Menu
Generate UI-viewable replay artifacts from a trained RL checkpoint. Use when user asks to generate replays, view agent behavior, or step through games.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Check Hetzner VM sweep status: games completed, errors/bugs with seeds, max fame leaderboard. Use when user asks about VM status, sweep progress, bugs, or fame.
View GitHub project board status. Shows issues by status (In Progress, Up Next, Backlog) with priority labels. Use when user asks 'what's the status', 'show me the board', or 'what's in progress'.
Complete work on current issue, create PR, and update project board. Use when implementation is complete and ready for review.
Link GitHub issues as sub-issues or blockers. Use when organizing epics or marking dependencies between issues.
Recommend what issue to work on next based on priority, complexity, and dependencies. Use when user asks 'what should I work on', 'what's next', or 'pick something for me'.
Start working on a GitHub issue. Use when user says 'work on #123', 'start issue 42', or 'pick up the next task'.
| name | replay |
| description | Generate UI-viewable replay artifacts from a trained RL checkpoint. Use when user asks to generate replays, view agent behavior, or step through games. |
| user-invocable | true |
| argument-hint | [run-name or checkpoint-path] [--seeds 1-5] [--no-oracle] |
Generate artifact JSON files from a trained RL checkpoint that can be viewed in the client replay viewer.
cd packages/python-sdk && source .venv/bin/activate
python scripts/generate_replays.py <checkpoint> --artifact --combat-oracle --commerce-oracle --seeds 1-5 --output-dir sim-artifacts/<run-name>
If run name provided (e.g., /replay explore-fix-v1):
# Find the latest checkpoint in the run
ls packages/python-sdk/training/runs/$1/checkpoints/ | sort | tail -1
Checkpoint path: training/runs/$1/checkpoints/<latest>.pt
If full checkpoint path provided: Use it directly.
If no argument: List available runs and ask which one.
ls packages/python-sdk/training/runs/
--seeds N-M — Seed range (default: 1-5)--no-oracle — Skip combat oracle (default: oracle enabled)IMPORTANT: Output directory must use sim-artifacts/ (hyphen, not underscore) to match the Vite server config.
cd packages/python-sdk && source .venv/bin/activate
python scripts/generate_replays.py \
training/runs/<run-name>/checkpoints/<checkpoint>.pt \
--artifact \
--combat-oracle \
--seeds 1-5 \
--output-dir sim-artifacts/<run-name>
Omit --combat-oracle if --no-oracle was specified.
Show a table of results:
Tell the user:
packages/python-sdk/sim-artifacts/<run-name>/bun run dev:rust, open replay viewer, and select from the dropdown (or drag & drop any JSON file)set_rl_mode(True) is already called by the script — this disables undo actionspackages/python-sdk/sim-artifacts/ via the /__artifacts endpointsim-artifacts/ directory is gitignored