What data for this category looks like
This category’s core documents include four types: test authorization letters, penetration test reports, vulnerability reproduction records, and rectification suggestions. Data sources cover target system logs, public vulnerability databases, automated scanner results, and red team manual verification records. Data updates follow cycles tied to individual projects or major versions. The structure of each document flexibly adjusts based on the test scope. Core fields include authorization number, vulnerability severity, attack path, remediation plan, and other relevant items. Token counts per individual document vary widely, with no fixed, uniform length limit.
Constraints imposed by these characteristics on context and token handling
Documents in this category have diverse types and widely varying token counts. Confidential documents such as authorization letters require full retention of all content; truncation may lead to legal authorization disputes. Data sources are scattered, requiring integration of target system logs, vulnerability database data, scan results, and manual records into the context window. Reasonable allocation of token proportions for content from different sources is needed. Update cycles adjust flexibly per project, so the content loaded into the context each time is not fixed, making fixed configuration unreliable. Error consequences are severe: context truncation may lead to missed vulnerabilities or business interruptions. Unlike vulnerability scanning services, this category requires integration of unstructured content from manual verification, placing higher demands on context coherence. As such, context and token configurations must balance completeness and resource efficiency.
The following configuration instructions are based on FastGPT v4.17.0.
How to set configurations
| Configuration Item | Recommended Range | Rationale |
|---|---|---|
maxContext | 8000-15000 token | Adapts to the total token volume of multiple documents for a single project, reserves redundancy to avoid truncation of core confidential documents such as authorization letters |
quoteMaxToken | 1200-1800 token | Matches the length of a single vulnerability reproduction record, fully retains details of attack paths and exploit chains |
| Chunk Length | 800-1000 characters | Adapts to the paragraph structure of penetration test reports, balances context coherence and token usage |
| Overlap Length | 100-150 characters | Adapts to cross-paragraph associations in vulnerability verification records, prevents loss of key verification steps |
| Context Recall Limit | Top 6 entries | Controls the total token count of integrated content, avoids exceeding model window limits |
Three common misconfiguration issues
- Symptom: A 400 error is returned when uploading a complete penetration test report, with a prompt indicating context length exceeded. Cause: No preprocessing was performed on the long document, and loading the full content directly exceeds the set
maxContextvalue. - Symptom: Attack path descriptions in vulnerability reproduction records are truncated, and key basis for risk rating results is missing. Cause: The
quoteMaxTokensetting is lower than the token count of a single reproduction record, leading to truncation of core content. - Symptom: Some scanner log content is lost when integrating scanner results and manual verification records. Cause: The chunk length is set too large, causing short log entries to fail to load correctly into the context window.
How to verify correct configuration
- Upload a single complete test authorization letter, check that the returned result includes all authorization number and authorization scope content.
- View context loading logs, confirm that the actual loaded token count does not exceed the set
maxContextvalue. - Trigger a context call for vulnerability reproduction records, check that full descriptions of attack steps are retrieved completely.
- Adjust the chunk length, compare context recall results across different configurations, confirm that cross-paragraph associated information is retained.
The values provided are common starting points and should be measured against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-10-19.