Problem
search is a one-shot result list and the MCP agent experience is detached from the web experience.
Proposed Solution
a visible, user-controlled scout workspace where a natural-language mission is converted into search criteria and progressively builds a shortlist with evidence-backed inclusion/exclusion explanations.
MVP Scope
mission composer with structured interpretation preview; user must confirm criteria before execution; progress states; candidate stream; inspect evidence; pin/reject; refine criteria; cancel/retry; page actions for find alternatives, explain match, narrow search, and build complementary shortlist; reuse existing search/MCP behavior rather than unrestricted chat.
Acceptance Criteria
every candidate explanation cites public evidence; user can inspect and modify parsed filters; cancellation stops further work; deterministic fallback if model unavailable; errors do not discard reviewed candidates; no introduction is sent without separate explicit confirmation; accessible status announcements; analytics around mission start/refine/pin/complete; integration tests for happy path, cancellation, failure.
Privacy and Trust Constraints
no fabricated agent activity, autonomous contact, hidden ranking criteria, or private-data enrichment.
Success Metrics
scout completion, meaningful actions/session, candidate-to-pin rate, refinement rate, introduction conversion after human review.
Related Issues
#73, #139, #203.
Problem
search is a one-shot result list and the MCP agent experience is detached from the web experience.
Proposed Solution
a visible, user-controlled scout workspace where a natural-language mission is converted into search criteria and progressively builds a shortlist with evidence-backed inclusion/exclusion explanations.
MVP Scope
mission composer with structured interpretation preview; user must confirm criteria before execution; progress states; candidate stream; inspect evidence; pin/reject; refine criteria; cancel/retry; page actions for find alternatives, explain match, narrow search, and build complementary shortlist; reuse existing search/MCP behavior rather than unrestricted chat.
Acceptance Criteria
every candidate explanation cites public evidence; user can inspect and modify parsed filters; cancellation stops further work; deterministic fallback if model unavailable; errors do not discard reviewed candidates; no introduction is sent without separate explicit confirmation; accessible status announcements; analytics around mission start/refine/pin/complete; integration tests for happy path, cancellation, failure.
Privacy and Trust Constraints
no fabricated agent activity, autonomous contact, hidden ranking criteria, or private-data enrichment.
Success Metrics
scout completion, meaningful actions/session, candidate-to-pin rate, refinement rate, introduction conversion after human review.
Related Issues
#73, #139, #203.