Flux Research — DEPRECATED

This skill has been merged into flux-drive. Use interflux:flux-drive with mode=research instead.

The /interflux:flux-research command automatically routes to flux-drive in research mode. This file is retained for reference only — it will be removed in a future release.

---

Original Implementation (archived)

You are executing the flux-research skill. This skill answers research questions by dispatching only relevant research agents from the roster, collecting their findings in parallel, and synthesizing a unified answer with source attribution. Follow each phase in order. Do NOT skip phases.

Input

The user provides a research question as an argument. If no question is provided, ask for one using AskUserQuestion.

RESEARCH_QUESTION = <the question the user provided>
PROJECT_ROOT = <git root of the current working directory>
OUTPUT_DIR = {PROJECT_ROOT}/docs/research/flux-research/{query-slug}

Where {query-slug} is the research question converted to kebab-case (max 50 chars, alphanumeric + hyphens only).

---

Phase 1: Triage

Step 1.0: Domain detection (reuse from flux-drive)

Check for domain context that can sharpen research queries:

Classify the project's domains using flux-drive's LLM classification (Step 1.0.1 in flux-drive SKILL.md): read README, build files, and key source files, then classify into known domains. For each detected domain, load the domain profile from ${CLAUDE_PLUGIN_ROOT}/config/flux-drive/domains/{domain-name}.md and extract the ## Research Directives section (if present).

Fallback: If no domains detected or no Research Directives sections exist, skip domain injection — agents run with the raw query only.

Step 1.1: Build Query Profile

Analyze the research question to determine:

query_profile:
  type: <one of: onboarding, how-to, why-is-it, what-changed, best-practice, debug-context, exploratory>
  keywords: [list of key terms extracted from the question]
  scope: <narrow | medium | broad>
  project_domains: [from Step 1.0, if any]
  estimated_depth: <quick | standard | deep>

Type detection heuristics:

• "how do I..." / "what's the best way to..." → how-to
• "why does..." / "why is..." → why-is-it
• "what changed..." / "when did..." → what-changed
• "best practice for..." / "conventions for..." → best-practice
• "help me understand this codebase..." / "how is this organized..." → onboarding
• "I'm debugging..." / "context for this bug..." → debug-context
• No clear pattern → exploratory

Depth estimation:

• quick (30s per agent): simple factual lookups, single-source answers
• standard (2min per agent): multi-source synthesis, pattern matching
• deep (5min per agent): comprehensive survey, cross-referencing, analysis

Step 1.2: Score agents

Score each research agent on a 3-point scale using the query-type → agent affinity table:

Query Type	Primary (score=3)	Secondary (score=2)	Skip (score=0)
onboarding	repo-research-analyst	learnings-researcher, framework-docs-researcher	best-practices-researcher, git-history-analyzer
how-to	best-practices-researcher, framework-docs-researcher	learnings-researcher	repo-research-analyst, git-history-analyzer
why-is-it	git-history-analyzer, repo-research-analyst	learnings-researcher	best-practices-researcher, framework-docs-researcher
what-changed	git-history-analyzer	repo-research-analyst	best-practices-researcher, framework-docs-researcher, learnings-researcher
best-practice	best-practices-researcher	framework-docs-researcher, learnings-researcher	repo-research-analyst, git-history-analyzer
debug-context	learnings-researcher, git-history-analyzer	repo-research-analyst, framework-docs-researcher	best-practices-researcher
exploratory	repo-research-analyst, best-practices-researcher	git-history-analyzer, framework-docs-researcher, learnings-researcher	—

Domain bonus: If a detected domain has Research Directives for best-practices-researcher or framework-docs-researcher, add +1 to their score (these agents benefit most from domain-specific search terms).

Selection: Launch all agents with score >= 2. Agents with score 0 are skipped entirely.

Step 1.3: User confirmation

Present the triage result via AskUserQuestion:

AskUserQuestion:
  question: "Research plan for: '{RESEARCH_QUESTION}'. Query type: {type}. Launching {N} agents ({agent_names}). Estimated depth: {estimated_depth}. Proceed?"
  header: "Research"
  options:
    - label: "Launch (Recommended)"
      description: "Dispatch {N} agents in parallel for {estimated_depth} research"
    - label: "Edit agents"
      description: "Add or remove specific agents before launch"
    - label: "Cancel"
      description: "Abort research"

If user selects "Edit agents", present a multi-select AskUserQuestion with all 5 agents and let them toggle.

If user selects "Cancel", stop immediately.

