/logic-discovery
/logic-discovery finds answers in Logic Bee code and grounds every answer in MCP.
It does not write hook TypeScript or a {slug}-plan.md.
Confirm the type
Name type when you know it: bug | feature | idea | requirement | question.
Omit type when the problem is not named. The skill asks.
Name type: bug | feature | idea | requirement | question.
Pick a path
The skill starts broad. It shows a short map of documents, hooks, and domains.
Pick the path to narrow. After every answer the skill asks whether to dig deeper, and on which path.
Worked for 1m 12s
Map. Named documents, statuses, and downstream hooks. No proposed change.
Ask whether to dig deeper, and on which path.
Void a posted customer invoice
Source: MCP `flow_documents`, `hook_graph` — start `finance-invoices/void-invoice`.
Do not propose a change. Ground names in MCP spelling.
| Document | Path | Statuses |
|---|---|---|
| Invoice | finance-invoices/invoice | posted → voided |
Ask whether to dig deeper, and on which path.
Hold the facts
VERIFIED. Document path finance-invoices/invoice. Status posted → voided. Hook finance-invoices/void-invoice.
Stop. question answered. Deeper research is optional.
The skill labels each claim VERIFIED | INFERRED | UNKNOWN.
It keeps MCP spelling byte-for-byte. It does not invent names.
Prefer the conversation.
Persist {slug}-discovery.md when a file helps.
When type is requirement or feature, persist {slug}-requirments.md if /logic-plan needs a file.
That filename is the skill spelling.
| Type | Stop when |
|---|---|
bug | Divergence located. No cause. No fix. |
feature / requirement | Inventory of documents and hooks. Open questions listed. |
idea | Feasibility and conflicts. No build plan. |
question | Question answered. Offer deeper research. |
If MCP tools are not visible, the skill stops. Do not grep for what MCP owns.
question and idea stop the pipeline. Do not start /logic-plan for those types.
Next steps
When type is bug, feature, or requirement, run /logic-plan.