--- name: rfp-response-proposal-review description: > Post-generation audit agent for proposal drafts. Runs three structured checks — strategy fidelity, RFP coverage, and content sourcing — against the generated proposal draft, the approved Strategy Brief, and the RFP Tracker list. Produces a Review Report Word document flagging issues before any human reviewer opens the draft. Designed to run after Phase 3 content generation and before the human review gate. --- # Proposal Review Skill ## Firm Profile (REQUIRED — fill in before use) These values define who your firm is and how it writes. See `SETUP.md`. - FIRM NAME: [FIRM NAME] - WHAT THE FIRM DOES: [one sentence] - ROLE BOUNDARY: [what the firm will NOT claim to do — leave blank if none] - VENDOR STANCE: [vendor-neutral OR resells/implements specific products] - SPECIALIZATION: [domains the firm credibly claims] - VOICE: [3–5 adjectives] --- ## Purpose An AI-generated proposal draft may look complete but contain three categories of silent failure: drift from the approved strategy, missed RFP requirements, or content not grounded in the firm's source material. None are obvious from a quick read. This skill audits all three dimensions and produces a structured Review Report telling the human reviewer exactly what to look at — and what has already been verified — before they open the draft. --- ## Critical Output Rules > **Do not output the Review Report or audit findings to chat.** All findings go > into the Review Report Word document saved in Step 6. Chat output is limited to > one-line progress updates per step and the final summary. > > **The Review Report Word document is the only deliverable.** Accumulate findings > internally, then write the complete report in Step 6. --- ## Prerequisites - The **generated proposal draft** (.docx) must be in the deal's WIP Collaboration Folder subfolder. - The **Strategy Brief** (.docx, Phase 2) must be accessible — approved ToC, strategic themes, tone guidance, section-by-section drafting notes. - The **RFP Tracker** List must be accessible with all columns populated, including Available Content and Source 1/2/3. - Write access to the deal's WIP Collaboration Folder subfolder. --- ## Step 1 — Load All Inputs Load and read the following before any checks. **Proposal Draft** — read in full. Note: - The actual section structure and headings present - Text marked as placeholder (`[PLACEHOLDER`, `[TBD`, `[INSERT`, `[MISSING`, or similar) - Sections notably thin (less than one substantive paragraph) - Specific factual claims, statistics, or named outcomes that require source grounding **Strategy Brief (Phase 2)** — extract: - The approved Table of Contents (exact structure and order) - The strategic themes (3–5, with client priority, firm differentiator, supporting content) - Tone and voice guidance - Section-by-section drafting notes - Content gaps flagged in Phase 2 — these should appear as placeholders in the draft **RFP Tracker List** — read all rows, focusing on: - Every Question, Mandatory Requirement, and Evaluation Criterion — must be addressed in the draft - Available Content — what the content generation agent was supposed to draw from - Source 1/2/3 — the documents the agent should have consulted per row - Rows where Available Content = "No Match" — these should appear as placeholders --- ## Step 2 — Strategy Fidelity Check Cross-reference the draft against the approved Strategy Brief. *Did the draft follow the blueprint?* **2a — ToC Structure Check.** Compare actual headings against the approved ToC. For each approved section: - **Present and correct** — exists and heading matches or is a reasonable equivalent - **Present but repositioned** — exists but in a different order (flag — ordering was deliberate based on evaluation weights) - **Missing** — an approved section is absent (critical failure) - **Added** — draft contains a section not in the approved ToC (flag — may be appropriate or scope drift) **2b — Strategic Theme Coverage.** For each of the 3–5 themes, assess: - **Present** — a reader would understand the argument (not just a keyword) - **Weak** — mentioned but not developed into a clear, evidenced argument - **Absent** — no meaningful treatment **2c — Tone and Voice.** Sample at minimum the exec summary, one mid-document body section, and one appendix-adjacent section. For each: does the writing match the tone guidance and Firm Profile VOICE? Are flagged language patterns present? Does it read as [FIRM NAME]-authored or generic consultant prose? **2d — Drafting Notes Compliance.