DSCC
home / cookbook

Cookbook

Runnable end-to-end scenarios. Each case is self-contained: a PROMPT.md you can pipe into DSCC, a README.md that explains setup, expected tool calls, and how to verify the result.

Every case picks one capability, keeps the prompt short, and says honestly “the model should typically do X” rather than claiming guaranteed behaviour.

Cases

# Folder Hook Capability demonstrated
01 01-code-review Review a Rust file in read-only mode and report planted bugs. grep_search + read_file + StructuredOutput, ReadOnly permission
02 02-test-generation Generate table-driven unit tests for an untested helper. read_file → write_file, WorkspaceWrite permission
03 03-refactor-rename Rename a function across 4 files using grep + edit. grep_search → edit_file (replace_all), WorkspaceWrite
04 04-web-research Summarise an online API reference into a table. WebSearch + WebFetch, ReadOnly
05 05-data-analysis Compute statistics from a 25k-row CSV and write a report. bash + file I/O, DangerFullAccess

How to read a case

  1. Open the case’s README.md — it has context, setup, the exact run command, expected tool sequence, and verification.
  2. Open PROMPT.md — copy-paste it into dscc prompt "$(cat PROMPT.md)" or pipe it in.
  3. Compare the model’s actual output to the verification checks.