Security Review Skill
Use this skill before closing a task, approving a proof bundle, or reviewing a risky diff.
Three-Tier Boundary System
Apply this explicit boundary system during review:
Always Do (No Exceptions)
- Validate all external input at system boundaries
- Parameterize all database queries
- Encode output to prevent injection
- Use HTTPS for external communication
- Hash passwords with strong algorithms
- Set security headers (CSP, HSTS, X-Frame-Options)
- Use httpOnly, secure, sameSite cookies for sessions
- Run dependency audits before releases
Ask First (Requires Human Approval)
- Adding or changing authentication flows
- Storing new categories of sensitive data (PII, payment info)
- Adding new external service integrations
- Changing CORS configuration
- Adding file upload handlers
- Modifying rate limiting or throttling
- Granting elevated permissions or roles
Never Do
- Never commit secrets to version control
- Never log sensitive data (passwords, tokens, full card numbers)
- Never trust client-side validation as a security boundary
- Never disable security headers for convenience
- Never use eval() or innerHTML with user-provided data
- Never store sessions in client-accessible storage
- Never expose stack traces or internal error details to users
Review flow
- Call
ck_context to load the domain pack, risk tier, open findings, instruction hierarchy, and design-drift signals.
- Check if agent frameworks are detected in workspace context. If agent frameworks (LangGraph, CrewAI, AutoGen, LangChain) are present, activate the agent-pattern-verification skill for additional agent-specific checks.
- Run
ck_validate on the relevant code or config slices, including trust-boundary metadata when the proposed action was influenced by web, tool, skill, or mixed-provenance content.
- Apply the three-tier boundary system to classify each finding: always-do, ask-first, or never-do.
- Walk the review checklist in references/review-checklist.md.
- Persist any missed issue with
ck_finding, including boundary tier classification in metadata.
- If external security or regression systems produce exploit or browser evidence, record that through
ck_regression_result when it affects release readiness.
- Summarize blockers, warnings, and follow-up proof requirements, grouped by boundary tier.
Agent Pattern Integration
When agent frameworks are detected in the workspace context, this skill automatically includes agent-specific pattern checks:
- Loop Safety: Detects
while True without break, recursive calls without depth limits
- Retry Limits: Validates retry decorators have explicit stop conditions
- Tool Registry: Cross-references tool definitions with prompt references
- Context Size: Monitors system prompt and tool description token counts
- Graph Cycles: Analyzes LangGraph graphs for unreachable END nodes
These checks complement the baseline security review with agent-specific anti-pattern detection.
Additional resources
- For anti-rationalization patterns to help recognize and reject security shortcuts, see references/anti-rationalization-patterns.md.
1---2name: security-review3description: Run a structured security review before marking a task done. Services as the single audit skill for OWASP, baseline pack, domain-pack (HR/legal/marketing/sales/real-estate/government/insurance/logistics/manufacturing/e-commerce/nonprofit), and compliance (GDPR/SOC2/HIPAA/policy packs) — subsumes the deprecated `compliance-audit` and `domain-audit` aliases. Use this for code, config, architecture, or release reviews; the domain/compliance layer is triggered by the session's active domain pack and data flows.4license: Apache-2.05---67# Security Review Skill89Use this skill before closing a task, approving a proof bundle, or reviewing a risky diff.1011## Three-Tier Boundary System1213Apply this explicit boundary system during review:1415### Always Do (No Exceptions)16- Validate all external input at system boundaries17- Parameterize all database queries18- Encode output to prevent injection19- Use HTTPS for external communication20- Hash passwords with strong algorithms21- Set security headers (CSP, HSTS, X-Frame-Options)22- Use httpOnly, secure, sameSite cookies for sessions23- Run dependency audits before releases2425### Ask First (Requires Human Approval)26- Adding or changing authentication flows27- Storing new categories of sensitive data (PII, payment info)28- Adding new external service integrations29- Changing CORS configuration30- Adding file upload handlers31- Modifying rate limiting or throttling32- Granting elevated permissions or roles3334### Never Do35- Never commit secrets to version control36- Never log sensitive data (passwords, tokens, full card numbers)37- Never trust client-side validation as a security boundary38- Never disable security headers for convenience39- Never use eval() or innerHTML with user-provided data40- Never store sessions in client-accessible storage41- Never expose stack traces or internal error details to users4243## Review flow44451. Call `ck_context` to load the domain pack, risk tier, open findings, instruction hierarchy, and design-drift signals.462. Check if agent frameworks are detected in workspace context. If agent frameworks (LangGraph, CrewAI, AutoGen, LangChain) are present, activate the agent-pattern-verification skill for additional agent-specific checks.473. Run `ck_validate` on the relevant code or config slices, including trust-boundary metadata when the proposed action was influenced by web, tool, skill, or mixed-provenance content.484. Apply the three-tier boundary system to classify each finding: always-do, ask-first, or never-do.495. Walk the review checklist in [references/review-checklist.md](references/review-checklist.md).506. Persist any missed issue with `ck_finding`, including boundary tier classification in metadata.517. If external security or regression systems produce exploit or browser evidence, record that through `ck_regression_result` when it affects release readiness.528. Summarize blockers, warnings, and follow-up proof requirements, grouped by boundary tier.5354## Agent Pattern Integration5556When agent frameworks are detected in the workspace context, this skill automatically includes agent-specific pattern checks:5758- **Loop Safety**: Detects `while True` without break, recursive calls without depth limits59- **Retry Limits**: Validates retry decorators have explicit stop conditions60- **Tool Registry**: Cross-references tool definitions with prompt references61- **Context Size**: Monitors system prompt and tool description token counts62- **Graph Cycles**: Analyzes LangGraph graphs for unreachable END nodes6364These checks complement the baseline security review with agent-specific anti-pattern detection.6566## Additional resources6768- For anti-rationalization patterns to help recognize and reject security shortcuts, see [references/anti-rationalization-patterns.md](references/anti-rationalization-patterns.md).
aryaminus