| name | source-strategy-skill |
| description | Design a bounded source and retrieval strategy that matches each sub-question to source types, query modes, freshness, and fallback paths. |
| allowed-tools | ["read_file","list_dir","grep_search","glob_files","fetch_url","web_search","browser_capabilities","browser_list_tabs","browser_get_state","browser_snapshot","browser_screenshot","browser_console","browser_network","browser_downloads","browser_list_screens","browser_get_screen","browser_find_element","browser_assert_visible","browser_open_tab","browser_navigate","browser_back","browser_forward","browser_reload","browser_stop","browser_wait","browser_select_tab","browser_close_tab","browser_reopen_tab","browser_click","browser_double_click","browser_tap","browser_hover","[Truncated]"] |
Source strategy
Overview
Source strategy decides where and how to look before collecting evidence. It
matches the information need to local, indexed, primary, official, or
independent sources and prevents high-volume search from substituting for fit.
When to Use
Use after question decomposition and before evidence collection, especially
when sources span workspace material, knowledge indexes, websites, papers, or
time-sensitive data.
Workflow
- Map each sub-question to the source type most capable of answering it.
Prefer workspace and user-provided material first, then primary or official
sources, then credible independent analysis.
- Decide whether each lookup needs semantic retrieval, literal keyword search,
entity lookup, or direct document inspection. Keep queries short and focused.
- Set scope filters, freshness windows, language or jurisdiction constraints,
result limits, and minimum source identity needed for later verification.
- Plan query variants for terminology, competing hypotheses, disagreement,
null results, and changed conditions. Avoid near-duplicate query flooding.
- Define fallback paths for unavailable sources, low recall, low precision,
or contradictory results. Never loosen scope invisibly.
- Record the ordered search plan and stop criteria, then hand it to evidence
collection. The plan may adapt, but every material expansion stays visible.
Verification
Red Flags
- Using semantic retrieval for exact strings, identifiers, or error codes.
- Using generated answers when raw source passages are required.
- Expanding to every available source because early results are inconvenient.