recon

• Map the competitive landscape across the dimensions that matter for a positioning or product decision
• Different from /empire-research:compare — compare evaluates options for a known purpose; recon scouts competitors for positioning intelligence
• Different from /empire-product:vet — vet pressure-tests an idea with go/no-go output; recon produces a structured matrix of competitor data without picking a winner

• Required inputs before dispatch:
  • Competitor list (at least 2; ideally 3–6 to keep matrix legible)
  • User's company / product (for relative comparison)
  • Industry / vertical
  • Dimensions to analyze (see default-dimensions)
  • Decision the analysis informs (positioning, pricing, feature roadmap, sales battlecard)
• If any input missing → ask one clarifying question at a time
• For dimensions, suggest a default subset based on the stated decision
• MUST confirm competitor list, dimensions, and decision goal before dispatch

Suggest the relevant subset based on the user's stated decision:

• Pricing: tier structure, anchor price, free tier, packaging, hidden costs, willingness-to-pay signals
• Features: core capabilities, recent launches, gaps vs. user's product, depth on top 3 jobs-to-be-done
• Positioning: tagline, value prop, audience targeting, narrative angle, perceived strengths
• Audience: target persona, ICP, geography, segment focus
• Distribution: channels, partnerships, content strategy, ad presence
• Tech stack: platform choice, language, infrastructure (when public)
• Funding / traction: funding stage, headcount estimate, public traction signals
• Moats: network effects, data, switching costs, brand, regulatory

User can add, remove, or reweight dimensions before dispatch.

• One agent per competitor for parallel scouting
• Inspect available subagents via the Agent tool's subagent_type parameter
• Pick the available agent whose name/description best matches competitor research, market analysis, or competitive intelligence. Bundled fallback: competitive-analyst. Optionally include market-researcher for market-sizing dimensions.
• MUST list chosen agents (subagent_type) per competitor + rationale BEFORE dispatch
• If confident in every pick → dispatch immediately
• If uncertain (multiple scouts equally fit, no clear-fit agent for a competitor) → MUST confirm roster with user before dispatch; allow swaps

• Send single message with multiple Agent tool calls (one per competitor)

• Each agent receives:

  • The specific competitor assigned to them (one competitor per agent)
  • The agreed-upon dimension list with descriptions
  • User's company name and decision goal
  • Required output format (see below)
  • Ethical scope: "Use only publicly available info — websites, public reports, reviews, public ads. NEVER suggest social engineering, credential stuffing, or unauthorized access."
  • Confidence-tagging requirement (see confidence-tagging)
  • "Do NOT post findings externally."

• Required per-competitor output format:

  Competitor: <name>
  
  Summary: <2-3 sentences>
  
  Per-dimension data:
  | Dimension | Finding | Source | Confidence | As of |
  |---|---|---|---|---|
  
  Notable observations:
  - <point>
  
  Open questions / unknowns:
  - <point>
  

• Cap each agent response under 500 words

• After all competitor-agents return, produce side-by-side matrix

• Required structure:

  ## Competitive Landscape Matrix
  
  | Dimension     | <User's Co> | Competitor A | Competitor B | Competitor C |
  |---------------|-------------|--------------|--------------|--------------|
  | <dimension 1> | ...         | ... [Conf]   | ... [Est]    | ... [Inf]    |
  ...
  

• Cells include the confidence tag ([Confirmed], [Estimated], [Inferred])

• Highlight cells where one competitor dominates or where a clear gap exists

• ## Gaps & Opportunities — dimensions where the user's product is clearly behind OR where an unmet niche is visible

• ## Positioning Angle — recommended narrative or wedge based on the matrix (where can the user own a position credibly?)

• ## Action Items — prioritized, time-bound, specific (e.g. "Match competitor B's free-tier file limit within 30 days")

• ## Caveats — known unknowns, freshness of data, dimensions that need primary research

• MUST cite sources from agent reports

• MUST present report then stop; ask user which action items to pursue

• Every cell, claim, and source MUST carry one tag:
  • [Confirmed]: cited official site, public docs, public report, public reviews, public ad creative
  • [Estimated]: derived from indirect evidence (third-party reports, community estimates, similarweb-style traffic data)
  • [Inferred]: agent reasoning without citation
• Each cell MUST include an As of date when the data was observed
• MUST never present [Inferred] data as confirmed fact
• Recommendation and positioning angle MUST cite the strongest [Confirmed] evidence

• Use only publicly available information
• NEVER suggest, plan, or perform: social engineering, credential stuffing, unauthorized access, scraping behind authentication, impersonating customers or employees of competitors
• Quoted public claims by competitors are fair game; private info obtained through deception is not
• If the user requests a tactic that crosses this line, refuse and explain

• MUST verify web-search capability available before dispatch (WebSearch or equivalent in scout agents' toolset)
• If unavailable → MUST refuse and tell user: scouting depends on public sites, reviews, ads, and reports. Without web access scout agents can only emit [Inferred] data, which violates the confidence-tagging contract.

• MUST gather competitor list, dimensions, decision goal before dispatch
• MUST verify web-search capability before dispatch (see preconditions)
• MUST confirm with user before dispatch
• MUST dispatch in parallel (single message, multiple tool uses), one agent per competitor
• MUST keep findings local in chat only
• MUST NOT post to Slack, GitHub, Jira, or external systems unless user explicitly authorizes
• MUST tag every cell with confidence + as-of date
• MUST refuse out-of-scope tactics (social engineering, unauthorized access)
• MUST NOT begin implementation of any action item — recommendation only
• If zero suitable competitive-research/market-research agents exist in environment → MUST stop and tell user; never inline-impersonate a scout