Research AQE Systems

1. Read AGENTS.md, the question, and any user-supplied artifacts completely.
2. Define the decision the research must support, its scope, and freshness needs.
3. Search local evidence broad-to-narrow with rg --files, rg, code navigation,

tests, configuration, documentation, and git history.

1. For current external claims, inspect primary sources: upstream repositories,

releases, commits, issues, pull requests, and official documentation. Record dates, versions, commit identifiers, and links.

1. Trace definitions to usages and public boundaries. Map dependencies, data

flow, CLI/MCP parity, configuration generation, and tests when relevant.

1. Classify each material statement using

[references/evidence-rules.md](references/evidence-rules.md). Never repeat a benchmark, capability, or roadmap claim as fact without direct evidence.

1. Synthesize only after gathering evidence. State contradictions, uncertainty,

missing access, and stale documentation explicitly.

1. Return:

- the decision-oriented summary; - verified findings with source locations; - inferences and their supporting evidence; - gaps and risks; - prioritized recommendations and the smallest useful next validation.

Use parallel agents only when the user requests delegation or repository instructions explicitly permit it. Treat Claude agent definitions as domain references; do not copy or execute their Claude-only hooks or tool syntax.