Download UFO Files Release 01 & 02 (war.gov)

Purpose

Download the two UAP/UFO document bundles published by the U.S. Department of War on its "Presidential Unsealing and Reporting System for UAP Encounters" page (war.gov/UFO/): Release 01 Documents (Release_1.zip, ~1.2 GB, 262 files) and Release 02 Documents (release_02_document_bundle.zip, ~70 MB). Both are publicly accessible ZIP archives of declassified PDFs/images. This skill retrieves the raw ZIPs and verifies them; it is read-only and never submits anything.

When to Use

A user wants the full Release 01 and/or Release 02 UAP document archives from war.gov pulled to local disk.
Archival / bulk-ingest pipelines that need the original ZIPs (not individual PDFs).
Re-validating that the official bundle URLs are still live and unchanged.
NOT for the separate video bundles — those are large CloudFront-hosted archives (see Gotchas) and are out of scope here.

Workflow

These ZIPs are too large for the lightweight Browserbase Fetch API (browse cloud fetch returns 502 The response body exceeded the maximum allowed size). They must be pulled through a real browser session: the file downloads inside the session's Chrome, syncs to Browserbase session storage, and is retrieved as a ZIP-of-downloads via the sessions-downloads API. A real Chrome session — even a bare one with no stealth — passes war.gov's WAF, so neither --verified nor --proxies is required (verified across both files; see Gotchas for why the pre-run probe disagreed).

Direct, stable bundle URLs (no auth, no Content-Disposition; application/zip, HTTP 200):

Release 01 Documents — https://www.war.gov/medialink/ufo/bundle/Release_1.zip (1,223,976,178 bytes)
Release 02 Documents — https://www.war.gov/medialink/ufo/052226/release_02/release_02_document_bundle.zip (69,986,448 bytes)

Create a bare keep-alive session. No stealth flags needed:
```
sid=$(browse cloud sessions create --keep-alive --timeout 2400 \
  | node -e "let s='';process.stdin.on('data',c=>s+=c).on('end',()=>process.stdout.write(JSON.parse(s).id))")
```
Use a generous --timeout (e.g. 2400s) for Release 01 — the 1.2 GB transfer can run several minutes and the session must outlive it.
Enable download syncing, then navigate to the ZIP. Browserbase only syncs a download to session storage when CDP download behavior is set with eventsEnabled: true before the navigation. The browse CLI cannot send raw CDP commands, so use a tiny Node CDP client against the session's connectUrl:
- Connect a WebSocket to connectUrl (from browse cloud sessions get <sid>).
- Send Browser.setDownloadBehavior { behavior: "allow", downloadPath: "downloads", eventsEnabled: true }.
- Target.attachToTarget (flatten) the page target, Page.enable, then Page.navigate to the ZIP URL.
- Wait for Browser.downloadProgress with state: "completed" (receivedBytes === totalBytes). The page itself stays on the prior URL — a ZIP triggers a download, not a navigation, so the title/URL won't change and Page.navigate may reject with net::ERR_ABORTED; that rejection is expected, not a failure.
Wait for the sync, then retrieve. After the in-browser download completes there is a propagation delay (~12–30 s, longer for 1.2 GB) before the file appears in session storage. Poll:
```
browse cloud sessions downloads get "$sid" --output downloads.zip
```
A 22-byte result means "nothing synced yet" — sleep and retry until the size jumps to the real bundle size. The returned downloads.zip is a wrapper ZIP: it contains the actual bundle renamed with a timestamp suffix, e.g. Release_1-<epoch_ms>.zip. Unzip one level to get the real bundle.
Verify. Unzip the wrapper, confirm the inner ZIP's byte size matches the expected content-length, and unzip -l it. Release 01 should list 262 entries under Release_1/ (FBI photos + *-HQ-* document PDFs); Release 02 lists UAP PDFs under release_02_document_bundle/ (DOE/CIA/DOW *-UAP-* files). The bundles include macOS __MACOSX/ resource-fork entries — ignore them.
Release the session: browse cloud sessions update "$sid" --status REQUEST_RELEASE.

