v0.8.68 WhaleFlow: verification gates (compile, test, lint, review as post-agent hooks)

[Hmbown]

Problem

Sub-agents self-report "done" but there is no automated verification. The Constitution (Article II) demands ground-truth verification, but today that verification is manual — the parent must poll worktree diffs, check compilation manually, and trust agent claims. With 30+ agents running in parallel, manual verification doesn't scale.

Scope

Add post-agent verification gates that run automatically after an agent completes:

1. Compile gate — cargo check -p codewhale-tui (or equivalent for non-Rust projects)
2. Test gate — cargo test -p codewhale-tui -- {{agent_specific_tests}}
3. Lint gate — cargo clippy -p codewhale-tui -- -D warnings
4. Review gate — Spawn a reviewer agent that reads the diff and flags risks

Gate behavior

• Pass → artifact is promoted, dependent agents can proceed
• Fail → agent is retried with the failure output as context (up to N retries)
• Timeout → treated as fail, retried
• Escalate → after N retries, escalate to parent for human decision

Non-goals

• Not replacing human code review. Gates are automated checks; human review comes later.
• Not enforcing gates on every agent type. Scouts and explorers skip gates.

Acceptance

• Agent completes → compile gate runs → test gate runs → lint gate runs
• Review gate runs on diffs exceeding a configurable threshold
• Failed agent retries with failure context appended to prompt
• After 3 retries, failure is escalated to parent
• Gates are configurable per agent type and per WhaleFlow node

Related

• v0.8.68 WhaleFlow: Conductor agent type for orchestrating agent ensembles #4010 (conductor agent — gates are an input to the conductor's decision logic)
• v0.8.68 WhaleFlow: agent checkpoint and resume on timeout/interrupt #4011 (checkpoint/resume — failed gates trigger checkpoint-based retry)
• Agent policy: crates/tui/src/core/engine.rs (mode/approval/trust)