| name | syspilot.impact-python |
| group | impact |
| description | Impact analysis using sphinx-needs dependency trees. Discovers affected specification elements by traversing traceability links. USE FOR: change scoping, blast radius analysis, element discovery before spec writing.
|
Skill: Impact Analysis (Python)
Implements: SYSP_SPEC_SKILL_IMPACT_QUERY, SYSP_SPEC_SKILL_IMPACT_EXCHANGE, SYSP_SPEC_SKILL_IMPACT_GROUP
Requirements: SYSP_REQ_SKILL_IMPACT_QUERY, SYSP_REQ_SKILL_IMPACT_EXCHANGE, SYSP_REQ_SKILL_IMPACT_GROUP
Instructions
Tool
.github/skills/syspilot.impact-python/scripts/get_need_links.py — run with --help for parameter details.
Requires docs/_build/html/needs.json — run sphinx-build first if stale.
Exchange Contract
To replace this implementation: create a new skill folder, provide the same
query-by-ID capability, update the description for Copilot discovery.
No agent changes required.
Rules
Mandatory Execution
Impact Analysis is mandatory for every change. File lists provided in a Change
Request are input hints, not the complete scope. This skill MUST be executed
before any spec changes are made — the result defines the actual scope.
1. Search from consumer elements, not from new elements
A newly created element has no incoming links. Run the query from each consumer
element that the new element satisfies.
Example: You created SYSP_US_SKILL_IMPACT which :links: to SYSP_US_DESIGN
and SYSP_US_PM. Run the impact query from SYSP_US_DESIGN and SYSP_US_PM
(not from the new US).
2. Use --direction in for level transitions
in = "who links to me" = "who implements me at the next level down".
REQs link upward to USes, SPECs link upward to REQs. So querying a US with
--direction in yields its REQ candidates.
3. Depth strategy
| Transition | Depth | Rationale |
|---|
| Level 0 → Level 1 | 1 | US → REQ, direct links |
| Level 1 → Level 2 | 1 | REQ → SPEC, direct links |
| Level 2 cross-check | 2 | Catches second-order consumers (e.g. doc SPECs aggregating workflow SPECs) |
4. Raw output at Level 0, assessment at Level 1/2
At Level 0 present candidates to the user without verdicts. Assessment
(affected / not affected) happens when writing the next level.
