Overview
/silver:deep-research is the Silver Bullet orchestrator for technology decisions, architecture spikes, tech comparisons, and competitive intelligence. Research always precedes implementation — findings are written to research/ and referenced by the receiving workflow.
It never treats ad hoc reasoning as the workflow. It clarifies the decision, selects a research path, runs direct evidence-based research in the current host session by default, produces a traceable artifact, and then hands off to the appropriate implementation workflow.
/silver:deep-research maps to WF-SILVER-DEEP-RESEARCH in the v0.52.0 APO catalog. Standard composition:
After validation, hand off to /silver:feature, /silver:ui, /silver:devops, or /silver reclassification. See Composable Flows for the full catalog.
search_orchestrator with skill_portals for landscape research, capability_score eval harness, validate_report language gates, branch-correct GitHub SKILL.md retrieval, and chain_tracker citation lineage — see changelog for release notes.
/silver:deep-research takes precedence over any other matched workflow — research informs the implementation workflow. If an instruction matches both research and implementation, run research first, then hand off.
When to Use
Entry trigger signals for /silver:deep-research:
- "how should we" / "which technology" / "compare X vs Y"
- "spike" / "investigate" / "architecture decision"
- "should we use" / "what's the best approach for"
- Any request where you need information before you can plan an implementation
Catalog Composition
| Atomic flow | Command | Outcome |
|---|---|---|
AF-CLARIFY | /silver:clarify | Decision-ready research question and scope boundaries |
AF-DECIDE | /silver:deep-research (decision mode) | Evidence-backed comparison, recommendation, and confidence levels |
AF-DOCUMENT | /silver:ensure-docs | Durable research artifact under research/ |
AF-VALIDATE | /silver:validate | Gap check before implementation handoff |
If findings should become a formal spec, run /silver:spec (AF-SPECIFY) after research — it is not part of the default research composition.
Atomic Flow Execution
AF-CLARIFY — Frame the research question
Invoke /silver:clarify. Socratic clarification defines the research question, success criteria, and constraints before evidence gathering begins.
AF-DECIDE — Choose research path and gather evidence
SB classifies the question type, selects a depth mode (quick, standard, deep, ultradeep) from phases.yaml, sets research_type (default, solution-landscape, or solution-compare), and runs evidence-based research using repository context, local artifacts, official docs, primary sources, and optional search-cli retrieval with host fallback:
- A. Market/landscape —
research_type: solution-landscape— "What tools/solutions exist for X?", market survey → top-5 shortlist → SCR per candidate - B. Tech selection — "Should we use X or Y?", weighted criteria, concrete recommendation
- C. Competitive/product intelligence — "How do competitors solve X?", "What can we learn from product Y?"
For solution-landscape and solution-compare types, a mandatory need-profile interview runs before retrieval (need_profile.json, reference/need-profile-interview.md). phase_gate.py blocks DR-RETRIEVE until interview_complete: true. Named comparisons use /silver:compare (research_type=solution-compare).
Research paths
Path A — Solution landscape (solution-landscape)
Market survey for tool/category decisions under AF-DECIDE. shortlist_candidates.py produces exactly five candidates; each gets a Solution Capability Report (SCR) under solutions/<slug>/scr.md. Output: research/<date>-<topic>/ with landscape/landscape-report.md, shortlist/shortlist.json, weighted comparison/comparison-matrix.md, and serverless report.html.
Path B — Tech selection
Comparative evaluation against weighted criteria with a concrete recommendation. For named solutions you already know, use /silver:compare instead. Output: research/<date>-<topic>/comparison-report.md (or full solution-compare artifact tree with SCRs and report.html).
Path C — Competitive/product intelligence
How other products solve the problem, gaps, and adaptable patterns. Output: research/<date>-<topic>/competitive-intelligence-report.md.
Research types and SPA report
research_type | Trigger | Deep dive |
|---|---|---|
default | General research questions | Standard DR artifacts (research_report.md, decision-record.md) |
solution-landscape | AF-DECIDE tool/category decision | Market survey → top-5 SCR → weighted matrix → report.html |
solution-compare | /silver:compare A B … | SCR × N named solutions → weighted matrix → report.html |
Solution types package via compare_solutions.py → generate_report_spa.py → serverless report.html (open in Cursor browser + open report.html — no HTTP server). Default mode for solution types is deep.
AF-DOCUMENT — Persist findings
Research artifacts are written under research/<date>-<slug>/ with citations, confidence levels, and actionable recommendations. Lineage is preserved for downstream planning.
AF-VALIDATE — Pre-implementation gap check
Invoke /silver:validate when plans or specs already exist, or run a lightweight validation pass on the research artifact before handoff.
Synthesis handoff
Optionally re-run /silver:clarify with the research artifact as primary context to produce a decision-ready brief for implementation.
Handoff to implementation
After research completes, SB asks which workflow should receive the findings:
/silver:feature— build a new feature based on research findings/silver:ui— product or interface work based on research findings/silver:devops— infrastructure/deployment change based on research findings/silver:plan— convert findings into an implementation plan- Done — research-only engagement, no implementation needed
If the research continues into implementation, the artifact path (research/<date>-<slug>/) is passed as context so downstream SB context, planning, and verification can reference it. Research lineage is always preserved.
Example Invocation
Silver Bullet routes to WF-SILVER-DEEP-RESEARCH. /silver:clarify runs the need-profile interview. Path A (solution-landscape) runs under AF-DECIDE: market survey, top-5 shortlist, SCR per candidate, weighted matrix, and serverless report.html under research/. Validated and synthesized into a decision-ready handoff. Route to /silver:feature, /silver:ui, /silver:devops, or /silver:plan as needed. For named solutions already chosen, use /silver:compare Backstage Port Cortex instead.