Pullzip

Add files from a folder into an existing zip/7z/tar archive (same paths are replaced). Secrets and generated folders are left out by default. Defaults to dryRun=true, which lists what would be added and replaced.

Status of the optional local classifier used by scan_security semantic=true (Horizon-Labs/prompt-injection-guard-small, Apache-2.0, about 300 MB). It runs offline on this computer; files are pinned to one revision and checked by SHA-256. Installing is the one other tool besides check_update that uses the network (refused when AF_OFFLINE is set). To install or remove it, the user runs "af model install" / "af model remove", or the write tool ai_model_install when the server allows writes.

Download (about 300 MB, once, from huggingface.co and cdn.jsdelivr.net) or remove the optional local injection classifier. Defaults to dryRun=true: show the user what will be downloaded and only run after they agree.

Apply a Pullzip patch to a base archive to produce the updated archive. Refuses if the base is the wrong version (hash mismatch) unless force is set; verifies the result contents. Defaults to dryRun=true.

Test every archive in a folder and compare with the previous check (ledger file .archiveflow-health.json in that folder): reports broken archives and silent corruption (bytes changed although the file was never modified), plus new, modified and missing ones.

Read and verify Pullzip's local hash-chained audit log (what was extracted / sealed / verified / renamed). verify=true checks the chain and reports any tampering (edited, reordered or deleted entries).

Safely extract many archives into one destination (each into its own folder when needed). One broken archive never stops the rest. Same options as extract_archive. Defaults to dryRun=true.

Collect every link inside text files and documents (folder, file or archive) and flag phishing look-alikes (e.g. naver-login.example.com), raw IP addresses, punycode, shorteners, credentials in URLs and program downloads. Offline: nothing is visited.

Has this exact file (by SHA-256) been seen before, and what was the verdict? Looks it up in the local opt-in reputation ledger — useful to spot a file a teammate already flagged, once you have imported their signed reputation bundle.

Compare this Pullzip with the latest published version and list what changed since (fetches https://pullzip.download/updates.json — nothing is sent). Set AF_OFFLINE=1 to disable.

For a split archive (x.7z.001, x.part1.rar, x.z01+x.zip, x.rar+x.r00): find all parts, report missing ones and which part to open.

Turn an archive or project folder into one Markdown document for an AI: file tree plus text files, secrets and personal data masked, .gitignore respected, secret files and binaries left out, within a token budget.

Re-pack an archive into another format (e.g. rar → zip for sharing, zip → 7z for size, anything → tar.gz for Linux). Goes through a safe extraction, so unsafe entries are dropped. Reports the size before and after. Defaults to dryRun=true.

Create an archive (zip, 7z, tar, tar.gz, tar.xz, tar.bz2) from a folder, with compression level/method, split volumes (e.g. "25m" for mail limits) and self-extracting .exe. Secrets (.env, private keys, credentials) and generated folders (node_modules, .git) are excluded by default. ZIP passwords use AES-256. Defaults to dryRun=true.

Make one archive per top-level item of a folder (or per listed item), e.g. one zip per customer folder. Defaults to dryRun=true.

Compare two versions of an archive and build a small patch that contains only the added and changed files (plus a manifest of removals). Ship the patch instead of the whole new archive. Defaults to dryRun=true.

Remove files or folders from a zip/7z/tar archive in place. Changes the archive permanently: run with dryRun=true first (the default) and confirm with the user.

Diagnose an archive that fails to open and give the one fix to run: encrypted (needs a password), damaged/truncated (repair), a missing split-volume part, a wrong extension (renamed file), a compression method the engine can't read (update 7-Zip), or garbled file names. Show the recommendation to the user.

Show files added, removed and changed between two archives (for example two versions of a project backup).

Show which 7-Zip engine is used, live engine metrics and which folders this server may access. With selfTest=true, runs an end-to-end engine check (create, list, password, read, test, extract).

A plain-language verdict for non-experts (Korean or English): is it safe, what is inside, why, where it came from, and what to do. Show this text to the user as-is.

Export the full file list of an archive as JSON, CSV (safe to open in Excel — formula injection neutralized) or an indented tree with folder sizes.

Export the local reputation ledger as a signed bundle a team can share (hashes and verdicts only; file names are stripped unless includeNames). Others import it with import_reputation after trusting your key.

Extract an archive (or selected entries / kinds of files) into a folder. Unsafe entries are always skipped, links are removed, Mark-of-the-Web is preserved, existing files are kept unless overwrite says otherwise. salvage=true recovers readable files from a damaged archive. Defaults to dryRun=true which only returns the plan.

