v1.8.0

plan-agent v1.8.0

Implementation plans now require executable test code, not just prose assertions

Mandatory Tests section between Steps and Acceptance Criteria in every generated plan
Tier 1 (code-touching): unit, integration, E2E, and objective-verification tests
Tier 2 (non-code plans): objective-verification test only — no empty stubs
Objective-verification test renders as a hero card asserting the plan's stated goal is met
Prose verification (per-step verify, criteria) and executable test code are now separate layers

Implementation plans have always had prose verification — "run this grep and confirm the output matches." But prose doesn't run. Tests do.

plan-agent v1.8.0 adds a mandatory Tests section to every generated plan, using a two-tier depth model:

- Tier 1 (code-touching plans): unit, integration, E2E, and an objective-verification test — a real test file that directly asserts the plan's stated objective is accomplished.
- Tier 2 (non-code plans): only the objective-verification test. No empty stubs.

The teachable pattern: per-step verification and acceptance criteria are prose assertions — they check that steps ran and outcomes matched. The Tests section is executable code committed to the codebase, running in CI and proving correctness long after the plan is closed.

Docs say what should be true. Tests say what is.

More agentic-dev tooling breakdowns — follow along.

#ClaudeCode #AITools #DeveloperTools #Plugins

Twitter/X

Plans have prose verification ("run this grep"). But prose doesn't run.

plan-agent v1.8.0 adds a mandatory Tests section: real test code that ships with the codebase. Docs say what should be true; tests say what is. #ClaudeCode #AITools

Bluesky

plan-agent v1.8.0 adds a mandatory Tests section to every plan.

The key distinction: per-step verify and acceptance criteria are prose. The Tests section is executable code that runs in CI after the plan is closed.

Docs say what should be true. Tests say what is.

Substack

What's the difference between "confirm the output matches" in a plan and writing a test?

One describes an expectation. The other enforces it.

plan-agent v1.8.0 adds a mandatory Tests section to every plan. Code-touching plans get unit, integration, E2E, and an objective-verification test. Non-code plans get just the objective-verification test — no empty stubs.

The principle: prose assertions document what should be true. Tests prove what is true, in CI, indefinitely.