| name | legion-oracle |
| description | Research institutional knowledge before escalating questions to users. Check docs/solutions/ and codebase patterns before asking humans. |
Legion Oracle
Research institutional knowledge before escalating questions to users.
Core Principle
Check docs/solutions/ first. This codebase captures learnings from past work.
When to Use
digraph oracle_decision {
"About to ask user a question?" [shape=diamond];
"Is it a preference/requirement?" [shape=diamond];
"Might be documented?" [shape=diamond];
"Ask user directly" [shape=box];
"Use oracle" [shape=box];
"About to ask user a question?" -> "Is it a preference/requirement?" [label="yes"];
"About to ask user a question?" -> "Ask user directly" [label="no - not asking"];
"Is it a preference/requirement?" -> "Ask user directly" [label="yes"];
"Is it a preference/requirement?" -> "Might be documented?" [label="no"];
"Might be documented?" -> "Use oracle" [label="yes"];
"Might be documented?" -> "Ask user directly" [label="no"];
}
Use oracle for: patterns, conventions, solved problems, technical approaches
Ask directly for: preferences, requirements, scope decisions, human judgment
Research Strategy
Run steps 1-2 first (parallel OK), then 3-4 if needed:
| Step | Tool | Query |
|---|
| 1. Institutional learnings | Task learnings-researcher | Search docs/solutions/ for [question] |
| 2. Codebase patterns | Task Explore | Find how src/ handles [topic] |
| 3. Framework docs | Context7 MCP | resolve-library-id → query-docs |
| 4. External practices | Task best-practices-researcher or WebSearch | Current best practices for [topic] |
Output
Found: Answer with source (file:line or URL)
Not found: "Checked docs/solutions/ and codebase - no relevant learnings found" → search externally OR escalate to user
Example
/legion-oracle How should I handle GraphQL pagination?
[learnings-researcher] → No matches
[Explore] → Found src/legion/state/fetch.py uses cursor-based pagination
Answer: Use cursor-based pagination per src/legion/state/fetch.py:42
