skills4_opp_repl

A composable Agent-Skills pack for opp_repl, the interactive Python REPL + CLI tools + MCP server for OMNeT++ simulations.

Upstream: https://github.com/omnetpp/opp_repl

Each subdirectory is a self-contained skill conforming to the Anthropic Agent Skills format (SKILL.md with YAML frontmatter, optional templates/ / references/ / scripts/). The skills are:

• portable — nothing opp_repl-version-specific in the frontmatter; they work on any client that honours Agent Skills.
• composable — skills reference each other by name under a "See also" heading at the bottom of each file.
• progressively disclosed — SKILL.md files stay short and hand off to bundled references / templates when more context is needed.

Quick links

• WINDSURF.md — step-by-step Windsurf installation (global, workspace, and UI options).
• USAGE_EXAMPLES.md — eight end-to-end prompt/response scenarios showing which skills load when.
• TROUBLESHOOTING.md — symptom-to-skill lookup for common build / run / analysis failures (read this when things break).
• VERIFICATION.md — end-to-end log of the M/M/1/K scenario being run through the skill pack. Proof that the pack works on stock OMNeT++ from zero files to validated numerical results within 3.2% of the analytical reference.
• opp-repl-overview/SKILL.md — the entry-point skill, with a decision tree that maps tasks to the exact siblings to load.
• opp-repl-ai-workflows/SKILL.md — cookbook for agents driving opp_repl via MCP or shell.

What's new since initial release

• 3 new skills targeting the hardest friction points real AI agents hit when using opp_repl:
  • opp-repl-project-scaffolding — create a new OMNeT++ project from zero (.opp, .oppbuildspec, .nedfolders, NED+C++ skeletons, correct make_makefiles() + build_project() sequence).
  • opp-repl-result-analysis — read .sca/.vec with either opp_scavetool or the omnetpp.scave.results Python API; bundles scripts/parse_scalars.py that works both ways.
  • opp-repl-troubleshooting — decode every common opaque error (exit 127, "Building X failed", "Class not found", ...) to its cause and fix.
• Hardened existing skills with the pitfalls these real users hit — opp-repl-running-simulations, opp-repl-tasks-and-results, opp-repl-opp-files all now explicitly warn about tr.subprocess_result.stderr being the source of truth on ERRORs, the executables=[...] requirement, and the Makefile pre-generation step.

How to load

For Claude (Claude Desktop, Claude Code, Claude API with

agent-skills enabled)

Place the skills in one of these locations (or symlink from here):

• Global: ~/.claude/skills/
• Project: .claude/skills/ in your workspace

The skill at opp-repl-overview is the entry point. Load it first; it lists every sibling skill and when to use each.

For Windsurf (Codeium Cascade)

See WINDSURF.md for the full guide. TL;DR:

# Global install for every Windsurf workspace
git clone https://github.com/tabgab/opp_repl-skill.git ~/opp_repl-skill
mkdir -p ~/.codeium/windsurf/skills
for dir in ~/opp_repl-skill/opp-repl-*; do
    ln -sfn "$dir" ~/.codeium/windsurf/skills/
done

Windsurf auto-discovers all 26 skills; @-mention them in Cascade.

For other agent frameworks

The SKILL.md front-matter fields (name, description) are the ones Anthropic standardised; most other frameworks read the body as plain markdown. You can treat the whole pack as a library of runbooks — feed the description lines into the agent's system prompt and let it pull full bodies on demand.

Skill index

Foundations

Skill	What it covers
opp-repl-overview	Entry point + decision tree + feature-to-skill map
opp-repl-installation	pip, extras, setenv, Ubuntu PEP-668
opp-repl-concepts	Workspace / Project / Config / Task / Result model
opp-repl-opp-files	.opp descriptor format (+ 7 ready templates)
opp-repl-project-scaffolding	New project from zero (+ 7 file templates)

Execution & control

Skill	What it covers
opp-repl-repl-usage	IPython launch, namespace, autoreload
opp-repl-cli-tools	Shell wrappers for CI
opp-repl-running-simulations	build_project + run_simulations
opp-repl-filtering	Shared filter vocabulary
opp-repl-tasks-and-results	Task hierarchy, rerun, drill-down
opp-repl-troubleshooting	Symptom -> cause -> fix catalogue

Regression tests

Skill	What it covers
opp-repl-smoke-tests	Start / terminate checks
opp-repl-fingerprint-tests	Event-level behavioural regression
opp-repl-statistical-tests	Scalar-level regression
opp-repl-speed-tests	CPU-instruction-count regression
opp-repl-chart-tests	Image-based regression
opp-repl-sanitizer-tests	ASAN / UBSan runs
opp-repl-feature-and-release-tests	Aggregate suites

Analysis

Skill	What it covers
opp-repl-result-analysis	Read .sca/.vec (+ bundled parse_scalars.py)
opp-repl-comparing-simulations	Cross-project / cross-commit comparison
opp-repl-parameter-optimization	scipy Nelder-Mead tuning
opp-repl-coverage-reports	LLVM line coverage
opp-repl-profiling	perf + Hotspot

Infrastructure

Skill	What it covers
opp-repl-overlay-builds	fuse-overlayfs layers
opp-repl-ssh-cluster	Dask over SSH
opp-repl-github-actions	dispatch_workflow / dispatch_all_workflows
opp-repl-opp-env-integration	Routing through opp_env run

AI integration

Skill	What it covers
opp-repl-mcp-server	MCP endpoint (execute_python, opp-repl:// URIs)
opp-repl-ai-workflows	Seven end-to-end recipes for agents

Dependency hints

Almost every skill has a "See also" section. Key paths:

• New to opp_repl → opp-repl-overview → opp-repl-installation → opp-repl-concepts → opp-repl-opp-files.
• Running sims → opp-repl-running-simulations (+ opp-repl-filtering + opp-repl-tasks-and-results).
• Test-authoring → pick the specific opp-repl-*-tests skill (all depend on opp-repl-running-simulations and opp-repl-tasks-and-results).
• Agent integration → opp-repl-mcp-server + opp-repl-ai-workflows (+ whichever task-specific skills the recipe uses).

Authoring notes

• Each SKILL.md deliberately stays under ~200 lines. When more detail is needed, linked files under templates/ or references/ carry the weight.
• opp-repl-opp-files/templates/ contains 7 canonical .opp descriptors ready to copy-paste.
• No scripts are bundled — opp_repl's own shell wrappers and the MCP execute_python tool cover programmatic use. If you need project-local scripts, add them under scripts/ in the relevant skill directory.
• Upstream doc URLs in each skill target the main branch on GitHub, so they stay current as opp_repl evolves.

License

Follows the upstream opp_repl project (LGPL-3.0-or-later). The skill texts themselves are freely reusable under the same terms.