Digital Delivery Preflight

A reviewable inventory for your next file delivery.

Purchases are not open yet. Explore the tools and their limits below.

Before a handover

For independent creators reviewing a prepared, stable product directory or ZIP. Find packaging concerns before you decide what to send.

Inventory paths and declared sizes; compute SHA256 where contents are readable.
Flag unsafe ZIP paths, symlinks, collisions, unusual compression and common packaging residue.
Run bounded possible-secret text heuristics with masked values and explicit scan status.

What is included

Portable SKILL.md instructions and Python checker
Synthetic ZIP fixture generator
19 regression tests and validation notes
Practice task and Chinese quick start

The package includes synthetic examples and regression tests. These cover selected scenarios, not comprehensive security assurance.

Input → report

Input: one directory or ZIP. Output: JSON on stdout; the default command does not extract, execute, upload or modify the package.

python3 scripts/preflight.py "delivery.zip"

Synthetic input

needs-review.zip
  ../outside.txt
  .git/config
  settings.env
  compressed.txt
  binary.dat
  link

These are deliberately defective test entries, not a recommended archive structure.

Actual example summary

{
  "error": 1,
  "warning": 5,
  "info": 2,
  "files": 6,
  "hashed": 4
}

6 entries, 4 hashes. High-ratio and symlink contents are skipped. Binary bytes are hashed without a text-secret scan.

Read the complete synthetic report

Boundaries worth knowing

Not antivirus, a malware scan, a security audit, a copyright or license audit, or proof of secret absence.
Defaults: 10,000 entries; 32 MiB per file; 128 MiB cumulative reads; compression ratio 100; 1 MiB text prefix. Skipped and partial checks matter.
ZIP metadata and directory listings enter memory before the entry limit. This is not a hostile-input sandbox; use a bounded, stable copy.
Nested archives are not expanded. Binary, other encodings, encrypted or unreadable entries may remain unscanned.
Manifest presence is a filename convention check, not content or hash validation. SHA256 establishes byte equality, not authenticity.
Reports retain paths, which can reveal private information. Review before sharing.

Quick questions

Does exit 0 mean the product is safe to sell?

No. It means no findings within the configured checks and limits. Exit 1 needs review; exit 2 indicates invalid or unreadable input.

Will it fix or remove flagged files?

No. The helper is read-only and offers findings for a human decision.

Does it inspect existing checksum manifests?

No. It computes its own hashes where possible; it does not validate a manifest’s contents.

Runtime & data handling

The scripts use Python 3.10+ features and only the standard library. Actual script testing: Python 3.12.14 on Linux. Other runtimes and operating systems are unverified.

Included SKILL.md instructions can guide an agent to run the script. Installation and automatic discovery in external hosts, including Claude and other Codex clients, have not been verified.

The helper processes files locally without network calls or source-file edits. This is not an end-to-end privacy promise: your agent host may transmit text to its model provider. Review its configuration before supplying confidential files.

Availability

We are preparing these tools for sale. Purchases and payment links are not available yet. No release date is promised.

Final prices, commercial licensing, refund terms, support and delivery details will be shown before any purchase is offered. No purchase license or service commitment is granted by this page.