| name | assertion-verifier |
| description | Define testable claims about your project, run them, and crystallize results into a seed. Use when you need to verify that theory matches implementation, track what passes and what breaks, and identify where potential is blocked. |
Assertion Verifier — Claims Meet Reality
Register testable assertions about your project. Each assertion has an ID, a claim, and a test function. The verifier runs them all and produces a structured report: PASS / FAIL / SKIP.
Assertion Format
Each assertion is a dict:
{
"id": "AUTH_01",
"claim": "Login endpoint returns 200 for valid credentials",
"test": lambda: check_login_endpoint(),
"source": "API spec v2.1",
"group": "auth"
}
The test function must return (passed: bool, detail: str). If it raises an exception, the assertion is marked SKIP.
Integration with Seed Cycle
The assertion verifier feeds directly into seed_cycle.py:
from seed_cycle import ASSERTIONS, run_cycle
ASSERTIONS.extend([
{"id": "DB_01", "claim": "Migration is up to date", "test": check_migration},
{"id": "API_01", "claim": "Health endpoint responds", "test": check_health},
])
run_cycle(update=True)
What the Results Mean
| Status | Meaning | Seed Effect |
|---|
| PASS | Claim confirmed by data | If ALL pass: "suspicious symmetry" tension |
| FAIL | Claim contradicted by data | High-priority "contradiction" tension |
| SKIP | Test can't run (missing prerequisite) | "Blocked potential" entry |
Usage Patterns
Nightly verification:
python -c "
from seed_cycle import ASSERTIONS, verify_assertions, crystallize_seed, load_seed
# ... register assertions ...
results = verify_assertions()
crystallize_seed(results, load_seed())
"
Interactive check:
> verify my assertions
The coder runs all registered assertions and reports results.
Key Principle
Add tests that CAN fail. If everything always passes, you're testing tautologies. The value is in the FAIL and SKIP results — they show where the system's real edges are.
$ARGUMENTS
Eval
Trigger Tests
Appropriate prompts for this skill -> activates
Unrelated prompts -> does NOT activate
Fidelity Tests
Given valid input: produces expected output
Given edge case: handles gracefully
Always reports what was done
