Research-Stack/audit/infrastructure_audit.md

Infrastructure Audit Report

Repository: https://github.com/allaunthefox/Research-Stack
Branch: main
Scope: 4-Infrastructure/, 5-Applications/, 6-Kernel-Shim/, scripts/
Audit Date: 2026-06-02

---

Summary

Category	Count	Severity
Total scoped files	2,725	-
Secrets found	3	1 CRITICAL, 1 HIGH, 1 MEDIUM
Security issues	6	2 HIGH, 3 MEDIUM, 1 LOW
Dead code functions/regions	3	3 LOW
Broken imports	2	1 MEDIUM, 1 LOW
Path mismatches	3	1 MEDIUM, 2 LOW
Build problems	2	1 MEDIUM, 1 LOW
Config drift	2	1 MEDIUM, 1 LOW

---

CRITICAL Issues

C1. Hardcoded Wolfram Alpha API Key (SECRET)

• File: 4-Infrastructure/shim/wolfram_verify.py
• Line: APPID = "HYJE3R3R63"
• Severity: CRITICAL
• Description: A live Wolfram Alpha API key is hardcoded in plaintext and committed to version control. This key is used to make API calls to api.wolframalpha.com. Anyone with repository access can extract and abuse this key, incurring API charges and potentially violating terms of service.
• Evidence:

  APPID = "HYJE3R3R63"
  # ...
  url = f"http://api.wolframalpha.com/v2/query?appid={APPID}&input=..."
  

• Remediation:
  1. Immediately revoke the exposed API key at Wolfram Alpha developer portal
  2. Replace with: APPID = os.environ.get("WOLFRAM_APPID", "")
  3. Add WOLFRAM_APPID to .gitignore and environment setup docs
  4. Add secret scanning to CI to prevent future commits
  5. Rotate any other keys that may have been exposed

---

HIGH Severity Issues

H1. Remote SSH Command Injection via shell=True

• File: 4-Infrastructure/shim/verify_all_shims.py
• Lines: subprocess.run(cmd, shell=True, ...) (multiple instances)
• Severity: HIGH
• Description: The verification script constructs SSH commands via string concatenation and executes them with shell=True. While the current inputs appear hardcoded, the pattern is dangerous. The run_remote() function builds: ssh {user}@{ip} "{cmd}" where cmd comes from the SHIMS list but could be tampered with.
• Evidence:

  def run_remote(cmd):
      ssh_cmd = f"ssh {CUPFOX_USER}@{CUPFOX_IP} \"{cmd}\""
      res = subprocess.run(ssh_cmd, shell=True, capture_output=True, text=True)
  

• Remediation:
  • Use paramiko or asyncssh library for SSH connections
  • If subprocess is required, use shell=False with list args
  • Validate all command components against an allowlist

H2. Hardcoded Remote Infrastructure Credentials

• File: 4-Infrastructure/shim/verify_all_shims.py
• Lines: CUPFOX_IP = "46.232.249.226", CUPFOX_USER = "root"
• Severity: HIGH
• Description: Production infrastructure IP address and privileged username (root) are hardcoded. Combined with SSH key-based authentication (not visible in file but implied by SSH commands), this exposes production compute node details.
• Evidence:

  CUPFOX_IP = "46.232.249.226"
  CUPFOX_USER = "root"
  

• Remediation:
  • Move to environment variables: CUPFOX_IP, CUPFOX_USER
  • Use non-privileged user for remote execution
  • Document SSH key management in secure vault

---

MEDIUM Severity Issues

M1. SQL Injection Pattern in Eigensolid Data Shim

• File: 4-Infrastructure/shim/ingest_eigensolid_data.py
• Function: generate_crossing_weights_insert(), generate_eigensolid_snapshot_insert(), generate_braid_strand_insert()
• Severity: MEDIUM
• Description: SQL generator functions use f-string interpolation to construct INSERT statements. While the current uuid.uuid4() values are safe, the pattern is dangerous if extended to user-provided data. Note: the insert_eigensolid_data() function correctly uses parameterized queries - there is an inconsistency.
• Evidence:

  def generate_crossing_weights_insert(...):
      return f"INSERT INTO ene.crossing_weights ... VALUES ('{cw_id}', '{recipe_id}', ...)"
  

