Check Feedback

Check Feedback - PMF Signals from Rejection Feedback

A read-only skill that surfaces product/pricing/business-model signals from rejection feedback recorded by /check-responses. Use this for PMF reflection — questions like "is there a missing feature people keep asking for?" or "are we losing to a specific competitor?".

This skill is not for sales tactics. Tactical signals from rejections (when to recontact, who to redirect to, which industries to deprioritize) are consumed automatically by /evaluate in the daily cycle. Do not use /check-feedback to drive operational decisions.

Before starting: Read ${CLAUDE_PLUGIN_ROOT}/references/workspace-conventions.md and follow the cross-cutting conventions there (data storage, MCP error handling, document writes, output discipline).

Steps

1. Verify Arguments

• Project ID: $0 (required)

Return an error if $0 is empty.

2. Fetch Two Windows in Parallel

Call get_rejection_feedback_summary twice in parallel, both with scope: "pmf":

1. projectId: "$0", windowDays: 30, scope: "pmf" -> recent
2. projectId: "$0", scope: "pmf" (no windowDays) -> all-time

If the tool returns "Project not found", instruct the user to run /leadace first and abort.

With scope: "pmf" the server filters to PMF-relevant reasons only and computes total + percentage within the PMF subset. recontactWindows is returned with all five buckets at {count: 0, samples: []} and decisionMakerPointers is an empty array — those signals are tactical and consumed by /evaluate, not by this skill.

3. Format Report

Render in this order. Always lead with featureGapNotes — that section drives product/roadmap decisions.

a. Feature Gap Notes (PMF signal)

Pull from featureGapNotes (already ordered most-recent-first). For each entry:

- {receivedAt, ISO date} — {organizationName} / {prospectName}
  "{freeText}"

If empty in both windows, render: "No feature_gap rejections recorded yet."

b. PMF-Relevant Reason Distribution

Render side-by-side: 30-day vs all-time. With scope: "pmf" the response's primaryReasonDistribution contains only:

• feature_gap — concrete missing capability (highest PMF signal)
• already_have_solution — incumbent vendor presence (competitive pressure)
• competitor_locked — multi-year contract / renewal-only window (competitive pressure)

count, percentage, and total are all computed within the PMF subset by the server — render them as-is. Skip rows where both windows are 0.

Reason                   30 days        all-time
feature_gap              N (PP%)        N (PP%)
already_have_solution    N (PP%)        N (PP%)
competitor_locked        N (PP%)        N (PP%)
PMF-relevant total       N              N

If total is 0 in both windows, render: "No PMF-relevant rejections recorded yet."

4. Closing Note

End with one short, plain-English summary. Decide which line to write in this order — first match wins:

1. Thin signal — if total (from the response) is < 3, say so and recommend continued data collection. Stop.
2. Capability clustering in featureGapNotes — scan featureGapNotes[].freeText for repeated capability/feature terms across entries. If ≥2 notes name the same capability, use the dominant-feature example.
3. Competitor-pressure share — if already_have_solution + competitor_locked together exceed half of the PMF-relevant total, use the competitor example. (Free-text is not returned for these reasons, so do not name a specific vendor.)
4. Otherwise — use the mixed example.

Example tones:

• Dominant feature_gap capability: "Multiple recent rejections cite <feature> — strongest PMF signal is for that. Consider a strategy revision (refine it via /leadace) or product roadmap input."
• High competitor pressure: "A large share of rejections cite an incumbent solution or contract lock-in — competitive pressure is the dominant PMF signal. Consider differentiation messaging (refine the strategy via /leadace)."
• Mixed: "Rejection volume is N over 30 days, no single PMF signal dominates. Continue monitoring."

Do not invent product/strategy actions when the data does not support them.

This is a read-only skill — no DB writes, no side effects.