# Ba Knowledge Ops

> Templates and quality checks for capturing BA knowledge and evaluating AI-generated BA deliverables.

- Skill: `garyld1962/ba-knowledge-ops` (Agent Skill)
- Install (CLI): `npx skillmds@latest add garyld1962/ba-knowledge-ops`
- Raw SKILL.md: https://api.skillmd.com/api/skills/garyld1962/ba-knowledge-ops/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: garyld1962 (https://skillmd.com/u/garyld1962)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/garyld1962/ba-knowledge-ops

---


# BA Knowledge Operations

Use this skill for two adjacent workflows:

- `#prompt:ba-knowledge-capture`
- `#prompt:ba-eval-harness`

## Relationship to Copilot built-ins

- Use this skill to make BA knowledge durable and reusable across sessions.
- Use `/plan` only after the captured decisions or evaluated deliverables are
  ready to drive execution work.
- Use `@file` context to bring saved outputs back into later Copilot CLI runs.

## Knowledge capture templates

Match the user intent to the lightest fitting template:

1. Decision capture
2. Stakeholder intelligence
3. Requirements insight
4. Process observation
5. Meeting debrief
6. Lessons learned
7. Business rule discovery
8. AI session save

For each template:

- capture only what the user tells you
- ask for missing fields selectively
- keep the result concise and searchable

## Eval harness design

When building a BA evaluation suite:

- inventory the deliverables the user most often drafts with AI
- identify the deliverables where quality most affects credibility
- create repeatable pass/partial/fail criteria
- include BA-specific failure modes such as hallucinated facts, vague
  requirements, missing stakeholder perspectives, and untestable acceptance
  criteria

## Eval output contract

Each test case should include:

- Deliverable name
- Sample task
- Evaluation criteria
- Known AI failure modes
- Scoring rubric
- Result log template

## Examples

- **Decision capture:** Record a meeting outcome as a concise decision note with
  the decision, rationale, impact, and any follow-up fields the user actually
  knows.
- **Eval harness:** Build a pass/partial/fail rubric for AI-generated user
  stories that explicitly checks for vague requirements, hallucinated facts, and
  untestable acceptance criteria.

## Guardrails

- Quality criteria must be observable, not aspirational.
- Do not invent sample tasks or org standards.
- If one input spans multiple capture templates, suggest splitting it.

## Do Nots

- Do not blur multiple capture intents into one overloaded record when separate
  templates would stay clearer.
- Do not turn an eval harness into generic writing advice with no scoring
  criteria.
- Do not imply organizational standards that the user did not provide.

## Closed Decisions

- Captured BA knowledge must stay reusable, concise, and grounded in provided
  facts.
- Evaluation criteria must be observable and scorable.
- Use the lightest matching capture template instead of forcing every input into
  a single format.