{
"name": "quarry",
"version": "1.0",
"_comment": "Every step retries. dagnats is a durable runner and the engine honours a per-step retry policy, but a workflow that declares none gets exactly one attempt — so a transient infrastructure failure became a red run. Concretely: each step begins by cloning the repo from this forge, and a deploy restarts the forge, so any check-in deployed within ~15s of its own CI run had its clone killed mid-transfer. 22 of the last 30 runs failed that way, every one with 'zit clone: cannot connect'. Three attempts with exponential backoff from 30s covers a restart (~15s) twice over. A genuinely broken test still fails, three times, and still reds the run.",
"steps": [
{ "id": "test", "task": "quarry-test", "retry": { "max_attempts": 3, "strategy": "exponential", "initial_delay": 30000000000, "max_delay": 120000000000 } },
{ "id": "fmt", "task": "quarry-fmt", "retry": { "max_attempts": 3, "strategy": "exponential", "initial_delay": 30000000000, "max_delay": 120000000000 } },
{ "id": "build", "task": "quarry-build", "retry": { "max_attempts": 3, "strategy": "exponential", "initial_delay": 30000000000, "max_delay": 120000000000 } }
]
}