Requirement Risk Assessment Skill (v1.0)
Version Changes (v1.0)
- 1st Round (Usable Version) Template Implementation: Added the universal template in to reduce output drift caused by "knowing two rounds of delivery but not how to write the first round".
- Standardization of Evidence Anchor Tags: Unified the
Evidence Anchor([Direct Quote]/[Paraphrase]/[Gap])
- Version History Moved Externally: The main Skill only retains current version changes; historical summaries are uniformly placed in the README to reduce token usage and execution drift.
0) Objectives (Do Not Confuse)
This Skill only answers three questions, and must provide reasons for each:
- Feasibility: Compliance + technical/resource deliverability
- Value for Money: Comprehensive match of price/benefit/opportunity cost/long-term burden (including payment collection/liability risks)
- Risk Mitigation Strategies: Contract terms + execution processes + technical controls + stop-loss mechanisms
Note: The total score is only a summary and cannot cover single-point fatal risks (red lines/hard gates take priority).
0.15) TL;DR (30-Second Quick Execution)
Output order: First give the three conclusion questions + Commitment Gate status, then provide evidence and actions (avoid "looks professional but not actionable").
- Phase 0: Structured extraction → Key field snapshot + information sufficiency (see 3 + 3.4)
- Phase 0.25/0.5: Pitfall signal + conflict detection (see 4 + 5)
- Phase 1: Red line check (see 6) → If hit, , skip full scoring
- Commitment Gates (see 2.6 / Appendix Q) → Decide "only pause/only ROM/milestone lockable/quote & schedule lockable"
- Phase 2: Structured scoring (see 7 + Appendix A/E/G/H) → Output range + capping + confidence level
- Phase 3-6: Pre-mortem → Risk register → Terms/execution → Minimum supplementary information set (see 8-11)
0.12) Two-Round Delivery Agreement (Default: Usable First, Detailed Later)
Purpose: Reduce omissions and drift caused by "one-time long reports", allowing users to first receive actionable conclusions before deciding whether to expand on scoring/risk register details.
Default practice (unless the user explicitly requests "full assessment in one go"):
- 1st Round (Usable Version): Only output +
[Key Field Snapshot (8 Items)]
[Conflict Detection (≤5)]
[Top Gaps/Assumptions (3-6)]
[Required Supplementary Information (≤10 Questions)]
- 2nd Round (Full Version, Optional): After information is supplemented, output
[Scoring Overview]/[Risk Register]/[Negotiation Terms]/[Execution Plan]/[ROM]/[Work Breakdown]
Exceptions (directly provide full version):
- The user explicitly states "please output according to the full template/provide the scoring table/provide ROM three scenarios + terms"; or materials are complete (information sufficiency ≥6/8 and conflicts can be resolved).
0.121) 1st Round (Usable Version) Universal Template (Recommended for Direct Reuse)
When user materials are incomplete, or you have not yet completed conflict resolution and Commitment Gate determination, prioritize using this template to deliver an "actionable first-round conclusion", and converge gaps into ≤10 high-leverage questions.
markdown
## [Overview] (See 1.4, Must Be Provided First)
...
## [Key Field Snapshot (8 Items)]
...(See 3)
## [Red Line Check]
...(See 6)
## [Pitfall Signals (≤5)]
...(See 4)
## [Conflict Detection (≤5)]
...(See 5)
## [Top Gaps/Key Assumptions (3-6)]
- ...
## [Required Supplementary Information (By Priority, ≤10 Questions)]
1) ...(Associated dimension/why it will change the conclusion)
0.2) Minimum Input Set (Layered; Default to Supplement First If Information Is Insufficient)
When the user's input is very short, or you find that the "key field snapshot" has a large number of missing items, prioritize guiding the user to supplement the following minimum information (write as much as possible; missing items will trigger capping/pause):
markdown
[Project/Requirement Name]:
[Scenario Type] Regular/Multi-requirement Comparison/Historical Reference/Emergency/Contract Renewal/Modification:
[Verifiable Objective]:
[Scope Boundary] in-scope / out-of-scope: (One of the 8 items for information sufficiency)
[Deliverable List] Code/Documentation/Deployment/Training/Source Code Ownership etc.: (One of the 8 items for information sufficiency)
[Acceptance Criteria] Acceptance checklist/test method/time limit/default pass conditions: (One of the 8 items for information sufficiency)
[Payment Structure] Advance payment/milestone payment/final payment ratio/overdue handling: (One of the 8 items for information sufficiency)
[Project Timeline & Milestones] Use absolute date and time (e.g., 2026-03-20 18:00 +08:00 or Asia/Shanghai): (One of the 8 items for information sufficiency)
[Key Dependencies] Customer-side accounts/interfaces/data/whitelists/qualifications and latest provision time: (One of the 8 items for information sufficiency)
[Data & Compliance] Involved data types/authorization/retention & deletion/cross-border sharing (if applicable): (One of the 8 items for information sufficiency)
[Decision/Acceptance Responsible Person]: (One of the 8 items for information sufficiency)
[Budget/Quote Range] (If known):
[Billing Method] Fixed price/man-day/milestone:
[Daily Rate/Man-day Unit Price] (Required for ROM quote/schedule, leave blank if unknown):
[Emergency (If Applicable)] Real Deadline, Sacrificable Scope/Quality Bottom Line, Acceptance of Emergency Premium/Channel Fee:
[Renewal/Modification (If Applicable)] Historical Performance/Outstanding Debts, Historical Payment Collection, Modification Frequency and Sign-off Mechanism:
[AI/LLM (If Applicable)] Performance Metrics & Evaluation Dataset, Data Authorization Boundaries, Cost & Latency Budget, Vendors & Deployment, Security & Unauthorized Access:
[Confidentiality Restrictions] Is it allowed to quote short original text (≤25 characters) in the output? If not: Can you provide desensitized short text/field values for audit purposes?