A reusable Node CDP downloader implementing step 2 is straightforward (~60 lines): get connectUrl, open a WebSocket, set download behavior, attach + navigate, resolve on Browser.downloadProgress.state === "completed".

Finding the links (optional discovery)

If the bundle URLs ever change, open https://www.war.gov/UFO/ and read the anchor hrefs whose text matches Download Release NN Documents. Note the Release 01 link lives in a hidden release-toggle tab — on page load the "Release 02" tab is active, so the Release 01 document link is not visible until you click the RELEASE 01 tab (or just read it straight from document.querySelectorAll('a'), since hidden anchors are still in the DOM).

Site-Specific Gotchas

Bare browser works; the probe's 403 is a non-browser artifact. A plain HTTP client (incl. the Browserbase Fetch API without proxies) gets 403 from war.gov's WAF — that is what the pre-run probe saw. A real headless Chrome session, even with no --verified/--proxies, loads the homepage and both ZIPs with 200. So set verified:false, proxies:false honestly. If the WAF ever tightens, adding --proxies/--verified is the escalation, but it was not required here and --proxies would add per-GB bandwidth cost (painful on the 1.2 GB file).
Fetch API can't download these. browse cloud fetch <zip> --proxies returns 502 The response body exceeded the maximum allowed size. Use a browser session to handle large responses. Don't waste time on the Fetch path for the bundles.
Downloads do NOT sync unless Browser.setDownloadBehavior(eventsEnabled:true) is set first. A plain browse open <zip-url> triggers the download in Chrome but Browserbase captures nothing — downloads get returns the empty 22-byte ZIP. The CDP download-behavior call is mandatory and must precede the navigation.
downloads get returns a wrapper ZIP, not the bundle directly. The real archive is nested inside and renamed with a -<epoch_ms> suffix (Release_1-1781017165599.zip). Always unzip one level.
Sync is asynchronous. Right after the in-browser download completes, downloads get can still return 22 bytes for 10–30 s. Poll with a delay; don't treat the first empty result as failure.
Release 01 is 1.2 GB. Budget disk (~2.5 GB transient for wrapper + extracted inner), session timeout (--timeout 2400), and wall time (several minutes). The in-browser download is reliable end-to-end but slow.
Release 01 download link is in a hidden tab. On war.gov/UFO/ the Release 02 tab is selected by default; the Release 01 documents anchor is present in the DOM but offsetParent === null until the RELEASE 01 tab is clicked. The direct URL is stable regardless, so prefer the hard-coded URL over scraping the visible tab.
Video bundles are a different host and out of scope. The same page also offers "Download Release 0N Videos" — those point at CloudFront (https://d34w7g4gy10iej.cloudfront.net/uapvideos.zip ~1.3 GB, uap052226.zip ~5.6 GB), not war.gov. This skill targets only the two document bundles.
A CSV manifest exists for Release 01 at https://www.war.gov/Portals/1/Interactive/2026/UFO/uap-release001.csv — handy for cross-checking the file list without unzipping 1.2 GB.

Expected Output

A confirmation object per downloaded bundle (paths are local to wherever you unzipped):

{
  "success": true,
  "verified": false,
  "proxies": false,
  "bundles": [
    {
      "release": "01",
      "label": "Download Release 01 Documents",
      "source_url": "https://www.war.gov/medialink/ufo/bundle/Release_1.zip",
      "content_type": "application/zip",
      "bytes": 1223976178,
      "inner_filename": "Release_1.zip",
      "entry_count": 262,
      "top_folder": "Release_1/",
      "valid_zip": true
    },
    {
      "release": "02",
      "label": "Download Release 02 Documents",
      "source_url": "https://www.war.gov/medialink/ufo/052226/release_02/release_02_document_bundle.zip",
      "content_type": "application/zip",
      "bytes": 69986448,
      "inner_filename": "release_02_document_bundle.zip",
      "top_folder": "release_02_document_bundle/",
      "valid_zip": true
    }
  ]
}

Failure shape (e.g. WAF tightened, or sync never produced a non-empty archive):

{ "success": false, "release": "01", "error_reasoning": "downloads get returned 22-byte empty archive after 6 retries / 120s" }

download-ufo-files