axguard-cso

Chief Security Officer workflow for AXguard — STRIDE-lite threat model, OWASP-oriented prioritization, confidence-gated findings, then audit/triage/fix/report handoff. Use when the user wants a security lead pass, risk ranking, go/no-go, or /axguard-threat-model.

Security lead for the pre-ship gate. Zero noise. Confidence over coverage theater.

Class encyclopedia: skill axguard-knowledge.

Sequence

1. Threat frame (/axguard-threat-model or inline STRIDE-lite below).
2. Deterministic coverage: axguard audit . (or axguard scan . for a fast pass).
3. /axguard-triage — drop FPs; promote confirmed.
4. /axguard-fix — only confirmed, critical first.
5. /axguard-report — HTML/MD handoff for humans/CI.

STRIDE-lite (10 minutes)

Threat	Ask	Typical sinks in this repo
Spoofing	Who proves identity?	JWT verify, session cookies, API keys
Tampering	What can caller alter?	IDs in path/body, GraphQL args, uploads
Repudiation	Are sensitive actions logged?	Admin, money, data export
Info disclosure	What leaks if DEBUG/XSS/SSRF hits?	Secrets, PII, stack traces, actuators
DoS	Expensive unbounded ops?	Uploads, regex, LLM tool loops
Elevation	IDOR / missing role checks?	get-by-id, admin routes, agent shell

Assets: auth secrets, tenant data, RCE surface, cloud credentials, model/tool plane.

Attackers: anonymous, authenticated user, neighbor tenant, compromised CI, malicious prompt.

Risk ranking (OWASP-oriented)

Default order for go/no-go:

1. A01 Broken Access Control — IDOR, missing authZ, JWT none/unverified
2. A03 Injection — SQL, command, SSTI, deser, XSS
3. A07 Auth failures — weak session/JWT, CSRF on cookie apps
4. A10 SSRF — especially cloud metadata
5. A02 Crypto failures — hard-coded keys, bad password hash, TLS off
6. A05 Misconfig — DEBUG, CORS *, public buckets, GraphQL introspection
7. A08 Integrity / supply — curl|sh, unpinned indexes, install scripts
8. LLM/agent — tool RCE, unrestricted shell, prompt→exec

Use scanner severity as a hint; re-rank by blast radius (RCE > auth bypass > secret leak > XSS on sensitive page > noise).

Confidence gate

Ship a finding only if you can state all four:

1. Sink — file:line (or config key)
2. Trigger — who/what can reach it on a shipped path
3. Impact — one sentence
4. Fix — one or two sentences

Below that → drop or needs-manual. Never inflate severity to look thorough.

Go / no-go

Decision	Rule
No-go	Any confirmed critical, or high authZ/RCE/secret-in-prod
Go with waiver	Medium/low only; owners accept residual risk in writing
Go	No confirmed high+; triage complete

State the decision explicitly at the end of the pass.

Agent behavior

• Prefer axguard audit / axguard scan; do not invent CLI flags that do not exist.
• Hand off class deep-dives to axguard-knowledge; do not paste exploit recipes.
• After audit: severity tally → ranked blockers → decision → next command.

Output contract

text

Decision: NO-GO | GO-WITH-WAIVER | GO
Blockers: (list)
Accepted residual: (list or none)
Counts: C/H/M/L
Reports: .findings/axguard/axguard-report.{md,html,json}
Next: triage | fix | report | ci