0.25) Quick Entry (Execution Order: Gates First, Scoring Later)
Default execution sequence (even if "quick triage" is selected for output, internally complete all "gates"):
- Scenario Identification (See 0.5)
- Phase 0: Structured Extraction + Information Sufficiency X/8 (See 3 + 3.4)
- Pitfall Signal Scanning + Conflict Detection (See 4 + 5)
- Red Line Check (Prerequisite) (See 6)
- Red line hit (with evidence) ⇒ Conclusion is only , skip full scoring; only output "red line evidence + minimal risk set + terms/stop-loss suggestions + next steps"
- Red line elements missing and unjudgeable ⇒ Conclusion prioritizes
Pause (Reject Commitment Before Supplementing)
- Only if no red line is hit and information is sufficient: Enter scoring (7) → Pre-mortem (8) → Risk register (9) → Decision/negotiation/execution (10) → Minimum supplementary information set (11)
Note: Scoring is only a summary; any "gate" (red line/hard gate/Score Capping/conflict) can directly override the upper limit of the conclusion.
0.3) Applicable Boundaries (Prevent Overreach and False Certainty)
- Do not replace legal advice: This Skill only provides risk framework and term suggestions; when disputes arise regarding "liability/compensation/IP/data processing/security obligations", it is recommended to involve legal counsel.
- Do not make factual assumptions: Historical references must be based on historical materials provided by the user; without materials, only "reusable models/suggestions" can be given.
- Do not fabricate evidence: Any "evidence anchor" must come from short quotes of the user's original text; if the user explicitly requests "no pasting/no quoting of original text", use according to Appendix M of
references/assessment-standard.md
0.5) Scenario Identification (Top Priority)
After receiving the user's input, first identify the scenario type:
| Scenario | Trigger Condition | Assessment Focus |
|---|
| Regular Assessment | Single requirement/project | Standard full process |
| Multi-requirement Comparison | "Compare A and B"/"Which is better"/"Prioritization" | Horizontal comparison, difference analysis |
| Historical Reference (Requires User-Provided Materials) | "How was that previous project"/"Reference for similar projects" | Historical pattern matching, experience reuse (no fabrication of historical results allowed) |
| Emergency Requirement | "Urgent"/"Emergency"/"Rush"/"Need it tomorrow"/"Today" | Emergency premium, channel fee, impact of priority conflict (affordability) |
| Contract Renewal/Modification | Hits "renewal/modification/iteration/phase 2" | Historical performance cost, modification drift assessment |
| AI/LLM (Additional Tag, Stackable) | "Large model/LLM/RAG/fine-tuning/Agent/conversational/generative" | Enable Appendix O special scanning (performance metrics/data authorization/cost & latency/injection & unauthorized access/vendor policies) |
Note: AI/LLM is an "additional tag" that can coexist with scenarios such as emergency/renewal/multi-requirement comparison; during assessment, append Appendix O checks and questions to the standard process.
1) Output Strategy (Two Tiers, Avoid False Precision)
A. Quick Triage (Applicable: Very Short Text/Large Information Gaps/Only Asking "Can We Accept It")
Only output (must first provide
):
- Overview: Overall decision + three conclusion questions + information sufficiency/confidence level + Commitment Gate (including strong commitment upper limit)
- Red line conclusion (✅/❌/⚠️) + evidence anchor/gap
- Hit pitfall signals (if any)
- 3-6 highest-leverage gaps/conflicts
- Minimum supplementary information set (≤10 questions)
B. Full Assessment (Applicable: Sufficient Evidence or Must Make Decision/Quote Negotiation)
Output a complete report following the full Skill process.
C. Multi-requirement Comparison (Applicable: Need to Compare 2+ Requirements Horizontally)
Output a comparison matrix:
- Independent scoring for each requirement
- Horizontal dimension comparison
- Strengths and weaknesses summary
- Recommendation suggestions
Recommended output format (assess independently first, then summarize and compare):
markdown
## [Multi-requirement Comparison: Conclusions and Ranking]
Recommended Ranking: 1) Requirement A 2) Requirement B 3) Requirement C
Ranking Basis: [1-3 auditable reasons: red lines/hard gates/payment structure/scope certainty/timeline risk…]
|------|------|-----------|------------|------------------|--------|-------------|---------------------|--------|
| A | ... | ... | .../100 | ... | ... | ... | ... | ... |
| B | ... | ... | .../100 | ... | ... | ... | ... | ... |
Input Specifications for Multi-requirement Comparison (Must Clarify If Information Is Insufficient):
- Each requirement must have an identifiable name (e.g., "Requirement A/Requirement B" or actual project name).
- If the user mixes multiple requirements in a single paragraph and boundaries are unclear: First request the user to provide them in separate paragraphs or mark boundaries with headings before scoring ("脑补分段" is prohibited).
- If is triggered: Must clearly state the reason "gaps prevent fair comparison" and provide "minimum supplementary information set (by priority)", then perform ranking after supplementation.
Default Ranking Rules for Multi-requirement Comparison:
- First bucket by conclusion: >
Negotiate Before Accepting
- Within the same bucket, rank by: fewer triggers of hard gates/red lines/Score Capping first; then consider and confidence level (confidence level takes priority over total score)
D. Historical Reference (Applicable: User Requests "Reference for Similar Projects/Review of Previous Projects")
If the user
does not provide historical materials (assessment reports, contract terms, reviews, key facts):
Only output "reusable failure patterns/high-frequency pitfall signals/recommended terms and stop-loss points", and clearly mark:
[Historical Materials Missing, Cannot Conduct Factual Comparison]
.
E. Pause (Applicable: Insufficient Information to Make Commitments)
The conclusion must allow output as
(not "Reject", nor "Negotiate Before Accepting") if any of the following conditions are met:
- Red line check shows
⚠️ Insufficient Information to Supplement
- Information sufficiency , and the user still requests "give a yes/no answer now";
- Dense conflict items (see Appendix I) that cannot be resolved with existing materials;
- Any one of payment structure, acceptance criteria, or scope boundary is "missing + non-inferable", making it impossible to form executable terms and stop-loss points.
Explanation:
= "No delivery commitment/quote locking/scheduling commitment will be made before information is supplemented; enter full assessment or quote negotiation after supplementation".
F. Output Truncation Rules (Default, Avoid "Long but Unusable")
Unless the user explicitly requests "expand details/attach full tables", default to the following output limits:
- Evidence ledger: Default to only output "Key Field Snapshot (8 Items)"; full ledger is only output when (multi-requirement comparison/dense conflicts/user request).
