fix-my-positioning

Fix My Positioning — orchestrator

This skill runs a six-stage pipeline against a B2B startup URL — extraction → Fletch critique → 5-framework scoring → 3-variant generation → 7-axis pros/cons → recommendation/message-house — and finally writes a structured markdown report. Stages are numbered 1–6; the report-writing step is Stage 6's final substep.

Inputs

Parse the user's invocation:

• <url> — required, the B2B startup homepage URL
• --stage — optional: seed, A, or B+. Shapes scoring thresholds (e.g., seed-stage gets cap on category-creation claims) and recommendation logic
• --acv — optional: SMB, mid, or enterprise. Adjusts Challenger framework rigor
• --motion — optional: PLG, sales, or hybrid. Adjusts buyer-committee axis weighting
• --grade-only — optional flag. If present, stop after Stage 3 (Scoring). Skip generation, pros/cons, and message house.

If the user provides no URL, ask once. Do not invent a URL.

Pipeline

Stage 1: Extraction (delegate to agent)

Dispatch the positioning-extractor agent with the URL. The agent returns a structured markdown document containing verbatim hero copy, social proof, category labels, and the Fletch detector module results (13 failure detectors + 7 positive detectors + 9-element clarity heatmap).

Use the Task tool with subagent_type: "positioning-extractor". Pass the URL and any context flags. Wait for completion.

If the agent reports "fetch status: failed", relay the failure to the user with the agent's specific reason and stop the pipeline. Do not attempt to score on missing data.

Stage 2: Fletch critique synthesis

Read the agent's output. The Fletch detector module already ran inside the agent. In this stage, organize the findings into:

1. Critical findings — failure-pattern detectors that are blocking (F1 with --stage seed/A, F2 with --stage seed/A, F10, F11)
2. Standard findings — other failure-pattern detectors that fired
3. Positive signals — positive-signal detectors that fired
4. Clarity heatmap — the 9-element scores

Apply the score boosts from positive-signal detectors (P1 → +0.5 to Dunford c5; P2 → +0.25 to Dunford c4; P3 → +0.5 to VPC; P4 → +0.25 split between Raskin and VPC; P5 → +0.5 to Dunford c1; P6 → +0.25 to Raskin; P7 → +0.25 to Challenger). Track these boosts — they apply during Stage 3.

Read references/fletch-detectors.md if rule details need clarification.

Stage 3: Five-framework scoring

Read references/frameworks.md for the full rubrics. Score each framework independently, in this order:

1. Dunford — Obviously Awesome (0–2.0)
2. JTBD + ODI overlay (0–2.0)
3. Raskin — Strategic Narrative (0–2.0)
4. Osterwalder — Value Proposition Canvas (0–2.0)
5. Challenger Sale (0–2.0)

For each framework:

• Apply the rubric strictly. Read the 0 / 0.5 / 1.0 / 1.5 / 2.0 bands and choose the closest match.
• Quote 1–3 verbatim extracts from the agent's extraction as evidence.
• Identify the single weakest element (one phrase).
• Apply any positive-signal boosts from Stage 2.
• Apply stage-aware adjustments:
  • If --stage seed/A and Dunford is being held high because of a category-creation claim, cap Dunford at 1.0 (per frameworks.md Framework 1 stage-aware threshold)
  • If --motion PLG + --acv SMB, soften the Challenger ceiling (per frameworks.md Framework 5 stage-aware threshold)
• Write a one-sentence improvement recommendation.

Sum to a composite 0–10. Map to interpretive band (0–3 broken / 3–5 weak / 5–7 competent / 7–8.5 strong / 8.5–10 exceptional).

If --grade-only was passed: skip Stages 4 and 5 entirely. Proceed directly to Stage 6's report-writing substep. The report includes only sections 1, 2, and 3 of the output schema (composite scorecard, per-framework breakdown, Fletch findings) plus a closing note explaining that variants were skipped. Then stop.

Stage 4: Three-variant generation

Read references/exemplars.md and references/fletch-detectors.md (especially the "Fletch generation rules" and "seven structural patterns" sections) before generating.

Generate exactly three variants:

• Variant A — narrow-and-sharp (Dunford-style; narrow ICP; explicit competitive alternative; fast sales cycle). Use the Who/What/Why or Outcome-first pattern.
• Variant B — category-led (Raskin + Play Bigger style; claims or reshapes a category; longer evangelism cycle; higher ambition). Use the Category-led or Mechanism-led pattern. If --stage seed/A, do not generate a category-invention H1 — anchor to existing category + modifier instead.
• Variant C — outcome-quantified (JTBD + VPC style; leads with a specific measurable outcome on a high-importance under-satisfied job). Use the Outcome-first or Problem-first pattern.

