Skip to content

Workflow agent

Orchestrate tools + APIs to complete multi-step tasks. The generic default blueprint — use this when none of the more specialized ones fit cleanly.

When to use

You're building a general multi-step automation: ingest something, call APIs, write somewhere, notify someone. Think: Zapier-with-judgment, or a custom internal tool that needs to be reliable.

The loop

decompose → for each step: call tool with retry → check result → log → next

What gets generated

File Role
AGENTS.md Workflow loop + conventions + forbidden lists
SOUL.md "Operational, low-drama execution layer" voice
TOOLS.md Per-tool docs + tool-call logging contract
MEMORY.md Per-task tool log + skill-derived heuristics
SKILLS/decompose-task/ Plan generator with idempotency markers
SKILLS/call-tool-with-retry/ Exponential-backoff + idempotency check
SKILLS/check-result/ Independent observation of expected effect
scripts/test_task.sh Sample task replay
scripts/verify_output.py Plan + tool log + idempotency check

Validators

Validator What it checks
structure Every blueprint-generated file present and parses
tool_log Every plan step that uses a tool has a tool_log entry
idempotent Final plan step is marked idempotent: true (safe to retry)
  • filesystem — read/write project files
  • fetch — HTTP requests
  • shell — run commands (refuses forbidden_commands)
  • github — PRs / issues / actions
  • postgres — read-only DB queries

Memory schemas

  • conversation — run_id + task + plan + tool_log + final_status + idempotent_marker

Hero demo

Airflow + workflow cookbook shows this blueprint on apache/airflow.