- Pitfall signal scanning: Maximum 5 items (sorted by impact on conclusion).
- Conflict detection: Maximum 5 items (sorted by impact on conclusion).
- Pre-mortem: 3 items are sufficient; each item does not exceed 6 lines.
- Risk register: 6-10 items; and highlight the top 3 priorities above the table (corresponding to negotiation/stop-loss actions).
- Negotiation terms: 3-7 mandatory conditions, with top 3 marked (the three that most affect the conclusion).
G. ROM Quote/Scheduling (Optional: User Requests "Quote/Scheduling/Man-days" or in Quote Negotiation)
Output Objective: Before uncertainty is eliminated, only provide ranges + assumptions + verification actions, and bind "conclusion upper limit/negotiation terms" with ROM to avoid "oral commitment first then backfire".
Default Rules:
- First determine Commitment Gates (see 2.6/Appendix Q): ROM belongs to G1; before reaching G3, "lock quotes/lock schedules/write fixed launch times" are prohibited.
- If in G0: Do not output ROM, only output + minimum supplementary information set (avoid "looks like a quote/schedule is given but information is insufficient").
- If information sufficiency or key conflicts/hard gates/capping are hit: Only output ROM range (do not provide single-point quotes/single-point schedules).
- ROM must explicitly list: Top 3-6 key assumptions, corresponding verification plans, and "top 3 mandatory negotiation terms".
- ROM range boundaries: Align with the three-scenario caliber of Appendix H using (and must not bypass Appendix E hard gates).
Refer to
Appendix N of
references/assessment-standard.md
for ROM reference caliber.
1.4) Output Header (Mandatory, Must Be Provided First for All Outputs)
Purpose: Enable "consistency of conclusions/overreach of commitments/auditability of evidence" to be reviewed within 30 seconds; also stabilize the summary of multi-requirement comparisons.
Hard Rules:
- Even if
Reject/Pause/Quick Triage
- If is not calculated: Write and explain the reason (e.g., early stop due to red line).
Unified Template:
markdown
## [Overview]
Overall Decision: Accept/Negotiate Before Accepting/Pause/Reject
Feasibility: Accept/Negotiate Before Accepting/Pause/Reject
Value for Money: Worth Doing/Feasible but Not Worth Doing/Pause/N/A (Early Stop Due to Red Line)
Risk Mitigation (One Sentence): Highest-leverage measures for [terms/process/technology/stop-loss] (1-2 items are sufficient)
Scenario Tags: Regular/Multi-requirement Comparison/Historical Reference/Emergency/Contract Renewal/Modification + AI/LLM(if applicable)
Overall Confidence Level: High/Medium/Low
Information Sufficiency: X/8 or -/8 (Not Evaluated)
Commitment Gate: G0/G1/G2/G3 (Reason Summary) | Strong Commitment Upper Limit: Only Pause/Only ROM/Milestone Lockable/Quote & Schedule Lockable
1.5) Minimum Output Template (Deliverable Even for Early Stop)
Purpose: When
Reject/Pause/Quick Triage
is triggered, avoid outputting "only one sentence conclusion" or "long but unexecutable" reports.
T1 Reject (Red Line Hit)
markdown
## [Overview]
Overall Decision: Reject
Feasibility: Reject (Red Line)
Value for Money: N/A (Early Stop Due to Red Line)
Risk Mitigation (One Sentence): Provide alternative paths/compliance process suggestions that do not trigger red lines (do not provide overreach implementation details)
Scenario Tags: ...
Overall Confidence Level: Medium/High (Can be high if red line evidence is clear; reduce to medium/low if citation is restricted or key elements are missing)
Information Sufficiency: -/8 (Early Stop Due to Red Line)
Commitment Gate: G0 (Red Line) | Strong Commitment Upper Limit: Only Reject/Only Pause
## [Red Line Evidence]
- [Red Line Item]: Evidence Anchor (Short Original Quote)
## [Minimal Risk Set (No Long Report Expansion)]
- Main Risk: ...
- Impact: legal/reputation/… (1 sentence)
## [Next Steps (If Still Want to Proceed)]
1) Need User to Supplement/Clarify: ... (Gap)
2) Or: Revised Scheme/Alternative Path (No Red Line Trigger): ...
T2 Pause (Assess After Supplementing; Reject Commitment Before Supplementing)
markdown
## [Overview]
Overall Decision: Pause (Reject Commitment Before Supplementing)
Feasibility: Pause
Value for Money: Pause
Risk Mitigation (One Sentence): First formalize scope/acceptance/payment/dependency fields and obtain written confirmation, then enter assessment/quote negotiation
Scenario Tags: ...
Overall Confidence Level: Low/Medium
Information Sufficiency: X/8
Commitment Gate: G0 (Reason Summary) | Strong Commitment Upper Limit: Only Pause/Only Assess After Supplementing Information
## [Key Gaps/Conflicts (Top 3-6)]
- Gap/Conflict: ... (Impact: Which dimensions/conclusion upper limits)
## [Required Supplementary Information (By Priority, ≤10 Questions)]
1) ...(Associated dimension/why it will change the conclusion)
T3 Quick Triage (Non-Red Line, But Large Gaps/Only Asking "Can We Accept It")
markdown
## [Overview]
Overall Decision: Accept/Negotiate Before Accepting/Pause/Reject (Quick Triage)
Feasibility: Accept/Negotiate Before Accepting/Pause/Reject
Value for Money: Worth Doing/Feasible but Not Worth Doing/Pause
Risk Mitigation (One Sentence): ...
Scenario Tags: ...
Overall Confidence Level: Low/Medium
Information Sufficiency: X/8
Commitment Gate: G0/G1 (Reason Summary) | Strong Commitment Upper Limit: Only Pause/Only ROM
Red Line: ✅/❌/⚠️ (See [Red Line Check] or give 1-line conclusion + evidence/gap here)
## [Pitfall Signals (Maximum 5 Items)]
- Hit: ... (Evidence Anchor) → Impact
## [Highest-Leverage Gaps/Conflicts (3-6 Items)]
- ...
## [Next Steps: Minimum Supplementary Information Set (≤10 Questions)]
1) ...
