Reusable AI skills for discovery, requirements, prioritization, process improvement, and outcome review.
Quick start · Source ZIP · Meet the agents · See an example · Compatibility · Contribute
Choose the relevant branch; these skills are not a mandatory end-to-end pipeline. The SVG is the editable source.
| Your situation | Start here | What you get |
|---|---|---|
| “We need a dashboard.” But why? | Problem framing | A focused problem brief and next investigation |
| People describe the same workflow differently | Stakeholder discovery | Source-linked findings and unresolved differences |
| A feature sounds clear until implementation starts | Requirements | Behavior, exceptions, and observable acceptance criteria |
| Everything is a priority | Prioritization | Explicit trade-offs, dependencies, and a decision record |
| Work gets stuck between teams | Process design | Current/proposed flow and improvement options |
| One change affects several systems | Change impact | Affected artifacts, risks, and transition needs |
| The pilot shipped, but did it help? | Outcome review | Evidence-based assessment and next action |
Prefer a download? Get the current source ZIP, extract it, and open this README. This tracks the default branch; use a reviewed commit or tag for a reproducible installation.
Clone the repository:
git clone https://github.com/Guilsc/business-analysis-agents.git
cd business-analysis-agentsTell your assistant:
Read skills/ba-orchestrator/SKILL.md and help me with this goal:
[describe the outcome, context, and evidence you have].
Select the smallest useful set of roles. Keep facts, assumptions, and
pending decisions separate. Return the artifact and the next action.
For automatic discovery in a compatible host, install the skill folders:
# Preview first. Use the root supported by your assistant.
python scripts/install.py --root /path/to/project/.agents/skills --dry-run
python scripts/install.py --root /path/to/project/.agents/skillsThe installer copies all nine skills and refuses to overwrite an existing skill. Python 3.10+ is needed only for the installer and validator. Reading the skills requires no Python, API key, or package installation.
ChatGPT or Rovo without local folder access? Start with the portable guide, supply the relevant role document, and use your workspace's supported instructions/knowledge interface. See host setup.
| Skill | Responsibility | Typical artifact |
|---|---|---|
| ba-orchestrator | Select roles and integrate decisions | Decision package and next action |
| ba-problem-framing | Establish the need and scope | Problem brief |
| ba-stakeholder-discovery | Gather and reconcile evidence | Research plan and evidence register |
| ba-requirements | Specify behavior and review acceptance | Requirement cards |
| ba-prioritization | Compare choices and sequence work | Decision record |
| ba-process-design | Improve operational flow | Current/proposed workflow |
| ba-change-impact | Follow a change through dependencies | Change note |
| ba-outcome-review | Evaluate observed value | Outcome review |
| ba-diagram-brief | Make models understandable | Diagram source and viewable artifact |
These are reusable roles. Your host can apply them in one conversation or delegate to separate agents when it supports that capability and you authorize it. Installing a skill does not create a running agent.
This is a content toolkit, not a hosted service or autonomous runtime. The host reads the skill instructions; the orchestrator selects focused roles and integrates their artifacts. See the architecture guide for the source, installation, and runtime boundaries.
- Frame the decision. Establish the outcome, evidence, constraints, and definition of done.
- Choose the next useful role. Route by missing work, rather than running every specialist.
- Produce and review an artifact. Preserve evidence and unresolved questions through handoffs.
- Integrate the recommendation. Explain the choice, trade-offs, and next action.
For example, “improve support routing” may need discovery before process design. If the routing pilot already exists and has measured results, start with outcome review. A request for a diagram can go directly to the diagram role when the underlying model is already known.
Read the handoff contract, routing scenarios, and three reusable workflows.
Northstar Support wants automatic ticket assignment. The example follows an initial request through a problem brief, evidence, requirements, a decision, a change note, and a pilot review.
The evidence is deliberately limited. The recommendation stays proportional to it, and the pilot review does not invent a successful outcome.
| Template | Purpose |
|---|---|
| Problem brief | Establish the problem, evidence, scope, and next step |
| Evidence register | Keep sources, interpretations, and confirmation states visible |
| Requirement card | Connect behavior, conditions, exceptions, and acceptance |
| Decision record | Explain a choice and what would reopen it |
| Change note | Assess consequences and preserve decision history |
| Outcome review | Separate intended, observed, and still-unknown results |
The diagram role integrates with Cathryn Lavery's diagram-design: clear HTML/SVG visuals for workflows, states, data, ownership, and dependencies.
Install that optional skill using its upstream instructions. It is maintained separately and is not bundled here. The repository's overview diagram includes editable HTML source and an exported PNG.
The instructions do not specify a model provider. Folder-based Agent Skills hosts can discover SKILL.md; other interfaces can use the portable role guides as instructions and reference material.
| Environment | Route |
|---|---|
| Claude Code, Codex, Cursor, GitHub Copilot, OpenCode | Native skill folders; use the host's supported root |
| Visual Studio Code | Use the active AI extension/harness's skill support |
| Hermes | Install together under the active profile's skill category |
| ChatGPT, Rovo, other text-capable assistants | Native import if available, otherwise portable instructions/knowledge |
Tool access, rendering, context limits, and agent delegation vary. See compatibility and verification; packaging checks are not a claim of live testing on every platform.
- Evidence, assumptions, recommendations, and authorized decisions stay distinguishable.
- Examples are fictional and labelled. No invented interviews or measured results.
- Roles are focused, with explicit deliverables and review questions.
- The validator checks skill metadata, references, registry targets, portable copies, and repository hygiene. The validation workflow runs these checks on Windows and Linux on pushes and pull requests; its presence is not evidence of a passing remote run.
- Behavioral evaluation cases describe expected observable behavior. They are prompts for testing a runtime, not fabricated benchmark scores.
See the release validation record for checks actually performed and their limits.
Run local checks with:
python scripts/validate.py
python -m unittest discover -s tests -vUseful contributions include tested examples, clearer acceptance criteria, better routing cases, and verified host instructions. Start with CONTRIBUTING.md and the roadmap. Please use synthetic or redistributable material.
If this helps your work, consider starring the repository or sharing a concrete example of how you used it.
Original repository content is MIT licensed. This task-oriented toolkit includes no member-book PDF, raw book extraction, or chapter-summary pack. It is not an official publication or certification product of a professional standards body.
Visual design follows the separately maintained diagram-design project. See NOTICE.md for attribution and provenance for content boundaries.