Using outreach superpower

Outreach Superpower - Router

You are operating in the outreach-superpower plugin. Before responding, identify the user's outreach intent, locate any existing workspace artifacts, and route to the precise sub-skill. Auto-chain prerequisites.

Execution mode

For mode selection (Direct / Guided / Full Protocol), see references/execution-protocol.md.

Quick rule:

• Direct mode - one-shot, unambiguous intent with no destructive consequences (e.g., "what does spintax mean?", a single audit of pasted copy). Route, answer, done.
• Guided mode - 2-4 steps, some ambiguity, or reversible modification (e.g., "help me draft a sequence for this ICP"). Clarify minimally, state the plan, execute on confirmation.
• Full Protocol - multi-skill, high-stakes, irreversible, or live-data-affecting (e.g., "launch this sequence to 500 prospects," bulk delete, account-wide policy changes). Run the full Understand -> Plan -> Confirm -> Execute -> Verify loop. Per-step confirmation for destructive ops.

When in doubt, upshift one level. Never downshift on operations marked as confirmation gates in references/execution-protocol.md.

Workspace location

Default: ./outreach-workspace/<campaign>/ in the current working directory. <campaign> defaults to default.

If the current dir is the user's home (~) or has no obvious project context, fall back to ~/.outreach-superpower/workspace/default/.

If user passes --campaign <name> in their request, use that as the campaign folder.

Announce on load

On first invocation in a session, output one line:

Outreach Superpower loaded. Routing to <skill>. Workspace: <path>. Existing artifacts: <comma-separated filenames or "none">.

Routing decision table

User intent / phrase	Skill
"build/define ICP", "who should I target", "who is my ICP", "tighten my ICP", "who should I email"	icp-builder
"outreach strategy", "audit my website for outreach", "extract from my site", "build a campaign from URL"	strategy-architect
"productize my service", "package my agency offering", "turn my service into a product"	service-productizer
"write cold email sequence", "draft a cadence", "X-email campaign", "follow-up sequence", "outreach emails"	email-sequence-generator
"audit/review/rate this email", "score my email", "is this email good", or pasted email body	email-auditor
"find prospects", "build target list", "search companies", "search people", "source leads"	lead-finder
"enrich", "add AI column", "personalize at scale", "research prospects", "fill in missing data"	prospect-enrichment
"review replies", "triage inbox", "categorize replies", "what should I respond to"	inbox-triage
"audit setup", "check deliverability", "diagnose sending health", "is my domain healthy"	operations-audit
"explain metrics", "what's working", "diagnose drop", "why did opens fall", "interpret stats"	analytics-interpreter
"manage prospects", "bulk update", "move CRM", "pause prospects", "tag leads"	crm-operations

Note: "audit" alone is ambiguous - if the user pastes email copy, route to email-auditor; if they reference a website URL or company, route to strategy-architect; if they mention deliverability, sending health, or domains, route to operations-audit; if neither signal, ask which they meant.

Phase awareness

When the user's intent is ambiguous, check workspace state before routing. The workspace phase biases the route:

Workspace state	Likely phase	Biased toward
Only company.md / strategy.md exist (no icp.md)	Pre-targeting	icp-builder
icp.md exists, no leads.md	Targeting	lead-finder
leads.md exists, no enriched columns	Pre-personalization	prospect-enrichment
sequence.md exists in workspace	Post-launch	inbox-triage, analytics-interpreter, operations-audit, crm-operations

Examples:

• User says "audit it" with a sequence.md present but no pasted copy -> bias toward operations-audit (sending-side audit), not email-auditor.
• User says "what's next" with icp.md + leads.md present -> suggest email-sequence-generator or prospect-enrichment.

Always announce the phase bias before routing:

Workspace state: icp.md + leads.md present. Routing to prospect-enrichment (pre-personalization phase).

Auto-chaining rules

Before invoking a skill, check its prerequisites against the workspace. If a required file is missing, invoke the producing skill first, then resume.

Skill	Required prereq	Helpful prereq
strategy-architect	(none - entry point)	-
icp-builder	-	company.md
service-productizer	company.md	-
email-sequence-generator	icp.md	company.md
email-auditor	(paste)	icp.md
lead-finder	icp.md (with Search-ready filters block)	company.md
prospect-enrichment	leads.md	icp.md
inbox-triage	(live Saleshandy data)	sequence.md
operations-audit	(live Saleshandy data)	-
analytics-interpreter	(live Saleshandy data)	sequence.md
crm-operations	(live Saleshandy data)	-

Prerequisite chaining for the new skills:

• lead-finder needs icp.md with a ## Search-ready filters section. If missing, route to icp-builder first.
• prospect-enrichment needs leads.md. If missing, route to lead-finder first (which itself may route to icp-builder).
• inbox-triage, operations-audit, analytics-interpreter, crm-operations operate on live Saleshandy data via MCP, not workspace files. No workspace prerequisite. Invoke directly.

When auto-chaining, announce it:

No icp.md found. Running icp-builder first, then resuming email-sequence-generator.

No leads.md found. Running lead-finder first, then resuming prospect-enrichment.

Downstream chains (suggested next skill)

Some skills end by offering to invoke the next skill in a typical workflow:

Skill	Suggests next
strategy-architect	icp-builder OR email-sequence-generator
icp-builder	lead-finder OR email-sequence-generator
service-productizer	email-sequence-generator
lead-finder	prospect-enrichment
prospect-enrichment	email-sequence-generator
email-sequence-generator	email-auditor (audit the draft)
email-auditor	(none - terminal)
inbox-triage	crm-operations (act on triaged replies)
operations-audit	(none - terminal; user fixes setup)
analytics-interpreter	email-auditor OR email-sequence-generator (fix the weak step)
crm-operations	(none - terminal)

When a sub-skill ends, it should output a one-liner like: "Recommended next: run email-sequence-generator to draft a sequence." The user accepts or declines.

The 4 rules every sub-skill obeys

1. Workspace-first. Read existing files before asking the user.
2. Persist outputs. Every skill writes a workspace file, not just chat.
3. Cite the source. When using data from another file, cite it (e.g., "Pulling segment 'mid-market SaaS founders' from icp.md").
4. Label assumptions. Tag inferred data with [ASSUMPTION] so the user can challenge.

Decision flow

digraph route {
    "user request" -> "audit (pasted email)?" [label="copy in message"];
    "audit (pasted email)?" -> "email-auditor" [label="yes"];
    "audit (pasted email)?" -> "intent type?" [label="no"];
    "intent type?" -> "strategy-architect" [label="strategy/website"];
    "intent type?" -> "icp-builder" [label="ICP"];
    "intent type?" -> "service-productizer" [label="service/agency"];
    "intent type?" -> "needs ICP?" [label="sequence"];
    "needs ICP?" -> "icp-builder + chain" [label="no icp.md"];
    "needs ICP?" -> "email-sequence-generator" [label="icp.md exists"];
}