2) Mandatory Rules (Improve Accuracy, Violations Are Considered Incomplete)
- Forbidden to Fabricate Citations: Any "evidence anchor" must come from short quotes of the user's original text; if the user requests no pasting of original text, use according to Appendix M of
references/assessment-standard.md
[No Original Text Evidence Found]
- Conflicts Take Priority: When key conflicts occur, related dimensions cannot be scored 4-5, and overall confidence level usually cannot be "High".
- Gap Triggered Upper Limit (Score Capping): Strictly follow Appendix E of
references/assessment-standard.md
- Non-Linear Decision Gates Take Priority Over Total Score: Strictly follow "hard gates" in Appendix E, and list triggered gates in the report.
- Stop-Loss First, Optimism Later: Best/worst scenarios can only be established on the premise that "key assumptions are explicitly stated", and cannot bypass capping/hard gates.
- Consistent Scoring Caliber: Dimension scores are uniformly 0-5; total score is uniformly /100 (formula see Appendix H). Any table must explicitly mark units.
- No "Brainstorming of Outcomes" for Historical References: Unless the user provides historical assessment reports/project reviews/key facts, only output "reusable risk patterns/term suggestions", and do not assert "what happened then/what the result was".
- Multi-requirement Comparison: First Independent, Then Comparative: Each requirement must first complete "red line/pitfall/conflict/scoring/Pre-mortem/risk register" according to the same caliber, then output horizontal differences and recommended ranking (do not only give a combined total score).
- Conclusions Must Be Split into Three Questions: The final output must separately provide "Feasibility / Value for Money / Risk Mitigation Strategies", and map to evidence/gaps/conflicts and term actions.
- Conclusion Set Must Include Pause: When "insufficient information/key red line gaps/unresolvable conflicts" are triggered, must be allowed as output, and do not force a score or force selection of "Accept/Negotiate Before Accepting/Reject".
- Time Must Be Auditable: When encountering expressions such as "tomorrow/next week/as soon as possible/the sooner the better", they must be converted to absolute time points (including time zone); if conversion is not possible, it is regarded as a key gap and confidence level is reduced/trigger pause or capping (especially affecting timeline feasibility and emergency identification).
- No False Precision in Quotes/Schedules: When the user requests "quote/schedule/man-days", default to outputting ROM ranges (Appendix N); if information sufficiency < 6/8 or key capping/conflicts are triggered, "single-point quotes/lock schedule commitments" are prohibited, only "ranges + assumptions + mandatory negotiation terms + verification plans" can be provided.
- Best/Worst Must Be Verifiable: As long as best/worst or "if...then..." scenario deductions are given, "verification plan (Top 3-6)" must be provided simultaneously to turn key assumptions into executable verification actions (Appendix N).
- AI/LLM Triggers Special Scanning: When keywords such as "large model/LLM/RAG/fine-tuning/Agent/conversational/generative" are hit, Appendix O red line/pitfall/questions and terms must be appended; if any of performance metrics/evaluation dataset/data authorization/cost & latency cannot be fielded, the conclusion prioritizes or at most
Negotiate Before Accepting
- Security and Operations Cannot Be Omitted: When requirements involve public networks, login authentication, payments, personal/sensitive information, external APIs, AI/LLM tool calls/data回流 etc., at least supplement the "minimum security/operations baseline" according to Appendix P of
references/assessment-standard.md
- Default to Split Work for High Uncertainty: When information sufficiency , or key assumptions ≥ 3, or ROM is marked as "For Exploration Only", default to recommending phased delivery of "Phase 0 (Verification/Discovery) → Phase 1 (Delivery)", and set Phase 0 as a pre-milestone and payment/acceptance node (see 10.6).
- Commitment Gates Must Be Explicitly Provided: When the user requests "lock quotes/lock schedules/strong commitment to performance/sign-off", output (see 2.6/Appendix Q), and clearly state the "allowed commitment upper limit" (e.g., Only ROM/Cannot Lock Schedule/Cannot Commit to Performance).
- Minimal Disclosure of Sensitive Information: If evidence anchors and examples contain phone numbers/emails/ID cards/bank cards/full customer names/specific addresses/accounts/keys/tokens/private links etc., they must be desensitized with placeholders (e.g., , , ), retaining only the minimal information needed to support judgment.
- Score→Conclusion Mapping Is Only a Summary: If "score ranges" are needed to assist summary conclusions, follow Appendix N3 of
references/assessment-standard.md
2.5) Decision Algorithm (Gates First, Scoring Later, Avoid Total Score Covering Fatal Risks)
Give conclusions in the following order (and clearly mark hit items in the report):
- Red Line (Appendix B): If clearly hit with evidence ⇒ ; if key red line elements are missing ⇒ .
- Score Capping + Conflicts (Appendix E/I): Triggered items ⇒ Apply upper limit to related dimensions; reduce overall confidence level according to Appendix G.
- Non-Linear Decision Gates (Appendix E): Any hit ⇒ Conclusion is at most
Negotiate Before Accepting
- Then Look at Total Score and Range (Appendix H): Total score is only a summary; recommendations must link back to (red lines/hard gates/risk register/negotiation terms).
Note: When "gates" conflict with "scores", "gates" take precedence.
2.6) Commitment Gates (Determine How Far You Can Commit)
Purpose: Converge strong commitments such as "accept/quote/schedule/performance" into auditable thresholds; default to downgrading to
or
when evidence is insufficient to avoid single-point commitment backfire.
G0-G3 (From Low to High)
-
G0: Pause Gate (Default Conservative)
Trigger: Red line
with missing key elements, information sufficiency
, or any one of scope/acceptance/payment is "missing + non-inferable".
Allowed Output:
(or
if red line is clear) + minimum supplementary information set.
Forbidden: Quotes/lock schedules/commit to launch times/commit to performance.
-
G1: ROM Gate (For Exploration Only)
Trigger: No red line hit, and information sufficiency
; allow ROM formation with "key assumptions + verification plans".
Allowed Output: ROM range (Appendix N) + top 3 mandatory negotiation terms + verification plan (Top 3-6).
Forbidden: Single-point quotes, lock schedules, strong commitment to performance (especially AI/LLM).
-
G2: Negotiable Commitment Gate (Milestone Lockable, But Strong Commitment Still Forbidden)
Trigger: Information sufficiency
, and scope/acceptance/payment/dependencies can be fielded; hard gates/conflicts have clear term safeguards.