• Remediation:
  • Remove unsafe SQL generator functions or use parameter placeholders (%s)
  • Ensure all SQL generation uses parameterized queries consistently

M2. Broken Import Path Manipulation

• File: 4-Infrastructure/shim/qaoa_adapter.py
• Lines: sys.path.insert(0, ...), from q16 import ...
• Severity: MEDIUM
• Description: Module uses sys.path.insert() to find q16.py and braid_search.py. If these files are not present in the expected relative locations, the import will fail at runtime. Only partial graceful fallback exists.
• Evidence:

  _sys.path.insert(0, str(Path(__file__).resolve().parent.parent / "lib"))
  _sys.path.insert(0, str(Path(__file__).resolve().parent))
  from q16 import Q16_SCALE, from_q16, to_q16
  from braid_search import _q16_signed
  

• Remediation:
  • Package modules properly with __init__.py
  • Add try/except ImportError with fallback implementations

M3. Kernel Module const-Cast Undefined Behavior

• File: 6-Kernel-Shim/pist_compress.c
• Function: pist_init_galois_table()
• Severity: MEDIUM
• Description: The function casts away const from galois_mul_table to write to it at runtime: (u8 *)galois_mul_table. This is undefined behavior in C and potentially a security concern in kernel code.
• Evidence:

  static const u8 galois_mul_table[256] = { ... };
  static void pist_init_galois_table(void) {
      u8 *tbl = (u8 *)galois_mul_table;  // const cast away
      for (...) tbl[i] = p;
  }
  

• Remediation:
  • Remove const qualifier from galois_mul_table
  • Or use a non-const working copy

M4. Config Drift - Extracted Components Still Referenced

• File: 4-Infrastructure/AGENTS.md
• Severity: MEDIUM
• Description: AGENTS.md states that kubernetes/, k3s-flake/, kube/, shim/ray-actors/, shim/vcn_*.py, shim/hermes/ were extracted to distributed-compute-fabric repository. However, references to these paths may still exist in code and scripts, causing confusion.
• Remediation:
  • Audit all remaining references to extracted components
  • Update cross-references to point to new repository
  • Add deprecation notices where appropriate

M5. Path Mismatch in Verification Script

• File: 4-Infrastructure/shim/verify_all_shims.py
• Line: local_path = os.path.join("4-Infrastructure/shim", shim)
• Severity: MEDIUM
• Description: Uses relative path "4-Infrastructure/shim" which assumes script is run from repo root. If run from elsewhere, files won't be found.
• Remediation:

  script_dir = Path(__file__).resolve().parent
  local_path = script_dir / shim
  

---

LOW Severity Issues

L1. Kernel Module Non-Destructive Statistics Read

• File: 6-Kernel-Shim/pist_neuromorphic.c
• Function: stats_show()
• Severity: LOW
• Description: The stats_show() function destructively reads byte_freq by zeroing entries during the "top 10" loop. This is not thread-safe and corrupts statistics for concurrent readers.
• Evidence:

  st->byte_freq[max_idx] = 0; /* zero out for next iter (destructive!) */
  

• Remediation: Copy array locally before processing. Use spin_lock_irqsave() to protect the read.

L2. Dead Code in Kernel Compression Module

• File: 6-Kernel-Shim/pist_compress.c
• Functions: pist_shifter_hachimoji(), pist_shifter_aegis(), pist_entropy_approx()
• Severity: LOW
• Description: Three shifter functions (Hachimoji, AEGIS, Natural DNA) and the entropy approximation function are defined but never called in the compression/decompression paths. Only pist_shifter_mirror is actually used in pist_compress_page().
• Remediation: Remove unused functions or wire them into the compression engine via the shifter chain bitmask.