---

Phase 2: Launch

Step 2.0: Prepare output directory

mkdir -p {OUTPUT_DIR}
find {OUTPUT_DIR} -maxdepth 1 -type f \( -name "*.md" -o -name "*.md.partial" \) -delete

Step 2.1: Build per-agent prompts

For each selected agent, construct a research prompt:

## Research Task

Question: {RESEARCH_QUESTION}

Query profile:
- Type: {type}
- Keywords: {keywords}
- Scope: {scope}
- Depth: {estimated_depth}

## Project Context

Project root: {PROJECT_ROOT}

[If domains detected AND Research Directives exist for this agent:]

## Domain Research Directives

This project is classified as: {domain1} ({confidence1}), {domain2} ({confidence2}), ...

Search directives for your focus area in these project types:

### {domain1-name}
{bullet points from domain profile's ### {agent-name} section under ## Research Directives}

### {domain2-name}
{bullet points from domain profile's ### {agent-name} section under ## Research Directives}

Use these directives to guide your search queries and prioritize relevant sources.

[End domain section]

## Output

Write your findings to `{OUTPUT_DIR}/{agent-name}.md.partial`. Rename to `.md` when done.
Add `<!-- flux-research:complete -->` as the last line before renaming.

Structure your output as:

### Sources
- [numbered list of sources with type: internal/external, authority level]

### Findings
[Your research findings, organized by relevance]

### Confidence
- High confidence: [findings well-supported by multiple sources]
- Medium confidence: [findings from single source or indirect evidence]
- Low confidence: [inferences, gaps in available information]

### Gaps
[What you couldn't find or areas needing deeper investigation]

Step 2.2: Parallel dispatch

Launch all selected agents via Task tool with run_in_background: true:

Task(interflux:research:{agent-name}):
  prompt: {constructed prompt from Step 2.1}
  run_in_background: true

Agent invocation:

Agent	subagent_type
best-practices-researcher	interflux:research:best-practices-researcher
framework-docs-researcher	interflux:research:framework-docs-researcher
git-history-analyzer	interflux:research:git-history-analyzer
learnings-researcher	interflux:research:learnings-researcher
repo-research-analyst	interflux:research:repo-research-analyst

Step 2.3: Monitor completion

Timeouts by depth:

Depth	Per-agent timeout
quick	30 seconds
standard	2 minutes
deep	5 minutes

Polling loop (every 15 seconds):

1. Check {OUTPUT_DIR}/ for .md files (not .md.partial)
2. Report progress:

   ✅ learnings-researcher (12s)
   ⏳ best-practices-researcher
   ⏳ framework-docs-researcher
   [1/3 agents complete]
   

3. If all expected .md files exist, stop polling
4. After timeout, report any agents still pending

Completion verification:

1. For agents that didn't complete, check background task output for errors
2. Create error stubs for failed agents:

   ### Sources
   (none — agent failed)
   ### Findings
   Agent {name} did not complete within timeout.
   ### Confidence
   No findings available.
   ### Gaps
   This agent's entire domain is a gap in the research.
   

3. Clean up .md.partial files

---

Phase 3: Synthesize

Step 3.1: Delegate to Synthesis Subagent

Do NOT read agent output files yourself. Delegate ALL collection, merging, ranking, and verdict writing to a synthesis subagent. This keeps agent prose entirely out of the host context.

Launch the intersynth research synthesis agent (foreground, not background — you need its result):

Task(intersynth:synthesize-research):
  prompt: |
    OUTPUT_DIR={OUTPUT_DIR}
    VERDICT_LIB=auto
    RESEARCH_QUESTION={RESEARCH_QUESTION}
    QUERY_TYPE={type}
    ESTIMATED_DEPTH={estimated_depth}

The intersynth agent reads all research agent output files, merges findings with source attribution, ranks sources, writes verdicts, and returns a compact answer. It writes {OUTPUT_DIR}/synthesis.md.

After the synthesis subagent returns:

1. Its return value is the compact answer (~5-10 lines) — display this immediately as the quick answer
2. Read {OUTPUT_DIR}/synthesis.md for the full report to present to the user
3. The host agent never touched any individual agent output file

Step 3.2-3.4: (Handled by intersynth)

Source ranking, merging, conflict resolution, and report generation are performed by the intersynth synthesis agent.

Step 3.5: Present unified answer

Read {OUTPUT_DIR}/synthesis.md and present to user. This file was written by the intersynth agent.

Output Summary

When complete, display:

Research complete!

Output: {OUTPUT_DIR}/synthesis.md
Agents: {N} dispatched, {M} completed, {K} failed
Sources: {total} ({internal} internal, {external} external)

Key answer: [1-2 sentence summary]