The Quartz Tribunal — AI-Driven Security Audit of Zenzic v0.7.0

"Assediato. Riparato. Sigillato."

Before a tool that guarantees documentation integrity can call itself stable, it must first face an adversary that knows exactly where to look for cracks. This is the account of that encounter.

---

The Objective

The Safe Harbor Guarantee is not a marketing claim. It is an engineering contract: if Zenzic exits 0 and emits the Sentinel Seal, the documentation source is structurally sound, link-complete, and secret-free. That contract can only be trusted if the engine itself has been tested by an adversary, not just by its authors.

CEO Directive 189 — "The Inquisitor's Mandate" — commissioned a structured security audit of Zenzic v0.7.0 prior to the stable release. Three independent AI teams were assembled. The mandate: find every crack in the Safe Harbor, fix it, and certify the result.

:::note Baseline 1,301 tests at audit time (D094). 1,307 after D095 sealed all remaining Known Limitations. Zero failures before, zero after. The audit made Zenzic stronger than it entered. :::

---

The Methodology — The Magistrate Model

The audit used a three-team structure designed to eliminate confirmation bias:

Team	Role	Bias
🔴 Red Team	Adversarial — find what the Shield misses	Maximum aggression
⚪ Blue Team	Defensive — verify that RULE invariants hold	Maximum rigor
🟣 Purple Team	Ethical Arbiter — separate real bugs from noise	Maximum objectivity

No team was permitted to declare a finding without Purple Team review. A finding that would not survive objectivity scrutiny was discarded as noise. Only findings that survived the Magistrate Model entered the Bug Registry.

---

The Siege Chronicle

D1 — @site/ Path Traversal (Docusaurus)

Vector: [config](@site/../../zenzic/pyproject.toml) — a link attempting to escape the documentation root via Docusaurus's @site/ alias mechanism.

Result: Z202 PATH_TRAVERSAL detected. The InMemoryPathResolver resolved the path outside the authorized perimeter and fired correctly.

However, a severity bug was discovered: Z202 was mapped to "error" severity → exit 1 (suppressible by --exit-zero). Per RULE R4, exit codes 2 and 3 are never suppressible. PATH_TRAVERSAL (Z202) must exit 3.

Fix applied: _check.py _to_findings() — the severity condition was expanded to include PATH_TRAVERSAL alongside PATH_TRAVERSAL_SUSPICIOUS.

🔴 Red Team blocked · ⚪ Blue Team severity bug sealed · Exit 3. Non-suppressible. ✅

---

D3 — JSX Prop Shield Bypass

Vector: <MyComponent apiKey="ghp_<EXAMPLE_TOKEN>" /> — a GitHub PAT embedded in a JSX component attribute, not in raw Markdown prose.

Hypothesis: Shield might skip JSX prop values since they are not prose.

Result: Shield detected ghp_ immediately. Exit 2. SECURITY BREACH DETECTED.

Shield Pass 1A uses enumerate(fh, start=1) on the raw file bytes. It never skips lines based on content type. JSX props are plain text to the file scanner. RULE R9 holds unconditionally.

🔴 Red Team blocked · RULE R9 confirmed. ✅

---

S2 — Base64-Encoded Frontmatter Secret → Sealed in v0.7.0

Vector: YAML frontmatter field api_token: Z2hwXzEy[…base64…] — — a base64-encoded GitHub PAT stored as a configuration value in frontmatter.

D094 result: Shield exited 0. This was documented as Known Limitation KL-001.

D095 resolution: Speculative Base64 decoding implemented in shield.py (CEO-194). The decoder extracts candidate tokens from every normalised line, decodes each as UTF-8, and re-scans the decoded text through the full _SECRETS pattern table. The canonical test vector (Z2hwXzEy[…base64…] → ghp_[TOKEN]) now triggers Z201 and exits 2.

False-positive guard: A minimum token length of 20 characters (before decoding) prevents incidental short base64 strings from triggering spurious findings.

🔴 Red Team attack vector SEALED. KL-001 closed. Exit 2 confirmed. ✅

---

The Sealed Cracks

ID	Area	Finding	Fix	Status
BUG-CEO189-01	cli/_check.py	Z202 exits 1 instead of 3 (suppressible)	Severity condition expanded to PATH_TRAVERSAL	✅ Sealed
BUG-CEO189-02	suppression-policy.mdx EN+IT	Z106 mislabeled ALT_TEXT_MISSING	Corrected to CIRCULAR_LINK	✅ Sealed
BUG-CEO189-03	structural-integrity.mdx EN+IT	Z106 in Dimension 3 (should be Z403)	Code + name corrected	✅ Sealed
BUG-CEO189-04	health-metrics.mdx EN+IT	Z106 penalty name wrong	Replaced with correct CIRCULAR_LINK / Z403	✅ Sealed
BUG-CEO189-05	Structural Map RULE R23	Z106 incorrectly described as Alt Text	RULE R23 corrected	✅ Sealed
BUG-CEO194-B64	shield.py	S2 Red Team vector (Base64 bypass)	Speculative Base64 decoder added	✅ Sealed
KL-002	resolver.py	False-positive PathTraversal on case-insensitive filesystems (APFS/NTFS)	os.path.normcase applied to boundary comparison	✅ Fixed (portability)

7 findings sealed across 2 sprints. 0 open security issues.

---

The Verdict

Certification Metrics

Metric	Value
Tests passing	1,307 (Python 3.11 / 3.12 / 3.13)
Code coverage	80.28% (≥ 80% required)
Interactive Acts in zenzic lab	20 (Acts 0–19)
Masterclass length	1,525 lines
Red Team attacks launched	3 (D1, D3, S2)
Red Team attacks blocked (D094)	2/3
Red Team attacks sealed (D095)	3/3 (S2 sealed)
Critical security fixes	2 (Z202 exit code, Base64 decoder)
Documentation bugs fixed	5 (across 7 files)
Open security issues	0

Invariant Verification

Invariant	Test	Result
RULE R4 — Exit 2/3 never suppressible	--exit-zero on Z202	✅ Exit 3, not suppressed
RULE R9 — Shield scans raw content	JSX prop ghp_	✅ Detected, exit 2
RULE R8 — Zero subprocesses	Full code audit	✅ No subprocess, no os.system
RULE R3 — Finding codes mandatory	All findings carry Zxxx	✅ codes.normalize() confirmed
ADR-013 — Z2xx inviolable	_INVIOLABLE_CODES guard	✅ Present in rules.py
RULE R20 — Machine silence	SARIF format, no Rich output	✅ Gated at every call site

The Quartz Certification

:::note Quartz Tribunal Certification — Zenzic v0.7.0 CERTIFIED. The Tribunal reviewed 3 attack vectors, sealed all open cracks, and verified all invariants. Exit 0. Sentinel Seal. ✨

"The Safe Harbor is not a promise. It is a proof." :::