Extract an archive and every archive inside it (zip in zip in 7z ...) level by level with the same safety rules, into one folder tree; inner archives are removed after they are unpacked unless keepInner=true.

Find identical files across folders and archives (e.g. the same photos in three backups) and how much space the copies waste.

Which files (including Word/Excel/HWPX documents, in folders or archives) contain resident registration numbers, phone numbers, card numbers, e-mail addresses, keys or tokens — counts per file only, never the values.

Rewrite a ZIP whose file names were saved in a legacy code page (Korean CP949, Japanese Shift-JIS, Chinese GBK/Big5, Russian CP866/1251) so names are stored as UTF-8 and display correctly on every system. File data is copied byte-for-byte; unsafe entries are left out. Defaults to dryRun=true.

Generate a random archive password without look-alike characters (0/O, 1/l/I), for create_archive or sharing.

Compute SHA-256 (or SHA-1, MD5, SHA-512, CRC32) of files and folders on disk, or — with archive set — of entries inside an archive without leaving them on disk.

Merge a signed reputation bundle from a teammate into the local ledger. The signature must be valid and the signing key must be in trustedKeys (or your own). Use dryRun first to see the signer.

Summarize an archive without extracting it: file counts by type, unpacked size, suggested extraction folder, detected projects, duplicates, archive comment, whether it came from the internet, and a security report (safe/caution/danger).

Inspect up to 100 archives at once: one line and a security level for each, dangerous ones first.

List the open-source licenses in a deliverable (folder or archive, dependencies such as node_modules included) from license files and package manifests, and flag strong copyleft (GPL/AGPL) and weak copyleft (LGPL/MPL). A first pass, not legal advice.

List files inside an archive with paging and an optional name filter (substring or glob like *.pdf).

Show files Pullzip moved into quarantine, with why, when, where they came from and their security findings.

Show the snapshots (versions) stored in a snapshot folder made by snapshot_folder: when, note, how many files were added, changed and removed.

Combine several archives into a single new archive; by default each source keeps its own folder (name clashes are renamed). Goes through a safe extraction, so unsafe entries are dropped. Defaults to dryRun=true.

Re-pack any archive as the smallest practical 7z (LZMA2, ultra, solid), keeping its password. Writes a new file next to it; the original is untouched. Defaults to dryRun=true.

Copy (or move) photos and videos from a folder or an archive into YYYY/YYYY-MM folders by the date they were taken (EXIF), skipping exact duplicates. Defaults to dryRun=true.

Pack a folder into several independent archives, each at most maxMb — unlike split volumes, every part opens on its own (good for e-mail or upload limits). Files bigger than the limit are reported. Defaults to dryRun=true.

Find photos (JPEG/PNG) that reveal where they were taken (GPS), the camera, or hidden XMP/IPTC/comments — in a folder, a single photo, or inside an archive — before you share them.

One step to send files safely: leaves out secrets (.env, keys), removes photo locations, adds a generated AES-256 password, splits into mail-sized parts, and writes the message to send (password to be sent separately). Defaults to dryRun=true.

Move a suspicious file into the Pullzip quarantine folder (renamed so it cannot be opened by double-click), recording why, its hash, security findings and download address. Can be restored later.

Read one file from an archive without extracting it. Text is returned wrapped as untrusted content; images (png/jpg/gif/webp up to 5 MB) are returned as images.

Record this archive's verdict in the local opt-in ledger (only the SHA-256 hash and verdict; the file name is stored only if shareName is set). Building this ledger, and sharing signed bundles, is what powers the network effect.

Write a copy of an archive with keys, tokens, passwords, resident numbers, phone numbers, cards and e-mails masked inside its text files. Binary files are copied unchanged; the original is untouched. Defaults to dryRun=true.

Rename many files inside a zip/7z/tar at once by rules: find/replace (text or regex), replace spaces, lowercase/uppercase, add a prefix/suffix, limited to selected globs. Unsafe results and collisions are skipped. Defaults to dryRun=true (shows the plan).

Rename or move files and folders inside a zip/7z/tar archive without re-compressing. Unsafe new names (../, absolute, reserved) and collisions are refused. Defaults to dryRun=true.

Rebuild a ZIP that will not open (cut-off download, missing central directory, junk in front) by reading its file records one by one. Writes a new "<name>-repaired.zip"; the original is untouched. Reports recovered and lost files, and tests the result. Defaults to dryRun=true.