dsh-rerun-task-fs

Task-source provider reading a suite from a directory tree.

<suite>/suite.yml                 name and per-task defaults
<suite>/tasks/<id>/task.yml       one task
<suite>/tasks/<id>/workspace/     its optional seed tree

Every field a task omits comes from the suite's defaults, and the RESOLVED result — never the file text — is what gets hashed, so reorganizing a suite file without changing any effective value leaves task identities intact.

A task's hash covers the seed tree's CONTENT. The same prompt against a different starting tree is a different task, and pooling their results would quietly average two experiments.

- id: rerun-task-fs
  name: 'dsh-rerun-task-fs'
  config: { root: . }

Model Experience

None. This package never contributes to a model request: it runs in the orchestrator process, and the model calls it cares about happen inside a separate attempt subprocess running a different build of DSH entirely.

KV Cache effect

None; it neither assembles nor sends a provider request.

Known Limitations and Deferred Work

• Hashing a seed tree reads every file. A suite with large fixtures pays that on every load, including for rescore.
• Symlinks are followed as their target's content. A seed whose meaning depends on the link itself is not represented faithfully.
• No schema validation beyond hand-written checks. Errors are precise but the rules live in code rather than in a published schema.