Allowed Output: Phased delivery (10.6) + milestone/acceptance/payment suggestions; can provide "conditional schedule/quote range" (with terms and dependencies as preconditions).
Forbidden: "Lock schedules and single-point quotes" before contracts/dependencies are finalized; forbidden to make strong commitments to AI/LLM performance.
-
G3: Strong Commitment Gate (Quote/Schedule Lockable, But Boundaries Must Still Be Stated)
Trigger: G2 conditions are met, and key terms are formalized (contract/order/email confirmation), key dependencies and acceptance environment are available, and liability boundaries and stop-loss mechanisms are executable.
Allowed Output: Under clear boundaries/assumptions/change mechanisms, quotes and schedules can be locked (still recommended to retain buffers and change terms).
Still Forbidden: Commit to "100% error-free/performance guaranteed/unlimited liability" and other uncontrollable results; AI/LLM can only commit to "evaluation caliber + threshold + grayscale and rollback".
Output Requirements (Mandatory)
- Add a line to in full assessments:
Commitment Gate: Gx (Reason Summary) | Strong Commitment Upper Limit: ...
- When the user explicitly requests "finalize quotes/schedules/performance today", first give the Commitment Gate conclusion, then provide ROM/gaps and next steps (do not calculate scores first).
3) Phase 0: Structured Extraction + Evidence Ledger (Must Be Done First)
Extract requirement text into fields, and record for each field:
- Status:
Clear / Implied(Assumed) / Missing / Conflicting
- Evidence Anchor: 1-3 short original quotes; if the user requests no pasting of original text, use according to Appendix M of
references/assessment-standard.md
- Remarks: If "Implied", must write "Key Assumption"
Three States of Evidence Anchor (Unified Writing, Reduce Caliber Drift):
- : Short original quote from user or desensitized short quote provided by user (desensitize first according to Appendix M if containing sensitive fields)
[Paraphrase] ... (Citation Restricted)
- : Evidence needed but not found in user materials (must affect confidence level/trigger capping/trigger pause gate)
Mandatory Requirements:
- First Complete Full Evidence Ledger (Internal Step): Used for subsequent scoring, capping, conflicts, and questions; avoid "scoring by feeling".
- Default to Only Output Key Field Snapshot Externally: Unless requested by user or needed for multi-requirement comparison/conflict tracing, do not expand full ledger.
Default Output (Key Field Snapshot, 8 Items):
markdown
## [Key Field Snapshot]
|----------|---------------------------|--------------------|------------------|
| Scope Boundary | ... | ... | ... |
| Deliverable List | ... | ... | ... |
| Acceptance Criteria | ... | ... | ... |
| Payment Structure | ... | ... | ... |
| Timeline/Milestones | ... | ... | ... |
| Key Dependencies | ... | ... | ... |
| Data & Compliance Points | ... | ... | ... |
| Decision/Acceptance Responsible Person | ... | ... | ... |
Expand On Demand (Full Evidence Ledger):
markdown
## [Evidence Ledger (Structured Extraction)]
|------|---------------------------|--------------------|----------------|--------------|
| Background | ... | ... | ... | ... |
| Objective | ... | ... | ... | ... |
| Scope | ... | ... | ... | ... |
| Deliverables | ... | ... | ... | ... |
| Acceptance | ... | ... | ... | ... |
| ... | ... | ... | ... | ... |
3.1 Standard Fields (Cover As Much As Possible)
- Background: Current pain points and drivers
- Objective: Business objective/success metrics (verifiable)
- Scope: in-scope / out-of-scope (boundaries)
- Deliverables: Code/documentation/deployment/training/source code ownership etc.
- Acceptance: Acceptance standards + process + time limit + default pass conditions
- Non-Functional Requirements: Security/performance/availability/compliance/observability/compatibility (for public networks/sensitive data/external APIs, supplement minimum baseline according to Appendix P of
references/assessment-standard.md
- Data & Compliance: Data type/sensitivity level/source authorization/retention & deletion/cross-border sharing
- Dependencies & Integration: Third-party APIs/internal systems/customer-side input/account qualifications
- Stakeholders & Decisions: Decision-maker, owner, acceptance responsible person, feedback loop
- Plan & Resources: Milestones, key dates, customer cooperation items, own resource constraints
- Commercial & Contractual: Budget, payment nodes, change mechanism, maintenance scope, SLA (if any)
- Liability & Boundaries: Liability upper limit, compensation scope, confidentiality, security obligations, dispute resolution
3.2 Special Fields for Renewal/Modification (Must Be Supplemented When Input Hits "Renewal/Modification/Iteration/Phase 2")
- Historical Performance: Past delays/rework/disputes, known遗留缺陷/technical debt
- Current Cost: Support/operations/emergency input (man-days/frequency)
- Payment Collection & Credit: Outstanding payments, historical overdue, entity changes
- Change History: Number of scope drifts, presence of change orders and sign-offs, frequency of emergency task insertion
- Assets & Intellectual Property: Ownership of existing code/documentation, third-party dependencies and licenses, material authorization
3.3 Special Fields for Emergency Requirements (Must Be Supplemented When Input Hits Hard Signals Such As "Urgent/Rush/Very Urgent/Insertion/Today/Tomorrow/48 Hours/No Time to Follow Process")
- Real Deadline: Customer-stated deadline vs actual executable window
- Impact of Priority Conflict: Impact of insertion on existing projects (delay, resource preemption, quality degradation), and convert to "Priority Conflict Affordability" (A6) during scoring
- Emergency Premium: Whether rush fees are included, whether additional resources need to be mobilized
- Channel Fee: Whether "emergency channel" needs to be occupied, whether there is additional communication cost
- Quality Sacrifice Level: Which quality/functions can be sacrificed to gain time
- Acceptance Risk: Increased probability of acceptance disputes in emergency mode
3.4 Quick Check of Information Sufficiency (Prevent "Looks Professional But Unreliable")
Count whether there is clear evidence for the following "key fields":
- Scope Boundary, Deliverable List, Acceptance Criteria, Payment Structure, Timeline/Milestones, Key Dependencies, Data & Compliance Points, Decision/Acceptance Responsible Person
Output:
Information Sufficiency: X/8
- If : Default to "quick triage" and prioritize (unless the user explicitly states "only for exploration, no commitments/no quotes/no schedules"); overall confidence level cannot be "High" in full assessments.
