Writing a solution

A solution is any shell-invokable program. trap runs it once per case as a subprocess and captures its stdout/stderr/files — it never has to import trap or know it exists.

Minimal trap.yaml

Place it next to your solution code, then run tp run from that directory:

cmd: uv run python solution.py
tasks:
  test:
    source: ../task        # local path or git+ URL to the task dir

Inputs and outputs

trap injects TRAP_MANIFEST, a JSON string with two absolute directory paths:

import json, os
from pathlib import Path
m = json.loads(os.environ["TRAP_MANIFEST"])
cfg = json.loads((Path(m["inputs_dir"]) / "config.json").read_text())  # read case inputs
(Path(m["outputs_dir"]) / "result.json").write_text("...")            # write outputs here

• inputs_dir — your case's input files (you know their names).
• outputs_dir — write here; trap owns nothing in it, so it's exactly what you produced.
• stdout/stderr/exit code are captured automatically.

To pipe one input file to stdin, declare it:

cmd: uv run python solution.py
stdin: input.json          # inputs_dir/input.json → stdin
tasks:
  test: { source: ../task }

Full field list: trap.yaml reference. Exact env schema: IO contract.

Cost tracking

If the solution calls an LLM API, trap measures tokens/spend per case with no code changes — it auto-activates when a key env var is set (ANTHROPIC_API_KEY, OPENAI_API_KEY, …) and shows up in the report. Disable with tp run --no-cost. Details: cost tracking.

docs for trap v0.0.14 · source: trapstreet/trap