Swamp Club: Prompting Agents for Repeatable Workflows

What Swamp Changes

Swamp Club targets a different problem than chat prompts: repeatable operational workflows an agent can build, review, and run again. Work lives in .swamp/ as typed models, YAML definitions, workflow DAGs, and versioned run data—not only in conversation history.

The Human Objective Prompt

Describe outcomes, not CLI steps. Let the agent discover Swamp primitives via bundled skills.

"Goal: inventory all EC2 instances missing a cost-center tag and produce a weekly report. Constraints: read-only first; no destructive API calls until I approve YAML. Use Swamp in this repo (swamp repo init if needed). Deliverables: model definitions, workflow DAG, sample run output, and vault plan for credentials."

Workflow Composition Prompt

Ask for the Swamp-shaped pipeline explicitly.

"1) Search existing Swamp models. 2) Create definitions with typed arguments. 3) Compose a workflow with parallel jobs where safe; chain outputs with data expressions. 4) Reference secrets via vault expressions—never paste credentials in chat or YAML. 5) Show me all generated files for review before production runs."

Verification Prompt

"Before swamp workflow run, list assumptions, failure modes, and rollback plan. After run, summarize versioned data artifacts and what to query next time."

Editor Integration

Swamp ships skills for Cursor, Codex, Claude Code, and others. Mention your editor so the agent loads the correct rules directory and CLI flags (--tool cursor, etc.).