4) Phase 0.25: Pitfall Signal Scanning (Mandatory)
Use Appendix C of
references/assessment-standard.md
to do two things:
- List hit original short quotes (evidence anchors)
- State which dimensions/conclusion tendencies it will affect (at least: confidence level reduction, entry into risk register)
Rule: Hit items must enter the "risk register", and at least affect confidence level or trigger negotiation terms (do not only treat as "complaints").
Output Format:
markdown
## [Pitfall Signal Scanning]
- Hit: [Tactics/Signal] - Evidence Anchor - Impact
5) Phase 0.5: Conflict Detection (Mandatory)
Detect key conflicts according to Appendix I of
references/assessment-standard.md
, output a list of hit items and evidence anchors, and state impacts:
- Impact on upper limit of dimension scores/confidence level
- Whether it triggers the tendency of "Negotiate Before Accepting/Reject"
6) Phase 1: Red Line Check (Prerequisite)
Use Appendix B of
references/assessment-standard.md
.
If any hit: Conclusion can only be
Reject or
Pause (Insufficient Information, Reject Commitment Before Supplementing).
Output Format:
markdown
## [Red Line Check]
✅ Pass / ❌ Fail / ⚠️ Insufficient Information to Supplement
- [Check Item]: Pass/Fail/Insufficient - Evidence Anchor or Gap
7) Phase 2: Structured Scoring (Evidence + Confidence Level + Three Scenarios)
Dimensions and weights follow Appendix A of
references/assessment-standard.md
; calculation formula follows Appendix H.
7.0 Caliber Clarification (Avoid Scoring Drift in Emergency Scenarios)
- Base Total Score(/100): Only use A1-A5 (total weight=100).
- Emergency Adjustment Score: Only calculated when identified as "emergency requirement", range (formula see Appendix H of
references/assessment-standard.md
- Emergency Total Score:
Base Total Score + Emergency Adjustment Score
7.1 Scoring Output Must Include Two Additional Items (Auditability)
- Triggered Score Capping: List item by item (from Appendix E)
- Triggered Non-Linear Decision Gates: List item by item (from Appendix E)
7.2 Score Caliber (Must Be Explicitly Stated)
- : 0-5
- Contribution Score: Converted to /100 (e.g., A1 Compliance Risk: )
7.3 Output Format
markdown
## [Scoring Overview]
Total Score(base): XX/100 | Range(best~worst): AA~BB/100 | Overall Confidence Level: High/Medium/Low
Information Sufficiency: X/8
Commitment Gate: G0/G1/G2/G3 (Reason Summary) | Strong Commitment Upper Limit: Only Pause/Only ROM/Milestone Lockable/Quote & Schedule Lockable
[Emergency Assessment]: Yes/No
- If Yes: Emergency Adjustment Score: (+/-)N (-5~+15) | Emergency Total Score: ZZ/100 (can note if capped) | Emergency Premium: X% | Priority Conflict Affordability: Low/Medium/High (higher = more acceptable to insert) | Quality Sacrifice Level: 0-5 (higher = worse)
Triggered Score Capping:
- ...
Triggered Non-Linear Decision Gates:
- ...
|------------|---------------------------------|--------------------|--------|----------------|----------|
| Compliance Risk(30) | x/x/x | yy.y | ... | ... | ... |
| Deliverability(20) | x/x/x | yy.y | ... | ... | ... |
| Timeline Feasibility(20) | x/x/x | yy.y | ... | ... | ... |
| Price-Benefit Match(20) | x/x/x | yy.y | ... | ... | ... |
| Collaboration & Communication Risk(10) | x/x/x | yy.y | ... | ... | ... |
| [Emergency] Priority Conflict Affordability(+10) | x/x/x | yy.y | ... | ... | ... |
| [Emergency] Emergency Premium Coverage(+5) | x/x/x | yy.y | ... | ... | ... |
| [Emergency] Quality Sacrifice Level(-5, Penalty Item) | x/x/x | yy.y | ... | ... | ... |
7.4 Supplementary Scoring for Emergency Requirements (Additional Scoring Items for Emergency Scenarios)
When identified as an emergency requirement, add the following scoring items:
| Additional Dimension | Weight | Scoring Instructions |
|---|
| Priority Conflict Affordability | +10 | Affordability of the impact of insertion on existing projects, 0-5 points (higher = more affordable) |
| Emergency Premium Coverage | +5 | Whether rush fees cover additional costs, 0-5 points |
| Quality Sacrifice Level | -5 | Penalty Item: Higher score means more sacrifice, 0-5 points (reverse, 0 is best) |
Note: Emergency "channel fee/insertion cost" should be covered first through emergency premium and milestone/change/acceptance terms, rather than bypassing red lines, hard gates, and capping with "subjective score increase".
8) Phase 3: Pre-mortem (Failure Paths + Interaction Items)
Output at least 3 "most likely failure paths", and must include:
- Trigger conditions / amplification mechanism / loss type
- Early signals (observable)
- Stop-loss points (quantifiable)
- Monitoring and liability (frequency + responsible party)
- Preventive actions (contract/process/technology)
Additional Requirement: Point out at least 1-2 "interaction risks" (e.g., unclear acceptance + large final payment + extremely short timeline will directly turn delays into bad debt/disputes).
9) Phase 4: Risk Register (Make Suggestions Actionable)
Output 6-10 risk items that truly "change conclusions or terms", converge with a table:
- Risk Statement (no abstract words)
- Impact (schedule/cost/quality/legal/reputation)
- Probability (1-5) and Impact Level (1-5), used for ranking (not included in total score)
- Trigger Conditions & Early Signals
- Prevention/Mitigation (contract/process/technology)
- Stop-loss/Exit Conditions (Kill Criteria)
- Risk Ownership (Who currently bears it? How to transfer/share?)
