---

name: crosswords description: "Routes 5x5 mini crossword workflows for Tree of Thoughts sampling, BFS, and DFS search over clue grids." disable-model-invocation: true metadata: disco-role: operating license: MIT

Crosswords

Use this sub-skill when the user wants to solve or inspect the 5x5 mini crossword workflow.

Covers

• task=crosswords runs.
• Naive standard and cot sampling baselines.
• The paper-style BFS path with proposal generation and value checks.
• The notebook-derived DFS search workflow.
• MiniCrosswordsTask / MiniCrosswordsEnv state handling and validation.

Excludes

• Arithmetic puzzle solving; use sub-skills/game24/.
• Text passage generation; use sub-skills/text/.
• New task authoring; see the shared references/workflows.md notes.

Read and run

• references/workflows.md for the board format, DFS helper, and sample command shapes.
• references/troubleshooting.md for invalid actions, row parsing, and DFS pruning problems.
• scripts/run_crosswords.sh for the bundled wrapper that dispatches to sampling or DFS mode.

Quick command shapes

• BFS-style sampling: bash sub-skills/crosswords/scripts/run_crosswords.sh bfs
• Naive standard: bash sub-skills/crosswords/scripts/run_crosswords.sh standard
• Naive CoT: bash sub-skills/crosswords/scripts/run_crosswords.sh cot
• DFS search: bash sub-skills/crosswords/scripts/run_crosswords.sh dfs

Acceptance cues

A future agent should be able to:

• distinguish the board state from the clue/answer state;
• explain why a candidate must be exactly five letters;
• choose DFS pruning only when the status counts still allow a viable path;
• tell whether a failed output is a parsing issue or a crossword-logic issue.