description: "Work package for strengthening glossary skills with model validation"
Work Packages: Strengthen Glossary Skills with Model Validation
Inputs: spec.md, plan.md Organization: one atomic package; the public route and detailed workflow are not split between implementers.
Subtask Index
| ID | Description | WP | Parallel |
|---|---|---|---|
| T001 | Capture the baseline for three synthetic smoke scenarios before changing the skills | WP01 | No |
| T002 | Add a short route to spk-doctrine-glossary/SKILL.md | WP01 | No |
| T003 | Add the model pressure-test to spec-kitty-glossary-context/SKILL.md | WP01 | No |
| T004 | Repeat the three smoke scenarios and compare them with the baseline | WP01 | No |
| T005 | Run validators, targeted tests, and the scope scan | WP01 | No |
WP01 - Three Model Quality Checks (P1, MVP)
Goal: atomically add a code cross-check, concrete edge scenario, and all-three ADR gate to the existing glossary workflow.
Independent validation: three synthetic prompts demonstrate the required behavior; validators and targeted tests pass; the diff remains within the allowed scope.
Prompt: tasks/WP01-model-pressure-test.md Requirements: FR-001-FR-006, NFR-001-NFR-003, C-001-C-004 Dependencies: none.
Included Subtasks
T001 Capture the failing-first/snapshot baseline for three synthetic smoke scenarios before changing the skills
T002 Add a short route to src/charter/offering/skills/spk-doctrine-glossary/SKILL.md
T003 Add a conditional model pressure-test to src/charter/offering/skills/spec-kitty-glossary-context/SKILL.md
T004 Repeat the same smoke scenarios and confirm the code cross-check, edge scenario, and ADR gate
T005 Run both skill validators, tests/doctrine/test_spk_skill_pack.py, and the scope scan
Actual Status
- T001-T004 completed: the failing-first baseline did not find the required mechanisms; the repeated smoke test confirmed all three.
- T005 completed: both validators passed, targeted tests reported
6 passed, and the scope scan found only the two allowed product files.
Risks and Mitigations
- Broad trigger: apply the pressure-test only to modeling work or contested terms.
- Duplication: detailed rules live only in the legacy detailed skill; the public skill only routes to it.
- Documentation noise: recommend an ADR only when all three gates pass.
- Unverified claim: require a hypothesis label when code evidence is unavailable.
Coverage Summary
| Requirements | Package |
|---|---|
| FR-001-FR-006 | WP01 |
| NFR-001-NFR-003 | WP01 |
| C-001-C-004 | WP01 |