- Risk Responsible Person/Party (Who monitors, who acts)
- Evidence/Gap/Conflict (including hit pitfall signals/conflicts)
Example Format:
markdown
## [Risk Register]
|------|----------|-----------|-----------|--------------|-------------------|----------|--------------|----------|--------|----------|
| ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... |
Recommended Caliber (For Quick Consistency):
- Probability 1-5: 1=Rarely Occurs; 3=May Occur; 5=Almost Inevitable
- Impact 1-5: 1=Negligible; 3=Requires Additional Man-Days/Delay but Controllable; 5=Major Loss/Legal or Reputational Incident
10) Phase 5: Decision + Negotiation Terms + Execution Plan
10.1 Conclusion (Four Options)
- Accept
- Negotiate Before Accepting (Must Have Condition List)
- Reject
- Pause (Assess After Supplementing, Reject Commitment Before Supplementing)
Conclusion must be consistent with: Red lines, capping, hard gates, risk ownership.
- Red line check shows
⚠️ Insufficient Information to Supplement
- Insufficient information sufficiency (<=4/8) and gaps are concentrated in acceptance/payment/scope/data compliance/key dependencies;
- Conflict items cannot be resolved with existing materials (Appendix I).
10.2 Negotiation Terms (Must Be Verifiable and Executable)
Use Appendix D of
references/assessment-standard.md
, only select conditions strongly related to Phase 3/4.
And must explicitly state: Which "failure path/risk item" each condition corresponds to.
Tip: If legal term disputes arise regarding "liability/compensation/IP/data processing/security obligations", it is recommended to involve legal counsel for review (this Skill provides a risk framework and does not replace legal advice).
10.3 Execution Plan (Turn Risks Into Plans)
At least include:
- Scope Convergence and Freeze Point (When to freeze, how to charge for changes)
- Milestone and Acceptance Rhythm (Acceptance window and default pass)
- Customer-Side Dependency Delivery List (Who/When/Format)
- Quality and Security Bottom Line (Minimum observability/backup/permissions etc.)
- Stop-Loss Triggers (Overdue payment, excessive changes, unavailable dependencies etc.)
10.4 ROM Quote and Scheduling (Optional: Explicitly Requested by User or in Quote Negotiation)
Output Requirements:
- Default to outputting ranges for (not single points).
- Each range must be bound to: Key assumptions (≤6 items) + top 3 mandatory negotiation terms + verification plan (Top 3-6).
- If information sufficiency < 6/8 or key hard gates/conflicts/capping are triggered: Clearly write
ROM For Exploration Only, Not For Locking Quotes/Schedules
Caliber and formula see
Appendix N of
references/assessment-standard.md
.
10.5 Verification Plan (Optional but Highly Recommended: Turn Key Assumptions Into Executable Verification)
Must output a verification plan if any of the following conditions are met:
- Best/worst scenarios are provided;
- Key assumptions ≥ 3 items;
- User requests "quote/schedule locking now", but information is insufficient or conflicts are unresolved.
Recommended Format:
markdown
## [Verification Plan (Top 3-6)]
|--------------|----------------------------------------|-------------------|--------|----------|
| ... | ... | ... | ... | ... |
10.6 Phased Delivery/Work Split Suggestion (Optional but Highly Recommended: Turn Uncertainty Into Milestones)
Applicable: Insufficient information/many key assumptions/customer requests "lock quotes or schedules now".
Objective: Turn "verification plan" into negotiable contract and payment nodes to avoid bearing full delivery risk at once.
Recommended Output Format:
markdown
## [Phased Delivery/Work Split Suggestion]
Phase 0 (Verification/Discovery, First Validate Key Assumptions):
- Objective: ...
- Deliverables: POC/Spike/Sample Data Verification Report + Clear Scope Boundary + Acceptance Draft + ROM Range Convergence
- Acceptance Criteria: ... (verifiable)
- Timeline: ... (range) | Cost/Billing Method: ...
- Key Stop-Loss Point: ... (If not met, can terminate/do not enter Phase 1)
Phase 1 (Delivery):
- Objective: ...
- Deliverables: ... (per frozen scope)
- Milestones and Payment: Advance payment/milestone payment/final payment (final payment ≤30%) + acceptance window + default pass for overdue
11) Phase 6: Minimum Supplementary Information Set (High-Leverage Questions)
Output 5-12 questions, sorted by "likelihood of changing the conclusion".
Each question must be marked with:
- Associated dimension/risk item
- Why it will change the conclusion (1 sentence)
Prioritize selecting/rewriting questions from
Appendix L: Minimum Supplementary Information Question Bank of
references/assessment-standard.md
, then tailor to current gaps.
Final Output Template (Unified)
markdown
# Requirement Risk Assessment Report
## [Overview]
Overall Decision: Accept/Negotiate Before Accepting/Pause/Reject
Feasibility: Accept/Negotiate Before Accepting/Pause/Reject
Value for Money: Worth Doing/Feasible but Not Worth Doing/Pause/N/A (Early Stop Due to Red Line)
Risk Mitigation (One Sentence): ...
Scenario Tags: ...
Overall Confidence Level: High/Medium/Low
Information Sufficiency: X/8 or -/8 (Not Evaluated)
Commitment Gate: G0/G1/G2/G3 (Reason Summary) | Strong Commitment Upper Limit: Only Pause/Only ROM/Milestone Lockable/Quote & Schedule Lockable
Reasons(1-3 items, must be auditable): ... (mapped to red lines/hard gates/conflicts/payment and acceptance/key dependencies)
## [Key Field Snapshot (Optional)]
|----------|---------------------------|--------------------|------------------|
| Scope Boundary | ... | ... | ... |
| Deliverable List | ... | ... | ... |
| Acceptance Criteria | ... | ... | ... |
| Payment Structure | ... | ... | ... |
| Timeline/Milestones | ... | ... | ... |
| Key Dependencies | ... | ... | ... |
| Data & Compliance Points | ... | ... | ... |
| Decision/Acceptance Responsible Person | ... | ... | ... |
## [Red Line Check]
...
## [Pitfall Signal Scanning]
...
## [Conflict Detection]
- Hit Items: ... (Evidence Anchor)
- Impact: ...
## [Key Assumption List]
- Assumption 1: ... (How it will affect conclusion/score/terms if not valid)
- Assumption 2: ...
