research

Files (16)

• README.md
• SKILL.md
• __init__.py
• adaptive_limits.py
• agent_filter.py
• layer2_filter.py
• orchestration.py
• query_complexity.py
• implementation-details.md · references/
• migration-notes.md · references/
• output-format.md · references/
• performance-tuning.md · references/
• troubleshooting.md · references/
• usage-examples.md · references/
• search_executor.py
• semantic_cluster.py

SKILL.md (first lines)

# Research (`/research`)

`/research` is the canonical research capability. It owns research intent,
capability routing, bounded lane execution, source opening, evidence
assessment, research artifacts, and grounded output.

Pipeline:

```text
research question
  -> evidence requirements
  -> capability routing
  -> QMD/MMX/Brave evidence gathering
  -> explicit source opening
  -> conservative assessment
  -> research-run.v1 artifact and output
```

The implementation is the existing Phase 1 engine. Provider output is not
authority; discovery is not evidence; evidence is not a claim; a claim is not
a decision. Unsupported claims remain refused.

Use `/research` for new research requests. `/all` remains a compatibility
wrapper and delegates to this workflow with compatibility caller telemetry.

## Modes

```bash
/research "question" --mode auto
/research "workspace question" --mode local-only
/research "current documentation" --mode unified
```

An experimental complementary lane may be selected explicitly through the
existing execution API or CLI (`--external-provider exa|duckduckgo`). This is
restricted/manual behavior; neither provider is an automatic candidate.

The runtime preserves immutable run-scoped artifacts, source identity,
assessment records, failures, quota/readiness telemetry, and conservative stop
behavior. Phase 2A, `/go`, `/search`, new providers, and `agy` are outside this
workflow's authorization.

From EndUser123/search-research, MIT.

More from EndUser123/search-research