E2E agentic testing pipeline for CC version validation

[buildoak]

Summary

Tier 2 testing: a semi-automated pipeline that spins up a real wet claude session in tmux, sends a test prompt, validates the full proxy -> statusline -> compression stack, and reports results.

Pipeline

1. Open tmux session wet-e2e-test
2. Start wet claude inside it
3. Inject minimal test prompt via session-ctl
4. Poll /_wet/status until session_requests > 0 (timeout 60s)
5. Validate: requests >= 2, input_tokens > 0, context_window > 0, statusline renders
6. Optional: run wet compress on the single tool result to validate compression
7. Collect results, tear down session
8. Report: pass/fail with diagnostic details

Trigger modes

• Manual: wet check --e2e or user triggers via TG
• Automatic: Fired by cron watchdog when static checks detect a new CC version

Cost

~1 minimal API prompt per run (~500 input tokens). Negligible even if daily.

Semi-manual aspect

Pipeline runs and produces a report. User reviews. No auto-remediation.

Implementation

• scripts/e2e-test.sh or wet check --e2e subcommand
• Depends on session-ctl for prompt injection
• Results to ~/.wet/compat-checks.log and optionally TG notification

Priority

Phase 2 — build after static checks (wet check) are solid.