L3. Dead Bit Flags in pist_compress.c

• File: 6-Kernel-Shim/pist_compress.c
• Severity: LOW
• Description: Eight shifter bit flags are defined but only PIST_SHIFT_PIST_MIRROR is checked in the compression function.
• Evidence:

  #define PIST_SHIFT_HACHIMOJI    BIT(0)  // unused
  #define PIST_SHIFT_AEGIS        BIT(1)  // unused
  #define PIST_SHIFT_NATURAL_DNA  BIT(2)  // unused
  #define PIST_SHIFT_PIST_MIRROR  BIT(3)  // used
  #define PIST_SHIFT_PNA          BIT(4)  // unused
  #define PIST_SHIFT_LNA          BIT(5)  // unused
  #define PIST_SHIFT_PRION        BIT(6)  // unused
  #define PIST_SHIFT_GALOIS       BIT(7)  // unused
  

• Remediation: Implement multi-shifter chain or remove unused flags.

L4. Missing Output Directory Handling

• File: 5-Applications/scripts/rgflow_gpu_pipeline.py
• Line: save_results(flat, Path("out"))
• Severity: LOW
• Description: Hardcoded output path "out" may not exist. While out_dir.mkdir() is called inside save_results(), the path is relative to cwd.
• Remediation: Use configurable output path with mkdir(parents=True, exist_ok=True).

L5. Heavy Monolithic requirements.txt

• File: requirements.txt
• Severity: LOW
• Description: Single requirements.txt includes GPU (wgpu), quantum (qiskit, cirq, perceval-quandela), distributed compute (ray), and database dependencies. Many will fail on systems without specialized hardware.
• Remediation:

  requirements.txt           # core only
  requirements-gpu.txt       # wgpu, CUDA
  requirements-quantum.txt   # qiskit, cirq, perceval
  requirements-db.txt        # psycopg2, boto3
  

L6. AGENTS.md Storage Agent Description Drift

• File: 4-Infrastructure/AGENTS.md
• Severity: LOW
• Description: AGENTS.md describes detailed trigger model for storage_agent.py with specific Q16_16 thresholds and receipt schema. The actual implementation may differ from documentation.
• Remediation: Add doc tests or sync documentation with code via automated extraction.

L7. Potentially Unused Cloudflare WASM Functions

• File: 4-Infrastructure/cloudflare/src/lib.rs
• Functions: reset(), derive_scalar()
• Severity: LOW
• Description: reset() and derive_scalar() are defined but may not be called from the JavaScript entry point (index.js).
• Remediation: Verify usage in index.js or remove if confirmed unused.

---

Per-Component Breakdown

4-Infrastructure/ (1,420 Python files, 142 Rust files, 18 C files, 68 Shell files, 125 YAML files)

Component	Files	Critical Issues	Warnings	Notes
shim/*.py	~50	1 (wolfram key)	4	Core adapter shims
storage/	~10	0	1	Storage agent subprocess calls
cloudflare/	3	0	1	WASM edge functions
kernel/	2 C files	0	3	Kernel modules have dead code, const cast
hardware/	~20	0	0	FPGA/sparkle hardware bring-up
NoDupeLabs/	~50	0	0	Third-party package

5-Applications/ (142 Rust files)

Component	Files	Critical Issues	Warnings	Notes
parquet_compressor/	~30	0	1	Cargo.toml dependency versions
scripts/	~20	0	1	Path handling in GPU pipeline

6-Kernel-Shim/ (2 C files mirrored from 4-Infrastructure/kernel/)

Component	Files	Critical Issues	Warnings	Notes
pist_compress.c	1	0	2	Dead code, const cast
pist_neuromorphic.c	1	0	1	Destructive stats read

Note: The C files in 6-Kernel-Shim/ appear to be mirrors of 4-Infrastructure/kernel/ files. Ensure they are kept in sync or use symlinks/submodules to avoid divergence.

