Skill Function Test Suite

Problem

Skill updates can fail after edits: broken references, boundary crashes, skipped configurations, and fixes without regression evidence. skill-function-test turns these risks into checkable stages: it backs up the target skill, builds a blueprint and constraint list from SKILL.md and scripts/, then runs scenario tests, functional tests, and S4 execution-fidelity tests before writing test-report.md and an HTML report. It treats configuration as the workflow, reducing prompts to ask users whether to enable fixes or skip steps.

Workflow and limits

The core is a dual-track testing model:
- S1-S3: checks trigger phrases, core capabilities, and multi-step workflows against declared behavior.
- D1-D6: inspects AST syntax, reference chains, data contamination, noise, calculation correctness, and boundary robustness.
- S4: validates rule fidelity under L1-L5 noise, including challenges, skip requests, reverse instructions, environmental contamination, and condition changes.

The flow is orchestrated by runner.py, with backup.py for ZIP backups, test_config.py for configuration, hooks.py for blocking, and gen_report.py for dual-format reports. Regression confirmation is required before accepting fixes, so F-0 BLOCK issues are not silently masked. Scope: it targets structured skill packages, not plain code snippets, manual tests, or ordinary unit tests; if the target lacks SKILL.md, a scripts directory, or parseable entry points, the scan and conclusions may be less reliable.