Repository navigation
[Bug]: Failed CrawlResult has links: {} / media: {} instead of the usual keys #2286
Description
crawl4ai version
0.9.4 (also on 0.9.2)
Expected Behavior
A result that failed (robots.txt refusal, fetch error, "All proxies failed") has the same links and media keys as a successful one:
"links": {"internal": [], "external": []},
"media": {"images": [], "videos": [], "audios": []}so a client can read links["internal"] without special-casing failures.
Current Behavior
CrawlResult.links and CrawlResult.media default to {} (models.py#L136-L137). Successful crawls fill them from the scraper's typed Links / Media models, but the results AsyncWebCrawler.arun builds itself never set them:
- robots.txt refusal (async_webcrawler.py#L385-L397)
- "All proxies failed" (#L640-L646)
- the exception path (#L710-L712)
- the anti-bot raw-HTML fallback, which is a successful result (#L588-L598)
So in a mixed batch /crawl returns "links": {}, "media": {} for exactly the entries a client has to handle differently, and in the SDK result.links["internal"] raises KeyError. Since #2134 keeps failed results in /crawl responses, every batch with a refused or failed URL has this shape. In one client that read links.internal.length, the resulting TypeError was taken for a failed request and the healthy pages of the batch were re-crawled several times over.
cc @nightcityblade (#2134) @ntohidi
Is this reproducible?
Yes
Inputs Causing the Bug
- Two URLs on one site, one of them disallowed by its robots.txt
- crawler_config params: {"check_robots_txt": true}Steps to Reproduce
1. Run the Docker image (0.9.4, or built from develop)
2. POST /crawl with both URLs and check_robots_txt=true
3. Compare `links` / `media` of the two resultsCode snippets
from crawl4ai.models import CrawlResult
# What AsyncWebCrawler.arun returns for a URL robots.txt disallows
r = CrawlResult(url="https://example.com/private", html="", success=False,
status_code=403, error_message="Access denied by robots.txt")
print(r.links, r.media) # {} {}
r.links["internal"] # KeyError: 'internal'OS
Linux (Docker image built from develop @ 1f68e5b)
Python version
3.12 (the image's)
Browser
Chromium (Playwright, bundled)
Browser version
No response
Error logs & Screenshots (if applicable)
POST /crawl, check_robots_txt=true, two URLs of a local test site (CRAWL4AI_ALLOW_INTERNAL_URLS=true), /private disallowed by its robots.txt:
[shape] HTTP 200
/index.html success=True status=200 links=['external', 'internal'] media=['audios', 'images', 'videos'] error=''
/private/index.html success=False status=403 links={} media={} error='Access denied by robots.txt'
Activity
nightcityblade commentedon Oct 9, 2026
Hi, I'd like to work on this. I'll submit a PR shortly.
nightcityblade commentedon Oct 9, 2026
I see #2289 already addresses this issue, so I'm stepping back to avoid duplicate work.
ntohidi commentedon Oct 12, 2026
Thanks for the clear report. Confirmed on
develop. Here is the root cause.Root cause
CrawlResultgivesmediaandlinksa default of{}(models.py#L136-L137):Only the success path fills in the full shape (
internal/external,images/videos/audios). Every other place that builds aCrawlResultkeeps the empty default:AsyncWebCrawler.arun: robots.txt refusal (L394), "all proxies failed" (L653), the exception path (L724), and prefetch mode, which setslinksbut notmedia(L764)async_dispatcher.py(L255, L307, L345, L678, L736). The issue does not list these, but they have the same problem.Since #2134 keeps failed results in
/crawlresponses, any batch with a failed URL returns two shapes.Fix direction
Change the default once, in the model, not at each call site. That covers every place above and any added later. #2289 does this and is under review.
ntohidi commentedon Oct 12, 2026
The fix is currently on the 'develop' branch and will be released soon.
ntohidi commentedon Oct 14, 2026
Fixed by #2289, released in v0.9.5. Upgrade with
pip install -U crawl4ai. Release notes: https://github.com/unclecode/crawl4ai/blob/main/docs/blog/release-v0.9.5.md