---

Security Analysis Details

Python Security Patterns Found

Pattern	Files	Risk Level	Notes
subprocess.run(shell=True)	verify_all_shims.py	HIGH	Remote SSH commands
subprocess.run()	storage_agent.py	MEDIUM	Calls to backup scripts
f-string SQL construction	ingest_eigensolid_data.py	MEDIUM	Generator functions
eval() / exec()	None found	-	Clean
pickle.loads()	None found	-	Clean
input()	None found	-	Clean
Hardcoded API key	wolfram_verify.py	CRITICAL	Wolfram Alpha APPID

Rust Safety Analysis

File	unsafe Blocks	Notes
cloudflare/src/lib.rs	0	Clean WASM bindings
parquet_compressor/src/*.rs	TBD (not in scope of read)	Likely has GPU unsafe code

C Kernel Safety Analysis

File	Issue	Severity
pist_compress.c	const cast away	MEDIUM
pist_compress.c	int_sqrt() usage (kernel API)	LOW (verify availability)
pist_neuromorphic.c	Non-atomic stat read	LOW
pist_neuromorphic.c	PAGE_SIZE buffer in sysfs show	LOW (potential overflow)

---

Build System Analysis

Cargo.toml (parquet_compressor)

[dependencies]
parquet = { version = "58.3", default-features = false, features = ["arrow", "zstd"] }
arrow = "58.3"
serde = { version = "1.0", features = ["derive"] }
serde_json = "1.0"
rayon = "1.8"
clap = { version = "4.4", features = ["derive"] }
indicatif = "0.18"
byteorder = "1.5"
sysinfo = "0.30"
tokio = { version = "1.35", features = ["sync", "rt", "rt-multi-thread"] }
anyhow = "1.0"
divsufsort = "2.0"
itertools = "0.14"
flate2 = "1.0"
zstd = "0.13"
wgpu = "0.19"
bytemuck = { version = "1.14", features = ["derive"] }
pollster = "0.3"
chrono = "0.4"
glob = "0.3"
pathdiff = "0.2"
regex = "1.11"
ahash = "0.8"
dashmap = "6.2"
dirs = "5.0"

Issues:

• itertools = "0.14" is very recent (June 2026); may not be stable
• divsufsort = "2.0" may have build issues on Windows
• No Cargo.lock visible in the tree (should be committed for binaries)

requirements.txt

Issues:

• Monolithic file with 20+ heavy dependencies
• GPU, quantum, and distributed compute deps all in one file
• No version pins (uses >= which can break with major version updates)
• perceval-quandela>=1.2.3 requires proprietary SDK

---

Remediation Priority Matrix

Priority	Issue	Effort	File
P0	Revoke and rotate Wolfram Alpha API key	15 min	wolfram_verify.py
P0	Remove hardcoded key from source	5 min	wolfram_verify.py
P1	Fix subprocess shell=True in verify_all_shims.py	1 hour	verify_all_shims.py
P1	Move SSH credentials to environment	30 min	verify_all_shims.py
P2	Fix SQL injection pattern	30 min	ingest_eigensolid_data.py
P2	Fix const cast in kernel module	15 min	pist_compress.c
P2	Add try/except for fragile imports	30 min	qaoa_adapter.py
P3	Remove dead code from kernel modules	1 hour	pist_compress.c
P3	Fix destructive stats read	30 min	pist_neuromorphic.c
P3	Split requirements.txt	30 min	requirements.txt
P4	Fix relative path assumptions	30 min	Multiple files
P4	Sync AGENTS.md with code	2 hours	AGENTS.md

---

Recommendations

1. Immediate Actions (Today):

   • Revoke the exposed Wolfram Alpha API key
   • Remove hardcoded IP/username from verify_all_shims.py
   • Run git filter-repo or similar to remove secrets from git history

2. Short-term (This Week):

   • Fix all subprocess shell=True usages
   • Fix SQL injection patterns
   • Fix kernel module const cast
   • Add secret scanning to CI (e.g., truffleHog or git-secrets)

3. Medium-term (This Month):

   • Remove dead code from kernel modules
   • Split requirements.txt into feature-specific files
   • Add proper error handling for missing files/paths
   • Sync documentation with actual code

4. Long-term:

   • Add automated security scanning to CI pipeline
   • Implement proper secret management (HashiCorp Vault, AWS Secrets Manager)
   • Add unit tests for kernel module sysfs interfaces
   • Create integration tests for shim verification pipeline