Update docs to help agents handle full lifecycle

[myurasov-nv]

Docs-only: three gaps that caused real failures while an AI agent drove full GPU-run lifecycles (create -> provision -> train -> collect -> delete). Verified 2026-07-25 on launchpad 2x H100, verda B300, massedcompute L40S, GCP T4.

Issues

Problem observed	Change
brev exec "nohup train.py &" (ml-training.md step 7) blocks for the whole run - exec holds the session even for nohup children; step 8 tail -f never exits	Detached-launch pattern (setsid nohup ... & disown via @launch.sh + completion marker to poll); tail -n instead of tail -f
Readiness lifecycle undocumented: RUNNING != SHELL READY; sshd refuses 1-2 min after READY; benign ErrorForbidden from the create poll	Lifecycle note under Create Instances
Billing model absent: per-second, RUNNING-only (BUILDING free, no round-up); stopped instances keep accruing storage; usage feed lags	"How Billing Works" section in SKILL.md + warning in cleanup.md

Files

• SKILL.md: detach pattern, lifecycle note, billing section
• reference/commands.md: long-running note under brev exec
• prompts/ml-training.md: fix background launch, monitoring
• prompts/cleanup.md: stopped-instances-cost warning

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 23.96%. Comparing base (83ace83) to head (9d21191).
⚠️ Report is 32 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #428      +/-   ##
==========================================
+ Coverage   20.09%   23.96%   +3.87%     
==========================================
  Files         140      136       -4     
  Lines       18451    18866     +415     
==========================================
+ Hits         3707     4522     +815     
+ Misses      14328    13868     -460     
- Partials      416      476      +60     

Flag	Coverage Δ
Linux	23.96% <ø> (+3.87%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.