.
-
Create an empty project folder and open it as a new project in Codex.
-
Give Codex this repository URL and ask it to follow the repository instructions:
Read this repository and install the Codex skill it provides: https://github.com/dueyama/auto-07p-codex
Codex should inspect the repository and install
skills/analyze-bifurcations-with-auto/. If it asks for the exact skill directory, provide:https://github.com/dueyama/auto-07p-codex/tree/main/skills/analyze-bifurcations-with-auto
-
After Codex reports that the skill was installed, send a new message:
Use the
analyze-bifurcations-with-autoskill. Check the dependencies, install AUTO-07p inside this project, and verify the installation.Codex will inspect the toolchain first and ask before installing a missing system package such as Homebrew
gcc. -
Run the bundled demonstration:
Run AUTO-07p's
abdemo and show me the bifurcation diagram.
The installed skill becomes available on the next Codex turn. It is copied to
$CODEX_HOME/skills/analyze-bifurcations-with-auto/; AUTO itself is then built
separately inside the current project's .auto/ directory.
Install the skill only once per Codex environment. For later analysis projects,
skip step 2 and ask Codex to use the already installed skill.
$CODEX_HOME/skills/
└── analyze-bifurcations-with-auto/ # reusable Codex skill
your-analysis-project/
├── .auto/ # project-local AUTO-07p, ignored by Git
└── runs/ # local calculations and plots
The skill does not edit .zshrc, .bashrc, or other shell startup files.
Each project can therefore use and verify its own pinned AUTO installation.
After the demo works, give Codex your equations and describe the parameter you want to vary. Initial values and a continuation range help when they are known, but you can also ask Codex to propose a bounded first run.
For example:
Here are my equations. Vary
mu, find the bifurcations, and show me the bifurcation diagram.
Codex can propose AUTO settings, explain what it found, continue periodic solutions when appropriate, and keep the model, logs, branch files, plots, and machine-readable exports together in a dedicated run directory.
Every accepted analysis includes a bifurcation diagram displayed in Codex.
Stable segments are solid, unstable segments are dashed, equilibrium and
periodic branches are distinguished, and special points such as HB, LP,
BP, PD, and TR are labeled when detected. Codex also saves PNG and SVG
figures plus tidy CSV and JSON results for later use.
examples/custom-hopf/ is an original, analytically solvable example rather
than an AUTO demo. It continues the supercritical Hopf normal form
x' = mu*x - y - x*(x^2 + y^2)
y' = x + mu*y - y*(x^2 + y^2)
from its equilibrium branch, detects the Hopf point at mu = 0, and continues
the stable periodic branch. Theory predicts radius sqrt(mu) and period
2*pi for mu > 0.
In the checked result, AUTO located the Hopf point at
mu = -3.0e-9. The maximum errors were 3.0e-9 for radius^2 - mu
and 2.1e-11 for the period relative to 2*pi.
The repository-local skill is
skills/analyze-bifurcations-with-auto/.
The root AGENTS.md tells Codex when to use it while developing this
repository. End users install the skill directory into their Codex skills
directory using the GitHub URL above.
The current implementation has been tested only on Apple silicon macOS.
Linux is expected to work with Python 3, Git, GNU Make, a C compiler, and
gfortran, but it has not yet been verified by this project. Native Windows
has not been tested or designed as a supported environment; WSL is the
recommended route and is also currently unverified.
On macOS, Homebrew supplies gfortran through its gcc formula. Codex checks
for it and asks for approval before installing a missing system package.
The installer uses the first gfortran on PATH, performs a compile-and-link
probe before building AUTO, and records the resolved compiler path and version
in .auto/install-manifest.json.
Verified on Apple silicon macOS:
| Fortran compiler | AUTO build | Official ab verification |
Original Hopf example and plots |
|---|---|---|---|
| Homebrew GCC 16.1.0 | Pass | Pass | Pass |
| Standalone GCC 12.2.0 | Pass | Pass | Pass |
Expected platform status:
| Platform | Status |
|---|---|
| Apple silicon macOS | Verified |
| Intel macOS | Expected to work; not tested |
| Linux | Expected to work; not tested |
| Windows through WSL | Expected to work; not tested |
| Native Windows | Not supported or tested |
AUTO's optional interactive plotting tools can require additional Tk/X11 packages. Their absence does not prevent continuation calculations or non-interactive plotting from saved branch data.
The command-line helpers are implementation and development tools, not the primary user interface. Contributors working in a checkout can run:
python3 scripts/autoctl.py doctor
python3 scripts/autoctl.py install
python3 scripts/autoctl.py verifyverify runs AUTO's official ab demo and requires actual Hopf and limit
points in the generated output. scripts/plot_bifurcation.py is the
non-interactive renderer used by the skill. The bin/auto-codex wrapper loads
AUTO's environment for one child process without changing shell startup files.
The integration code in this repository is MIT-licensed. AUTO-07p is a separate upstream project with its own authorship and distribution terms.
This is a focused instruction update, not a change to the AUTO numerical engine or a requirement to use a particular Codex model.
- Keep conversation as the interface: a request such as "run the ab demo and show its bifurcation diagram" uses the demo defaults without unnecessary clarification. Ask when missing information changes the scientific question.
- Resolve helper scripts from the installed skill directory and pass explicit project and AUTO paths, so a separate analysis project does not need a clone of this repository.
- Clarify that plots should be displayed in Codex, not merely linked, and that NumPy and Matplotlib must be available to the plotting interpreter.
- Check for branches omitted because axis columns are missing, and avoid treating missing stability information as evidence of instability.
- Separate installation verification from saved-output inspection; do not rebuild AUTO or run an unrelated smoke test just to inspect existing data.
- Retain project-local defaults while allowing user-selected output locations; no machine-specific storage paths are prescribed.
Validation for this update: skill structure validation and all 12 existing unit tests passed. AUTO builds and numerical continuation runs were not repeated; the platform results above refer to earlier testing.