Security assessment and data handling practices for AISEACT.

AISEACT is a documentation-based skill (also known as a "prompt-based skill" or "instruction skill"). It contains no executable code, no external dependencies, and no network endpoints.

Security Status: ✅ Low Risk

• No code execution
• No data persistence
• No credential requirements
• No external service dependencies (beyond standard search APIs)

1. Skill Loading: AI platform reads SKILL.md and reference documents
2. User Request: You explicitly request AISEACT methodology (or have enabled autonomous mode)
3. Methodology Application: AI follows the documented guidelines for search strategy
4. Standard Search: AI uses platform's existing search tools (no custom search implementation)
5. Output: AI provides answer with source citations following the methodology

User Query
    ↓
AI Agent (with AISEACT skill loaded)
    ↓
Search Strategy (methodology from SKILL.md)
    ↓
Platform Search API (standard search tools)
    ↓
Search Results
    ↓
Analysis & Filtering (methodology guidelines)
    ↓
Response to User

• ❌ No personal data collection
• ❌ No browsing history access
• ❌ No file system access (beyond reading skill files)
• ❌ No environment variable access
• ❌ No credential access

Description: The skill includes source quality guidelines that may filter certain sources.

Severity: Low (behavioral/policy risk, not security risk)

Mitigation:

• Skill is optional (manual by default)
• User can override filtering per-query
• Source lists are transparent and documented
• User retains full control over final output

User Control:

"包含 [source]" - Include specific source regardless of rating
"显示所有来源" - Show all found sources without filtering
"不用AISEACT" - Skip methodology entirely

Description: Source ratings may reflect Western-centric or mainstream biases.

Severity: Low (methodological limitation, not security risk)

Mitigation:

• Acknowledged in TRUST.md
• Source evaluation criteria are transparent
• User can request any source regardless of rating
• Configuration options to adjust strictness

Description: Earlier versions had mismatch between "ALWAYS USE" language in SKILL.md and actual registry configuration.

Severity: Fixed ✅

Resolution:

• SKILL.md now uses optional, user-controlled language
• "ALWAYS USE" replaced with explicit user request patterns
• Clear distinction between manual (default) and autonomous (opt-in) modes

Description: Skill files could theoretically be modified to inject malicious instructions.

Severity: Low

Mitigation:

• Install from trusted source (official repository)
• Review SKILL.md before installation
• No executable code to exploit
• AI platform's safety filters still apply

1. Verify source: Install from official repository

   git clone https://github.com/stephenlzc/aiseact.git

2. Review files: Check SKILL.md and reference documents before installing

3. Manual mode: Start with manual invocation (default) before enabling autonomous mode

1. Understand methodology: Review TRUST.md to understand source evaluation approach

2. Test outputs: Verify a few outputs meet your expectations before relying on the skill

3. Use overrides: When in doubt, use override commands:

   • "显示所有来源" to see unfiltered results
   • "包含 [source]" to include specific sources

4. Check source ratings: Review references/unreliable-sources.md to understand rating criteria

1. Default to manual: Keep disable-model-invocation: true (default) unless you specifically want autonomous behavior

2. Review autonomous outputs: If enabling autonomous mode, monitor initial outputs for unexpected behavior

Overall Security Rating: ✅ Low Risk / Safe

Primary Considerations:

• Behavioral: Source filtering methodology (user-controllable)
• Methodological: Potential bias in source ratings (acknowledged)
• Not a technical security risk

If you discover a security issue with AISEACT:

1. Do not open a public issue
2. Contact the maintainer via the repository's security contact
3. Provide details about the potential vulnerability

This security information is current as of March 2026.