bzfs Experimental Script Generator

Core Outcome

Generate reviewable snapshot management scripts, not command transcripts. Keep scripts minimal, idiomatic, and explicit about safety knobs and assumptions. Enforce safety-first dry-run defaults for mutating flows (bzfs --dryrun; bzfs_jobrunner --dryrun (without --jobrunner-dryrun) and do not execute non-read-only CLIs. For bzfs_jobrunner outputs, fit all idiomatic patterns from bzfs_testbed/bzfs_job_testbed.py (both syntactic and semantic), including action routing and dict construction/format/passing.

Hard Safety Rules

1. No TDD by default: Do not create unit tests or integregation tests unless the User explicitly requests it.
2. Generate scripts only. Never execute generated scripts.
3. Never execute CLI commands while using this skill, except optional read-only zfs list ... and zpool list ... commands for pool/dataset/snapshot discovery.
4. Classify create/replicate/prune/rollback/restore flows as state-changing.
5. For state-changing flows, default scripts to dry-run:
   • bzfs ... --dryrun (default send; use --dryrun=recv only if asked).
   • bzfs_jobrunner ... --dryrun.
6. Use DRYRUN (Bash) / dryrun (Python) as the dry-run toggle variable. Keep safe defaults enabled (DRYRUN=1, dryrun=True).
7. Keep destructive flags (--force*, --delete-*) opt-in, documented, and disabled by default.
8. Prefer bzfs and bzfs_jobrunner over direct mutating zfs commands.
9. Keep scope limited to bzfs and bzfs_jobrunner snapshot management workflows; decline general ZFS administration or non-bzfs tooling requests.
10. Do not favor Bash over Python; provide both Bash and Python script outputs unless the user asks for one language.
11. For bzfs_jobrunner scripts, mirror the action set from bzfs_testbed/bzfs_job_testbed.py.
12. For bzfs_jobrunner scripts, mirror dict handling from bzfs_testbed/bzfs_job_testbed.py: build native dict/list objects in Python and pass them via --flag={value} formatting; in Bash, keep explicit quoted dict literals.
13. Distill and apply bzfs_job_testbed.py idioms at both syntactic and semantic levels.

Step by Step Reasoning Workflow:

• Think systematically through what's been asked of you, break down the problem, work through it step by step, and reason deeply before responding.

Script Generation Workflow

1. Gather required inputs:

   • datasets, recursion scope, hostnames and their roles, replication schedule, retention plans, monitoring plans, log path.
   • If inputs are missing, ask the User corresponding questions via the request_user_input tool or similar, if available.

2. Classify the request:

   • read_only: snapshot list compare, snapshot monitoring, inventory/listing.
   • state_changing: snapshot creation, replication/backup, snapshot pruning, restore.

3. Choose the CLI:

• Prefer direct bzfs for:

  • adhoc/manual snapshot creation,
  • adhoc/manual replication or restore,
  • adhoc/manual snapshot pruning,
  • adhoc/manual monitoring of snapshots,
  • snapshot list comparison.

• Prefer bzfs_jobrunner for:

  • periodic or automatic workflows,
  • multi-host or fleet-wide orchestration,
  • one shared jobconfig that drives create snapshot, replicate, prune, and monitor actions,
  • cron/systemd style wrappers around a shared Python config.

4. bzfs_jobrunner Host filtering:

   • This is a complex area. Think deeply. Source-side actions usually scope with --src-host, destination-side actions with --dst-host. But do not follow this template blindly; depending on which specific source/destination host subsets the workflow is actually intended for (for example third-party-host orchestration, testing one src -> dst route, each destination host pulling independently or each source host pushing independently, or high-frequency pair jobs), add or omit --src-host and/or --dst-host filters.
   • You have plenty of time; go slow and make sure everything is correct.

5. Generate code:

   • For bzfs_jobrunner NEVER merge any job actions (for example --create-src-snapshots, --replicate, --prune-src-snapshots, --prune-src-bookmarks, --prune-dst-snapshots, --monitor-src-snapshots, --monitor-dst-snapshots, --dryrun, --verbose) into the underlying jobconfig script code; instead keep them in a separate launcher bash script so different actions can run with the same jobconfig configuration settings.
   • Prefer plan-based convenience flags such as --include-snapshot-plan, --create-src-snapshots-plan, and --delete-dst-snapshots-except-plan over hand-written --include-snapshot-times-and-ranks chains when standard secondly/minutely/hourly/daily/weekly/monthly/yearly policies are sufficient.
   • Keep code idiomatic and minimal. Rigorously apply the KISS principle: Keep it simple stupid. Leave out all fluff and unnecessary indirections/abstractions. Do not add any CLI flag. Do not add any environment variable beyond DRYRUN.
   • Bash: #!/usr/bin/env bash + set -euo pipefail.
   • Python: #!/usr/bin/env python3 + subprocess.run([...], check=True).
   • Build commands as arrays/lists, not unsafe concatenated shell strings.
   • Model command shape after bzfs_testbed/bzfs_job_testbed.py action conventions.
   • For bzfs_jobrunner dict args, follow bzfs_job_testbed.py style: Python dict/list objects rendered into --src-hosts=..., --dst-hosts=..., --src-snapshot-plan=..., etc.
   • Return paired Bash and Python variants by default.

6. Add operator gates:

   • A single dry-run switch defaulting to enabled.
   • Printed command preview before execution.
   • For Python previews, use shlex.join(cmd).
   • For bzfs_jobrunner scripts, include --dryrun when DRYRUN=1; do not include --jobrunner-dryrun unless explicitly requested otherwise.

7. Return three artifacts:

   • scripts (Bash + Python, unless one language is explicitly requested),
   • what it does,
   • exactly what must be edited before first run.

Output Contract

• Prefer single-purpose scripts.
• Provide Bash and Python variants together unless the user requests otherwise.
• Keep comments short and safety-focused.
• Keep naming aligned with bzfs docs (org, target, *_plan).
• Include cron/systemd-ready command lines for periodic tasks.
• For "restore", default to restore rehearsal into non-production target datasets unless explicitly instructed otherwise.

References

• Read references/safety_and_semantics.md before generating state-changing workflows.
• Use references/task_recipes.md for canonical task-to-command patterns.
• Start from references/script_templates.md for minimal Bash/Python script skeletons.
• Check references/simulation_examples.md for calibrated examples and quality bar.
• Match bzfs_jobrunner structure to bzfs_testbed/bzfs_job_testbed.py when generating periodic orchestration scripts.