Sales

Tier 0 Doctrine

This is a routing hub. It receives a /sales command, identifies the target skill, and dispatches execution. It does not perform analysis itself — subagents and skills do the work. The orchestrator's job is classification, routing, and aggregation.

Governed by: Hub-and-Spoke

• Hub: This orchestrator — receives all /sales commands
• Spokes: 13 skills + 5 agents that do the actual work
• Rim: The user — invokes commands, receives outputs

IMO

Ingress: User invokes /sales <command> [arguments]. Schema validation: command must match a registered route. Arguments must match the target skill's expected input type (URL, prospect name, description, or none).

Middle: Route to the correct skill. For /sales prospect, launch 5 parallel subagents, collect results, compute composite score, assemble final output. For all other commands, dispatch to the matching skills/sales-<command>/SKILL.md.

Egress: Markdown file output (or terminal output for /sales quick). Read-only view of the analysis. No logic in the output — it is a report.

Constants

Command Registry (Fixed)

Prospect Score Framework (Fixed)

Composite Score = Weighted average of all 5 categories (0-100).

Grade Scale (Fixed)

Output Standards (Fixed)

1. Actionable over theoretical — every recommendation is executable
2. Personalized — tailored to the specific prospect, not generic
3. Revenue-focused — every insight connects to deal probability
4. Evidence-based — cite specific sources and data points
5. Ready to use — outreach emails are copy-paste ready

Variables

Workflow

BLOCK 1: Command Classification

Governed by: C&V

Constants:

• Every /sales command maps to exactly one skill.
• Unknown commands produce an error with the command registry table.
• /sales prospect is the only command that launches parallel subagents.

Variables:

• The specific command entered by the user.
• The specific arguments provided.

IMO:

• Ingress: Raw user input — /sales <command> [args].
• Middle: Parse command. Match against command registry. If no match, return the registry table as help output. If match, identify the target skill path.
• Egress: Classified command + target skill path + parsed arguments.

Go/No-Go: Command matches a registered route. Arguments are present if required.

BLOCK 2: Skill Dispatch

Governed by: Hub-and-Spoke

Constants:

• Each skill lives at skills/sales-<command>/SKILL.md.
• The orchestrator reads the skill, then executes it — never modifies it.
• For /sales prospect, dispatch 5 subagents in parallel (not sequential).

Variables:

• The specific skill being dispatched.
• The specific arguments passed to the skill.

IMO:

• Ingress: Classified command from Block 1.
• Middle: If /sales prospect → launch 5 parallel subagent calls:
  1. agents/sales-company.md → Company research
  2. agents/sales-contacts.md → Contact intelligence
  3. agents/sales-opportunity.md → Opportunity assessment
  4. agents/sales-competitive.md → Competitive analysis
  5. agents/sales-strategy.md → Outreach strategy For all other commands → read and execute the target skill.
• Egress: Skill execution results (raw outputs from subagents or skill).

Go/No-Go: Skill file exists and is readable. Arguments match skill's expected input.

BLOCK 3: Score Aggregation (Prospect Only)

Governed by: IMO

Constants:

• Aggregation applies ONLY to /sales prospect output.
• Weights are fixed: Company 25%, Contacts 20%, Opportunity 20%, Competitive 15%, Strategy 20%.
• Each subagent returns a 0-100 score for its category.
• Composite = weighted sum of all 5 scores.
• Grade assignment uses the fixed grade scale (A+ through D).

Variables:

• The 5 individual category scores.
• The resulting composite score and grade.

IMO:

• Ingress: Raw outputs from all 5 parallel subagents.
• Middle: Extract category score from each subagent output. Apply weights. Calculate composite. Assign grade. If any subagent failed, mark that category as "incomplete" and note it in the executive summary.
• Egress: Composite score, grade, and per-category breakdown.

Go/No-Go: At least 3 of 5 subagents returned valid scores. If fewer than 3, flag as incomplete analysis — do not assign a grade.

BLOCK 4: Output Assembly

Governed by: CTB

Constants:

• Output file includes: prospect URL, date, overall score, executive summary.
• Output is structured with clear headers and tables.
• Output follows the target skill's specified file format.

Variables:

• The specific content assembled from skill/subagent outputs.
• The output filename.

IMO:

• Ingress: Skill output (Block 2) + score aggregation (Block 3 if prospect).
• Middle: Assemble the final output file:
  • Trunk: The output file itself.
  • Branches: Major sections (summary, detailed findings, recommendations, next steps).
  • Leaves: Specific data points, scores, action items. Save to the current directory with the filename from the command registry.
• Egress: Written markdown file (or terminal output for /sales quick).

Go/No-Go: Output file written successfully. Contains all required sections.

Rules

1. Never hardcode industry-specific content into the orchestrator. Industry detection rules, vertical-specific signals, and company type heuristics live in runtime config.
2. Never execute analysis in the orchestrator. The orchestrator routes — skills analyze.
3. Never modify a skill file during execution. Read and execute, never write.
4. Never run subagents sequentially when they can run in parallel.
5. Never assign a grade when fewer than 3 subagents returned valid scores.
6. Never produce output without citing data sources. "Based on our analysis" is not a source.
7. Never skip the command classification step. Unknown commands get the help table, not a guess.

Cross-Skill Integration

Quick Snapshot (/sales quick <url>)

This is the only command handled inline (no external skill file):

1. Fetch homepage via WebFetch.
2. Evaluate: company size signals, industry fit, tech indicators, growth signals, decision maker visibility.
3. Output: scorecard with top 3 opportunities + top 3 concerns.
4. Max 30 lines of terminal output. No file saved.

Reference Pointers