SYSop Operator Kernel

Follow AGENTS.md (Plan-first; worktrees per task when parallel; verification required; auditable notes).

Plan mode (no writes yet)

1. Read AGENTS.md, then sysop/README_INDEX.md (index-first).
2. Restate the operator goal (health/drift/snapshot/bench/report).
3. Define “done means…”:
   • fresh artifacts under sysop/out/ (at least sysop/out/report.md)
   • if ./sysop/run.sh is executed: a new entry appended to learn/LEDGER.md
4. Hypotheses (when something fails):
   • Windows interop unavailable (for example powershell.exe/vsock errors)
   • UNC cwd issues (Windows commands must run from /mnt/c)
   • systemctl bus blocked in the Codex runner (expected)
5. Minimal evidence (read-only) to gather as needed:
   • ./sysop/preflight.sh
   • ./sysop/healthcheck.sh
   • command -v powershell.exe || command -v pwsh.exe (if Windows snapshot is requested)
6. Propose the exact ./sysop/run.sh ... command(s) and STOP for approval:
   • ./sysop/run.sh all (writes under sysop/out/ and appends to learn/LEDGER.md)
   • Optional (repo-scoped auto-fix + script generation): ./sysop/run.sh all --apply-fixes

Execution mode (after approval)

• Run the approved ./sysop/run.sh ... command(s) from repo root.
• Return results using the repo output format (see AGENTS.md), including:
  • report path: sysop/out/report.md
  • “Top bottlenecks” excerpt (if present)
  • latest learn/LEDGER.md excerpt (3–8 lines)

Safety

• No destructive ops (rm -rf, git reset --hard, git clean -fdx).
• No writes outside the repo or to /etc.
• No internet fetches.

For details and rationale, see references/OPERATOR_KERNEL.md.