v1.0 to v2.3

- description: Run ReacNetGenerator on LAMMPS trajectories to generate reaction networks and reports. Use when the user wants to analyze reactive MD trajectories with ReacNetGenerator (dump/xyz/bond). Handles common LAMMPS dump quirks like x/y/z vs xs/ys/zs scaled coordinates by converting to x/y/z (orthorhombic + triclinic supported via reacnet-md-tools >= 0.1.1), infers atomname order from a LAMMPS data file when available, runs via local reacnetgenerator or via `uvx --from reacnetgenerator`, and writes outputs into `out/<input_basename>/` with logs and a summary.

- compatibility: Requires uv and internet access (uses `uvx --from reacnet-md-tools ...` and `uvx --from reacnetgenerator ...`).

- version: '1.0'

- ## Quick start (agent)

- 1. Ask for or infer:

- - input trajectory path(s)

- - input type: dump | xyz | bond (default dump if LAMMPS "ITEM:" is detected)

- - atom names order for `-a` (infer from LAMMPS data file if present)

- 2. Prefer running the pipeline tool from PyPI (do not hand-compose long commands):

- - Use `reacnet-md-tools` → `rng-pipeline`.

- 3. Avoid interactive prompts in non-interactive runs:

- - Prefer passing `--data path/to/file.data` explicitly.

- - Use `--pick-data` only as an interactive fallback when you have a TTY.

- ## Commands (preferred)

- ### Non-interactive (recommended default)

- ```bash

- uvx --refresh --from 'reacnet-md-tools>=0.1.1' rng-pipeline \

- --input /path/to/trajectory.lammpstrj \

- --type dump \

- --outroot out \

- --data /path/to/system.data \

- --nohmm \

- --stepinterval 10 \

- --maxspecies 50

- ```

- ### Interactive fallback (TTY only)

- If multiple `.data` files exist and you are running interactively, you may use `--pick-data`:

- ```bash

- uvx --refresh --from 'reacnet-md-tools>=0.1.1' rng-pipeline \

- --input /path/to/trajectory.lammpstrj \

- --type dump \

- --outroot out \

- --pick-data

- ```

- ## What this skill does

- - Detect whether a LAMMPS dump uses `x y z` or scaled `xs ys zs` coordinates.

- - If scaled coords are present:

- - Convert to `x y z` using BOX BOUNDS (orthorhombic + triclinic with tilt factors `xy xz yz`).

- - Requires `reacnet-md-tools >= 0.1.1`.

- - Write the converted dump as `out/<basename>/dump_with_xyz.lammpstrj`.

- - Infer `-a/--atomname` order from a LAMMPS data file in the same directory (optional):

- - Parse `Masses` section and map common masses to elements (C, H, O, Cl, N, F, S, P, Br, I, etc.).

- - If inference is ambiguous, stop and ask the user.

- - Run ReacNetGenerator:

- - Prefer `reacnetgenerator` if available on PATH.

- - Otherwise run via `uvx --refresh --from reacnetgenerator reacnetgenerator ...`.

- - Put all outputs in `out/<basename>/`:

- - `run.log` (stdout/stderr)

- - generated `*.html`, `*.svg`, `*.json`, `*.species`, `*.reaction`, etc.

- - `summary.md` listing key outputs and parameters

- ## Version check (agent)

- uvx --refresh --from 'reacnet-md-tools>=0.1.1' python -c "import reacnet_md_tools; print(reacnet_md_tools.__version__)"

- If you still see triclinic conversion errors:

- - ensure your run is resolving `reacnet-md-tools >= 0.1.1` (use `--refresh`), or

- - re-dump trajectories with `x y z` instead of `xs ys zs`.

- ## Default parameters (safe, adjustable)

- - `--nohmm` (default ON unless user explicitly wants HMM)

- - `--stepinterval 10` for a quick first pass (ask user; default to 1 if they want the full trajectory)

- - `--maxspecies 50` (adjustable)

- - `--split 1`

- ## Usage pattern (what to ask the user)

- Ask only what you need:

- - "What is the trajectory file path? (e.g., .lammpstrj/.xyz/.reaxc)"

- - "Which input type should be used: dump/xyz/bond? If unsure, I can auto-detect."

- - "Do you have a LAMMPS data file (.data) for inferring the element order? If yes, give me its path; otherwise provide the atomname list (e.g., C Cl H O)."

- - "What stride/stepinterval should be used? For a quick sanity run, 10 is fine."

- ## Notes / limitations

- - If your dump already contains `x y z`, no conversion is needed.