cso.md

/cso — Security Audit (Chief Security Officer)

Chief Security Officer audit performing secrets archaeology, dependency supply chain analysis, CI/CD pipeline security review, LLM/AI integration security checks, and OWASP Top 10 / STRIDE threat modeling. Produces a Security Posture Report without making code changes.

---

Trigger

Event	Condition	Manual
pull_request	Automatic when the PR has the security-audit label	Comment /cso on an issue

The /cso skill is label-gated — it only triggers automatically on pull requests that carry the security-audit label. It can also be invoked manually by commenting /cso on any issue.

---

How to Use

Automatic: Add the security-audit label to a pull request. The audit runs on the next PR event.

Manual: Comment on an issue:

/cso

Arguments

Command	Mode	Description
/cso	Daily (default)	Full audit, all phases, 8/10 confidence gate for zero noise
/cso --comprehensive	Monthly deep scan	All phases, 2/10 confidence bar — surfaces more findings
/cso --infra	Cloud and deploy only	Phases 0–6, 12–14
/cso --code	Code only	Phases 0–1, 7, 9–11, 12–14
/cso --skills	Skill supply chain	Phases 0, 8, 12–14
/cso --diff	Branch changes only	Combinable with any scope flag
/cso --supply-chain	Dependency audit	Phases 0, 3, 12–14
/cso --owasp	OWASP Top 10 only	Phases 0, 9, 12–14
/cso --scope auth	Focused domain	Audit a specific domain (e.g., auth, payments)

Scope flags are mutually exclusive. If multiple scope flags are passed, the skill errors immediately — security tooling must never silently ignore user intent.

--diff is combinable with any scope flag and with --comprehensive.

---

What It Does

The /cso skill runs a multi-phase security audit. It does not make code changes — it produces a read-only Security Posture Report with findings, severity ratings, and remediation plans.

Audit Phases

Phase	Name	What It Checks
0	Architecture Mental Model	Builds a security-relevant map of the codebase
1	Attack Surface Mapping	Entry points, exposed APIs, trust boundaries
2	Secrets Archaeology	Scans git history for exposed keys, tokens, credentials
3	Dependency Supply Chain	Known CVEs, typosquatting, unmaintained packages
4	CI/CD Pipeline Security	Workflow permissions, artifact integrity, secret exposure in logs
5	Cloud & Network Review	Cloud config, network exposure, storage permissions
6	Webhook Security	Inbound webhook validation, signature verification
7	LLM/AI Security	Prompt injection, output trust boundaries, data leakage
8	Skill Supply Chain	Third-party skill/plugin security review
9	OWASP Top 10	Injection, broken auth, SSRF, security misconfiguration, etc.
10	STRIDE Threat Model	Spoofing, Tampering, Repudiation, Information Disclosure, DoS, Elevation
11	Data Classification	PII handling, encryption at rest/transit, retention policies
12	False Positive Filtering	Removes noise with confidence-based filtering
13	Findings Report	Structured report with severity, evidence, and remediation
14	Trend Tracking	Compares against previous audits for progress/regression

Two Confidence Modes

• Daily mode (default): 8/10 confidence gate — only surfaces high-confidence findings for zero noise in routine audits.
• Comprehensive mode (--comprehensive): 2/10 confidence bar — surfaces everything for monthly deep scans. More findings, more noise, more coverage.

---

Example Output

## Security Posture Report — PR #55

**Mode:** Daily (8/10 confidence gate)
**Scope:** Full audit

### Secrets Archaeology
✅ No exposed secrets found in git history

### Dependency Supply Chain
⚠️ `lodash@4.17.20` — known prototype pollution (CVE-2021-23337), upgrade to 4.17.21+
⚠️ `jsonwebtoken@8.5.1` — algorithm confusion vulnerability, upgrade to 9.0.0+

### CI/CD Pipeline
✅ No secret exposure in workflow logs
⚠️ `.github/workflows/deploy.yml` — uses `pull_request_target` with checkout of PR head (potential code injection)

### OWASP Top 10
🔴 **A03:2021 Injection** — unsanitized user input in `src/search.ts:89`
⚠️ **A07:2021 Authentication** — session token not rotated after privilege change

### Threat Model (STRIDE)
| Threat | Risk | Notes |
|--------|------|-------|
| Spoofing | Low | Auth uses standard JWT with proper validation |
| Tampering | Medium | API accepts unsigned payloads on 2 endpoints |
| Repudiation | Low | Audit logging present |
| Info Disclosure | Low | No PII in logs |
| Denial of Service | Low | Rate limiting configured |
| Elevation of Privilege | Low | RBAC properly enforced |

### Remediation Plan
1. 🔴 Fix injection in `src/search.ts:89` — parameterize query (est. 15 min)
2. ⚠️ Upgrade `lodash` to 4.17.21+ (est. 5 min)
3. ⚠️ Rotate session token on privilege change (est. 30 min)

---

Configuration

In config.json:

{
  "cso": {
    "enabled": true,
    "trigger": "pull_request",
    "labelGated": true,
    "label": "security-audit"
  }
}

Field	Type	Description
enabled	boolean	Set to false to disable the skill
trigger	string	GitHub event type — pull_request
labelGated	boolean	When true, the skill only triggers if the specified label is present
label	string	The label that gates activation — security-audit

---

Requirements

Requirement	Details
Browser	❌ Not needed
Model	Default model from config (gpt-5.4) recommended
Permissions	admin, maintain, or write on the repository
Tools	Bash, Read, Grep, Glob, Write, Agent, WebSearch

---

Results

Security audit outcomes are persisted to:

.github-gstack-intelligence/state/results/security/pr-{N}.json

Each report includes findings with severity, evidence, and remediation plans. Trend tracking compares against previous audit runs to show security posture changes over time.

---

Related Files

File	Purpose
skills/cso.md	Skill prompt definition
skills/references/cso-acknowledgements.md	Credits for security research sources that informed the skill
config.json	Skill enablement and trigger configuration
lifecycle/router.ts	Event routing — maps GitHub events to skills

---

See Also

• /review — PR code review (complementary — /review finds code issues, /cso finds security issues)

---

← Back to Command Reference