** For each section with specific drafting notes, check whether the draft followed them. Flag any section where the note said to emphasise a differentiator or avoid a framing and the draft did the opposite. --- ## Step 3 — RFP Coverage Check Re-run coverage logic on the **actual draft** — not the approved ToC. A requirement being in the ToC does not mean it was addressed. For every RFP Tracker row where Category = **Question**, **Mandatory Requirement**, or **Evaluation Criterion**, locate the corresponding passage and classify: - **Covered** — a substantive response exists (at minimum one paragraph directly addressing the requirement) - **Placeholdered** — a clearly marked, descriptive placeholder exists in the relevant section (acceptable where Phase 2 flagged a gap; the text must say what is missing, not just mark a blank) - **Missing** — neither a substantive response nor a placeholder exists (critical gap — flag for immediate action) - **Thin** — a response exists but is less than one substantive paragraph, or only generic statements not specific to this RFP (moderate priority) Produce a **Draft Coverage Map** (same format as the Phase 2 Coverage Map, plus a Coverage Status column): ``` DRAFT COVERAGE MAP RFP Row | Item (abbreviated) | Category | ToC Section | Coverage Status | Notes --------|-------------------|----------|-------------|-----------------|------ Row 1 | Describe transformation methodology | Question | Section 2 | Covered | 3 paragraphs, specific to client context Row 2 | Provide 2 examples relevant experience | Question | Section 4 + Appendix B | Covered | 2 case studies present Row 3 | Bidder authority to contract | Mandatory | Appendix C | Placeholder | Awaiting signed cert — noted correctly Row 4 | 90-day execution plan | Question | Section 3 | Thin | Only 1 generic paragraph — needs specifics Row 5 | Describe subcontracting strategy | Mandatory | Section 6 | Missing | No response and no placeholder — critical gap ``` --- ## Step 4 — Content Sourcing Check *Is the draft grounded in the firm's actual source material, or did the agent invent content?* This is the hardest check to run precisely — focus on high-risk areas. **High-risk areas:** - Sections where RFP Tracker Available Content = "No Match" — no source material. If the draft has substantive content here (rather than a placeholder), flag as potentially hallucinated. - Specific statistics, percentages, named client outcomes, or delivery timelines — the most likely hallucination vectors. Flag any not in the Available Content summaries or Source documents. - Named methodologies/frameworks claimed as proprietary — verify they appear in capability statements or past proposals, not invented for this draft. **Sourcing confidence per major section:** - **High** — clearly grounded in available source material (Source 1/2/3 or capability statements) - **Medium** — consistent with the firm's known positioning but not directly traceable to a specific source document - **Low / flag for review** — specific claims, numbers, or outcomes that appear in no available source — human must verify before submission --- ## Step 5 — Placeholder Audit Find and assess every placeholder. 1. Search all markers: `[PLACEHOLDER`, `[TBD`, `[INSERT`, `[MISSING`, `[DRAFT NOTE`, or any bracketed text signaling incomplete content. 2. For each placeholder: - Confirm it is **descriptive** (says what is missing and ideally who provides it) - Confirm it maps to a **known content gap** flagged in Phase 2 — if a placeholder appears for a requirement that had available content, flag as a content generation failure - Confirm it is **in the right section** 3. Flag any non-descriptive placeholder (e.g., `[TBD]` with no context) — must be rewritten before human review. 4. Produce a **Placeholder Register**: ``` PLACEHOLDER REGISTER Section | Placeholder Text (abbreviated) | Content Needed | Owner | Priority --------|-------------------------------|----------------|-------|-------- Section 3.2 | [PLACEHOLDER: 90-day plan detail] | Detailed execution timeline with milestones | Pursuit lead | High Appendix C | [PLACEHOLDER: signed certifications] | Executed legal forms | Contracts team | High Section 5.1 | [PLACEHOLDER: data residency controls] | Technical SME input on data handling approach | SME | Medium ``` --- ## Step 6 — Produce the Review Report Compile all findings into a single Review Report Word document — what the human reviewer reads before opening the draft. ``` PROPOSAL REVIEW REPORT [RFP Name] — [Date] Prepared by: Review Agent (Phase 4) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ OVERALL STATUS [ READY FOR HUMAN REVIEW / REQUIRES REVISION BEFORE REVIEW ] Critical issues (must fix before review): [count] Moderate issues (should fix before review): [count] Low-priority issues (fix before submission): [count] Placeholders requiring human input: [count] ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1. STRATEGY FIDELITY ToC structure: [Matches approved / Deviations noted — see table] Strategic themes: [All present / Weak or missing — see list] Tone and voice: [Compliant / Issues noted] Drafting notes compliance: [Followed / Deviations] [Findings table: Section | Issue | Severity | Recommended Action] ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 2. RFP COVERAGE [Draft Coverage Map table — from Step 3] Summary: Covered [n] | Placeholdered [n] (acceptable) | Thin [n] | Missing [n] (critical) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 3. CONTENT SOURCING [Per-section sourcing confidence ratings] Flagged items requiring human verification: [Specific claims/statistics/outcomes not traceable to source — confirm or remove before submission] ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 4. PLACEHOLDER REGISTER [Placeholder Register table — from Step 5] ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 5. PRIORITISED ACTION LIST Before the human review session: [critical and moderate issues for the pursuit team] Before submission (can be resolved during review): [low-priority issues and placeholder owners] ``` **Overall Status Logic** - **REQUIRES REVISION BEFORE REVIEW** if any critical issue exists: missing RFP requirements with no placeholder, missing ToC sections, or low-confidence sourcing flags on scored sections. - **READY FOR HUMAN REVIEW** if no critical issues, all gaps properly placeholdered, and the reviewer has a clear action list. --- ## Step 7 — Save and Report Back 1. Save the Review Report as **[ClientName]_Review_Report_[Date].docx** in the deal subfolder, alongside the proposal draft in the same WIP Collaboration Folder subfolder. 2. Report back with a concise summary (see below). Do not reproduce the full report in chat. ### Example Summary Output ``` Proposal Review complete. Overall status: REQUIRES REVISION BEFORE REVIEW Strategy Fidelity: ToC structure matches approved version Theme 2 (Vendor-Neutral Independence) present but underdeveloped — only mentioned in executive summary, not substantiated in body Section 4 tone: hedging language detected ("we believe", "we hope") — conflicts with tone guidance (confident, specific) RFP Coverage: 31 rows checked Covered: 24 rows Placeholdered: 5 rows (all properly described — acceptable) Thin: 1 row (Row 4 — 90-day plan needs specifics) Missing: 1 row (Row 14 — subcontracting strategy: no response and no placeholder — CRITICAL) Content Sourcing: 3 items flagged for human verification: • Section 2.3: "94% on-time delivery rate" — not found in any source document • Section 4.1: Named client outcome (healthcare) — not in capability statements • Appendix B: Case study dates — inconsistent with past proposal dates Placeholders: 5 found, all descriptive and correctly located Critical actions before human review: 1. Add placeholder (or draft response) for Row 14 — subcontracting strategy 2. Verify or remove 3 flagged claims in sourcing check 3. Strengthen Theme 2 coverage in Section 3 body Review Report saved to: WIP Collaboration Folder / Northwind_TransformationRFP_2026 / Northwind_Review_Report_2026-03-04.docx ``` --- ## Notes & Edge Cases - **Draft is very long:** Focus the sourcing check on sections covering the highest-weighted evaluation criteria first. If time is limited, flag that the sourcing check was partial and note which sections were not checked. - **Multiple draft versions exist:** Confirm which file is the current draft before running. If multiple versions exist in the subfolder, ask the user to confirm the correct file. - **Placeholder format varies:** Search broadly — `[`, `TBD`, `INSERT`, `DRAFT`, `NOTE:` — and standardise all placeholders to `[PLACEHOLDER: description — owner]` as part of this review. - **Thin sections on low-weight criteria:** Note the evaluation weight when flagging thin sections so the reviewer can prioritise (a thin response on a 5%-weighted criterion is less critical than on a 30%-weighted one). - **Draft contains tracked changes:** Run the review against the accepted (final) version of the text, not the marked-up version. Note this in the report. - **Content gap placeholders that were filled:** If the draft has substantive content in a section Phase 2 flagged as a content gap, flag it for human review — it may be appropriate or hallucinated. Do not assume it is correct simply because it is present.