## [Verification Plan (Top 3-6, Optional but Highly Recommended)]
|--------------|----------------------------------------|-------------------|--------|----------|
| ... | ... | ... | ... | ... |
## [Scoring Overview]
...(including score caliber, triggered Score Capping and non-linear decision gates)
## [ROM Quote and Scheduling (Optional)]
Caliber: best/base/worst (range, no single-point commitment)
...
## [Phased Delivery/Work Split Suggestion (Optional but Recommended)]
Phase 0 (Verification/Discovery): ...
Phase 1 (Delivery): ...
## [Pre-mortem: Most Likely Failure Paths]
1. ...
2. ...
3. ...
## [Risk Register]
...
## [Negotiation Terms (Must Be Met)]
- [ ] Condition ... (Mapped to: Failure Path/Risk Item)
- [ ] ...
## [Execution Plan (Turn Risks Into Actions)]
- Scope and Freeze: ...
- Milestones and Acceptance: ...
- Dependencies and Liability: ...
- Technical Bottom Line: ...
- Stop-Loss Triggers: ...
## [Required Supplementary Information (By Priority)]
1. ...
Common Failure Modes (Must Avoid)
- Total Score Covers Up Red Lines: Always execute Phase 1 + Appendix E non-linear gates
- High Score With Missing Evidence: Execute Score Capping + low confidence level + questions
- Only List Risks, No Actions: Each risk must be mapped to contract/process/technical controls
- Inconsistent Conclusions: Conclusions must be consistent with gates, payment, acceptance, liability/risk ownership
- Ignore Conflicts/Pitfall Signals: Hit items in Phase 0.25/0.5 must be reflected in scoring, confidence level, and conclusions
- Fabricate Citations: Must use user's original text, if not found write
[No Original Text Evidence Found]
- Missed Assessment of Emergency Scenarios: Identified emergency requirement but did not execute emergency scoring logic
- No Horizontal Analysis in Multi-requirement Comparison: Only assessed individually, no dimension comparison
- False Precision in Quotes/Schedules: Provide single-point quotes/schedule locking with insufficient information; should output ROM range + assumptions + verification plan
- Best/Worst Without Verification Measures: Scenario deduction without verification plan, leading to unactionable suggestions
- No Work Split for High Uncertainty: Still commit to full delivery with many key assumptions/insufficient information; should output phased delivery suggestion of Phase 0→Phase 1 (see 10.6)
- Security/Operations Left Blank: Did not supplement minimum security/operations baseline for public network/sensitive data scenarios, leading to launch accidents or liability out of control (Appendix P)
Trigger Scenarios
- User sends PRD/meeting minutes/chat records/emails
- User asks "Can we accept this requirement/should we initiate it/what are the risks/should we renew/should we do the modification"
- User says "Help me compare A and B"/"Which is better"/"Prioritization"
- User says "How was that previous project"/"Reference for similar projects"
- User says "Urgent"/"Emergency"/"Rush"/"Need it tomorrow"
- User requests "lock quotes/lock schedules/provide man-days" (need ROM range + assumptions + verification plan, split work if necessary)
- User mentions "large model/LLM/RAG/fine-tuning/Agent/generative", hoping to assess performance metrics, data authorization, cost and delivery risks
- When input is very short: Default to quick triage first, then suggest supplementing information to enter full assessment
Quick Usage Example
When the user sends requirement text, directly execute the full assessment process and output the report.
When the user asks the following questions, trigger the Skill:
- "Help me see if we can accept this requirement"
- "What are the risks of this project"
- "Assess this PRD"
- "Is this outsourcing requirement reliable"
- "Can we initiate this project"
- "Help me compare these two requirements"
- "Which has higher priority"
- "How was that previous e-commerce project"
- "This is urgent, need it tomorrow"
Stress Test Scenarios (For Verifying "True Compliance with Rules")
- Contract to Be Signed Later: User says "Do it first, sign the contract later". Expected: Hit pitfall signal; conclusion at most
Negotiate Before Accepting
- Payment on Result + Unclear Acceptance: Expected: Trigger hard gate; conclusion cannot be ; risk register and negotiation terms must cover payment and acceptance.
- Emergency Insertion: User says "Launch tomorrow/need it today/insert this task". Expected: Identify emergency; output emergency adjustment score; clearly state insertion impact and emergency premium/channel fee/sacrifice items.
- Illegal Use: User explicitly describes gray market/illegal use. Expected: Red line hit ⇒ , skip full scoring.
- Multi-requirement Mixing With Unclear Boundaries: User mixes A/B/C in one paragraph with unclear boundaries. Expected: First request user to split into paragraphs or mark boundaries; prohibit "brainstorming boundaries" before scoring and ranking.
- Historical Inquiry Without Materials: User asks "How did that previous project turn out". Expected: Mark
[Historical Materials Missing, Cannot Conduct Factual Comparison]
- Information Sufficiency <=4/8: Expected: Default to prioritizing
Pause (Reject Commitment Before Supplementing)
- Request Total Score Without Evidence: User says "Give total score first, no citations". Expected: Reject fabrication, request materials or switch to quick triage and reduce confidence level.
- Citation Restricted (No Pasting of Original Text): User says "Materials cannot be forwarded/do not quote original text". Expected: Use as evidence anchor according to Appendix M of
references/assessment-standard.md
- Strong Commitment to AI/LLM Performance: User says "100% accuracy/no errors/performance must be like ChatGPT", but does not provide evaluation dataset or acceptance criteria. Expected: Enable Appendix O; conclusion at most
Negotiate Before Accepting
- Lock Quotes/Schedules With Insufficient Information: User requests "Finalize quotes/schedules today", but information sufficiency < 6/8 or many key assumptions. Expected: Prohibit single-point commitment; output ROM range + key assumptions + verification plan, and suggest phased delivery (10.6).
- Public Network/Sensitive Data Without Security/Operations Discussion: User hopes to launch quickly and involves login/permissions/personal information. Expected: Supplement minimum security/operations baseline according to Appendix P; gaps enter risk register and negotiation terms; conclusion if necessary.
- Insist on "Full Version First" But Refuse to Supplement Information: User says "Give full report first/fill all tables first, ask questions later", and information sufficiency <= 4/8 or key fields such as red line/payment/acceptance/scope are missing. Expected: Still deliver 1st round usable version according to , and lock conclusion to or at most
Negotiate Before Accepting