name	test-intent-ledger
description	Use before adding or reviewing clnkr tests to keep generated tests integration-oriented, behavior-focused, and low-slop.

Test Intent Ledger

Use this before adding or accepting new tests in clnkr.

For each new test or fixture, record:

Behavior under test: user-visible behavior, protocol contract, or persistence contract.
Real boundary exercised: CLI, clnkrd JSONL adapter, shell executor, provider adapter, session store, or agent loop.
Dependency realism: real, fixture, fake, or mock.
Fake/mock reason: why a real dependency or fixture is not suitable.
Failure proof: command, sabotage patch, or reason fail-before/pass-after is impractical.
Refactor tolerance: implementation detail that should be able to change without breaking the test.

Prefer real shell, temp filesystem, httptest.Server, subprocess helpers, and JSONL transcript fixtures over isolated helper tests when practical.
Keep WET test cases when repeated protocol examples make behavior clearer.
Use provider fakes only at the model/API boundary; do not fake the behavior being claimed.
Add helpers only when they remove incidental setup noise, not when they hide behavior.

Flag tests that:

assert private implementation shape without checking agent, CLI, provider, or transcript behavior
duplicate an existing scenario without a new behavior claim
use fake/mock terminology without explaining why a real boundary is unsuitable
update protocol fixtures without explaining the user-visible or provider contract that changed
cannot be made to fail by sabotaging the intended behavior

name	test-intent-ledger
description	Use before adding or reviewing clnkr tests to keep generated tests integration-oriented, behavior-focused, and low-slop.

Use this before adding or accepting new tests in clnkr.

For each new test or fixture, record:

Behavior under test: user-visible behavior, protocol contract, or persistence contract.
Real boundary exercised: CLI, clnkrd JSONL adapter, shell executor, provider adapter, session store, or agent loop.
Dependency realism: real, fixture, fake, or mock.
Fake/mock reason: why a real dependency or fixture is not suitable.
Failure proof: command, sabotage patch, or reason fail-before/pass-after is impractical.
Refactor tolerance: implementation detail that should be able to change without breaking the test.

Prefer real shell, temp filesystem, httptest.Server, subprocess helpers, and JSONL transcript fixtures over isolated helper tests when practical.
Keep WET test cases when repeated protocol examples make behavior clearer.
Use provider fakes only at the model/API boundary; do not fake the behavior being claimed.
Add helpers only when they remove incidental setup noise, not when they hide behavior.

Flag tests that:

assert private implementation shape without checking agent, CLI, provider, or transcript behavior
duplicate an existing scenario without a new behavior claim
use fake/mock terminology without explaining why a real boundary is unsuitable
update protocol fixtures without explaining the user-visible or provider contract that changed
cannot be made to fail by sabotaging the intended behavior

test-intent-ledger