Problem
The project has no .claude/commands/ directory, so Claude Code has no project-specific skills (slash commands) to assist contributors with common tasks. Every contributor has to re-describe routine workflows from scratch in each session.
What are Claude Code skills?
Skills are reusable prompt templates stored as Markdown files under .claude/commands/. They are invoked with /skill-name inside a Claude Code session and expand into a full instruction set, giving Claude the exact context it needs to perform a well-defined task correctly every time.
Proposed skills
/run-sim — Launch a simulation
Prompts Claude to pick the right python -m agentevac.simulation.main invocation: ask for scenario, messaging, metrics, GUI/headless, and produce the ready-to-run command.
/new-scenario — Scaffold a new information regime
Guides Claude through adding a new scenario key to agentevac/agents/scenarios.py (SCENARIO_CHOICES, load_scenario_config, apply_scenario_to_signals, filter_menu_for_scenario) and the matching test stub in tests/test_scenarios.py.
/new-agent-param — Add a new psychological parameter to the agent profile
Walks Claude through the full change surface: agent_state.py profile dict, belief_model.py/departure_model.py usage, spawn defaults in simulation/main.py, and the relevant tests.
/sweep — Run a parameter sweep study
Produces the correct agentevac-study invocation by asking for reference metrics path, sigma/delay/trust ranges, scenario, and output directory.
/replay — Replay a previous run
Resolves the run ID from outputs/ and produces the --run-mode replay --run-id <id> command.
/commit — Conventional commit
Summarises staged changes and produces a correctly formatted Conventional Commits message scoped to this project's modules.
Acceptance criteria
Problem
The project has no
.claude/commands/directory, so Claude Code has no project-specific skills (slash commands) to assist contributors with common tasks. Every contributor has to re-describe routine workflows from scratch in each session.What are Claude Code skills?
Skills are reusable prompt templates stored as Markdown files under
.claude/commands/. They are invoked with/skill-nameinside a Claude Code session and expand into a full instruction set, giving Claude the exact context it needs to perform a well-defined task correctly every time.Proposed skills
/run-sim— Launch a simulationPrompts Claude to pick the right
python -m agentevac.simulation.maininvocation: ask for scenario, messaging, metrics, GUI/headless, and produce the ready-to-run command./new-scenario— Scaffold a new information regimeGuides Claude through adding a new scenario key to
agentevac/agents/scenarios.py(SCENARIO_CHOICES,load_scenario_config,apply_scenario_to_signals,filter_menu_for_scenario) and the matching test stub intests/test_scenarios.py./new-agent-param— Add a new psychological parameter to the agent profileWalks Claude through the full change surface:
agent_state.pyprofile dict,belief_model.py/departure_model.pyusage, spawn defaults insimulation/main.py, and the relevant tests./sweep— Run a parameter sweep studyProduces the correct
agentevac-studyinvocation by asking for reference metrics path, sigma/delay/trust ranges, scenario, and output directory./replay— Replay a previous runResolves the run ID from
outputs/and produces the--run-mode replay --run-id <id>command./commit— Conventional commitSummarises staged changes and produces a correctly formatted Conventional Commits message scoped to this project's modules.
Acceptance criteria
.claude/commands/directory created and committedCLAUDE.mdso contributors know they exist