For each variant:

1. Generate an H1 — 6–12 words, ~55–70 characters. Apply Fletch generation rules: category + modifier (G1), capability over benefit (G2), single-line constraint (G3), named ICP (G5), no forbidden words (G6).
2. Generate a subhead — 12–25 words, doing exactly one job (elaborate / specify audience / name mechanism / provide proof / contrast status quo / quantify outcome). Apply pairing rules from fletch-detectors.md (G4).
3. Identify the variant's best-fit ICP — persona + company type + context.
4. Derive 2–4 value pillars for the variant using Dunford's "list every feature, ask 'so what?', group by value outcome" method.
5. Select 3 reinforcing features mapped to the pillars. Each feature must be:
   • Differentiated vs. named competitive alternatives
   • Directly enable a pillar's outcome
   • Demonstrable in a sales pitch or demo
   • Expressible as a Fletch capability statement (verb + concrete object — what the user does)

Validate each generated H1 against:

• Length (6–12 words; 13+ requires punctuation justification)
• Forbidden-words list (G6)
• Harry Dry test (visualizable / falsifiable / unique)
• Fletch one-sentence test

If a generated H1 fails validation, regenerate it. Do not ship a flagged H1.

Stage 5: Seven-axis pros/cons

Read references/tradeoff-axes.md for the full axis descriptions.

For each variant, score all seven axes (+ / 0 / −) and write a 1–2 sentence rationale per axis:

1. TAM vs. specificity
2. Category creation vs. category entry
3. Competitive defensibility (Hamilton Helmer's 7 Powers)
4. Messaging clarity vs. expansiveness
5. VC-readable vs. customer-readable
6. Sales cycle and GTM fit
7. Buyer committee multi-threading

Apply stage / ACV / motion adjustments per the rules in tradeoff-axes.md.

After all three variants are scored, run the coherence check: flag any variant whose axis combination is structurally infeasible for the company's stage (e.g., seed-stage + category-creation + enterprise-buying-committee + PLG motion).

Stage 6: Recommendation, message house, and surface adaptations

Recommendation logic:

• If --stage was provided, tag one variant as recommended per the table in tradeoff-axes.md ("How the recommendation is chosen"):
  • seed → Variant A (unless coherence-flagged; if flagged, fall back to Variant C)
  • A → Variant A or C (whichever has higher coherence + stronger axis 1 and axis 5 scores; if tied, prefer A)
  • B+ → Variant B or C (whichever has higher coherence + stronger axis 2 and axis 4 scores; if tied, prefer C)
• If --stage was NOT provided, present all three variants neutrally with no recommendation tag, and add the "your stage matters" note.

Message house (only for the recommended variant — skip if no recommendation):

• Roof — single-sentence positioning combining category + audience + capability + outcome
• Three pillars — pillar names from Stage 4, each with a one-sentence pillar statement and 2–4 proof points
• Tagline (optional, 3–7 words)

Surface adaptations (only for the recommended variant — skip if no recommendation):

• Pitch deck title slide
• LinkedIn company tagline
• Sales cold-email opener
• Product Hunt launch
• Sales one-liner ("elevator")

Read references/output-schema.md for surface-specific constraints (character limits, format).

Stage 6 (final substep): Write report and print summary

Read references/output-schema.md for the exact report structure.

1. Determine output filename:
   • Extract domain from URL (strip https://, www., take everything before the first .)
   • Use today's date (ISO YYYY-MM-DD)
   • Filename: positioning-<domain>-YYYY-MM-DD.md
   • If the file already exists for today's date and same domain, append -v2, -v3, etc.
2. Ensure directory exists: run mkdir -p ~/Desktop/claude-code/
3. Write the report using the Write tool, following the schema in references/output-schema.md exactly. Do not add or omit sections.
4. Print the terminal summary following the format in references/output-schema.md "Terminal summary structure".

Defaults and conventions

• Do not weight the five frameworks. The orthogonality argument depends on equal weighting.
• Score in 0.5 increments only. Do not output 0.3 or 0.7.
• Quote homepage copy verbatim — never paraphrase, never invent.
• When in doubt between two adjacent score bands, choose the lower one. Generous scoring corrupts the signal.
• Generated H1s and subheads must pass all Fletch generation rules and both ruthless tests before being included in the report.
• The report file is the canonical output. The terminal summary is a pointer.

Additional resources

Reference files

For detailed rubrics, detectors, exemplars, and output structure, consult:

• references/frameworks.md — Five scoring rubrics with all 0–2.0 bands, diagnostic questions, and stage-aware thresholds
• references/fletch-detectors.md — 13 failure-pattern + 7 positive-signal detectors, the 9-element clarity heatmap, the seven structural patterns, the Fletch generation rules (G1–G6), and the two ruthless tests (Harry Dry + Fletch one-sentence)
• references/tradeoff-axes.md — Seven pros/cons dimensions, coherence-check rules, and stage-aware recommendation logic
• references/exemplars.md — Eight high-signal homepage exemplars (Linear, Stripe, Notion, Gong, Superhuman, Attio, Vanta, Clay) with H1, subhead, pattern, and per-exemplar lesson
• references/output-schema.md — Exact markdown report structure, terminal summary structure, file naming convention, and ~/Desktop/claude-code/ directory handling