Permanent Underclass

Terminal game about surviving AI acceleration.

npx --yes permanent-underclass@latest

Screenshots

Title screen:

Character select:

Live turn UI:

What This Is

You pick a character, live through a collapsing job market, and try to make it to an ending without going broke or burning out.

• Short, replayable runs (simple: 12 turns, standard: 32 turns)
• Character-specific event pools with different pressure profiles
• No single correct path; tradeoffs are the point
• You are not locked to the three presented choices.

Press 4 and type your own move:

• start a podcast
• cry in the bathroom
• go on a meditation retreat
• short nvidia

With LLM mode enabled, your local codex or claude CLI resolves that action into real in-game consequences.

Play

# stable
npx --yes permanent-underclass@latest

# beta
npx --yes permanent-underclass@beta

Local Quick Start

cargo run

CLI help:

cargo run -- --help

Controls

• Ctrl+C: quit
• Enter: confirm / continue
• Arrow keys: navigate cards, menus, settings
• 1 / 2 / 3: choose a card
• 4: enter freeform mode (Enter submit, Esc cancel)
• s: manage subscription (Free / Pro / Ultra / Giga)
• m: open in-run settings
• Esc: back/cancel

Tempo Modes

• simple: 12 turns (default)
• standard: 32 turns

Use --tempo simple|standard in --cli, --headless, and --llm-bench modes.

LLM Mode (Optional)

# auto provider (default): prefers Claude if installed, else Codex
cargo run -- --llm

# Codex
codex login
cargo run -- --llm --llm-provider codex

# Claude
claude auth
cargo run -- --llm --llm-provider claude

Optional flags: --llm-model, --llm-reasoning low|medium|high|xhigh, --llm-profile.

Advanced Local Modes

CLI Interactive (No TUI)

cargo run -- --cli --character junior-dev --seed 1
cargo run -- --cli --character junior-dev --seed 1 --tempo standard

Headless Autoplay

cargo run -- --headless --character junior-dev --seed 1
cargo run -- --headless --character junior-dev --seed 1 --tempo standard

Headless With LLM + Snapshot Save/Resume

cargo run -- --headless --llm --save-file run.snapshot.json --character junior-dev --seed 1
cargo run -- --headless --llm --resume-file run.snapshot.json

LLM Benchmark (No TUI)

cargo run -- --llm-bench --llm-provider auto --llm-bench-iters 1
cargo run -- --llm-bench --llm-provider auto --llm-bench-iters 1 --tempo standard

Balance Simulation

cargo run --bin balance_sim -- --seeds 20
cargo run --bin balance_sim -- --seeds 400 --character all --tempo standard --output-jsonl data/sim/ending_sweep_latest.jsonl

Command Center

npm run dev

Opens http://localhost:5173/command-center to browse events, endings, characters, flags, and sprites. If data/sim/ending_sweep_latest.jsonl exists, the Sim tab renders ending/policy/character outcome summaries. Source lives in apps/command-center/.

Tests and Security Checks

# full release gate (single pass/fail summary)
./scripts/release_preflight.sh

# individual checks
cargo test --lib
cargo test --test sim
# one-time: cargo install cargo-audit
cargo audit
# one-time: cargo install cargo-deny
cargo deny check advisories
# one-time: brew install gitleaks  # or install from https://github.com/gitleaks/gitleaks
gitleaks detect --source . --no-git --redact --exit-code 1

Licensing

• Code license: LICENSE (MIT)
• Non-code asset licensing and attribution status: ASSET_LICENSES.md