Codex fails to read Documentation link pointing to GitHub complaining about "cache-miss"

What version of Codex CLI is running?

codex-cli 0.98.0

What subscription do you have?

Free

Which model were you using?

gpt-5.2-codex

What platform is your computer?

Darwin 25.2.0 arm64 arm

What terminal emulator and version are you using (if applicable)?

wezterm

What issue are you seeing?

The codex CLI fails to read a documentation link that claude has no problem reading and just completely goes off the deepend.

Full transcript in the repro below

What steps can reproduce the bug?

❯ codex
╭──────────────────────────────────────────────────────╮
│ >_ OpenAI Codex (v0.98.0)                            │
│                                                      │
│ model:     gpt-5.2-codex high   /model to change     │
│ directory: ~/projects/vxsort-cpp/…/smallsort/codegen │
╰──────────────────────────────────────────────────────╯

  Tip: New Codex is included in your plan for free through March 2nd – let’s build together.

■ '/model' is disabled while a task is in progress.

• Model changed to gpt-5.2-codex xhigh


› Please read the documentation about free-threaded z3 solver here: https://github.com/Z3Prover/z3/blob/38d6107de47a7997c4798525e1a0aaf993b36ca3/README.md#free-threaded-python-support-python-313t


• Searched

• I'm considering triggering a search via web.run when opening a file fails due to a cache miss, then opening the raw result afterward.

• Searched Z3 README free-threaded python support python 3.13t 38d6107de47a7997c4798525e1a0aaf993b36ca3

• I'm preparing to access the README file directly from the raw GitHub content at the specified commit to find the needed section. This avoids relying on incomplete search results and enables precise content extraction.

• Searched

• The user-provided GitHub blob URL fails to open due to a cache miss; it may require performing a search with the URL or anchor first to get a cached result before opening. Next, I'll try searching with the URL to obtain a valid search
  result, then open that to bypass the cache miss.

  Searching to bypass cache miss

  I'm searching for the README anchor on GitHub to get a search result ID, then I'll open that to avoid the cache miss on the direct URL.

• Searched "free-threaded" "python 3.13t" "Z3" README.md

• Searched Z3 free-threaded python 3.13t README

• I'm attempting to open the README from the repo search result to check if it shows the full content or a truncated version, aiming to locate the specific "free-threaded" anchor or section.

• Searched https://github.com/Z3Prover/z3

• I'll try reopening the README at the exact commit URL again despite the earlier cache miss, since the user specifically wants that version's documentation. If it fails, I'll attempt navigating from the repo commit page to reliably
  access the commit-specific README.

  Investigating cache miss with commit-specific open

  I'm analyzing that the cache miss likely happens because open expects prior cached data from a search or user message; reopening the commit-specific README directly fails. I'll try searching for the exact commit URL or read the README
  on master to locate the needed section, noting that master’s README may suffice if the commit content is similar.

• Searched 38d6107de47a7997c4798525e1a0aaf993b36ca3 README.md Z3Prover/z3 blob

• I'm preparing to search for the raw README URL from the specific commit to access free-threaded Python support details, aiming to find it within allowed source URLs.

• Searched raw.githubusercontent.com/Z3Prover/z3/38d6107de47a7997c4798525e1a0aaf993b36ca3/README.md

• The commit-specific README file can't be reliably opened due to cache misses, and only exact URLs from search results or user messages are allowed; fallback to master README is possible but doesn't meet the commit-specific requirement.

  Analyzing cache restrictions on commit URLs

  The tool's cache likely blocks opening unindexed URLs even if user-provided, requiring URLs to appear in search results first; since the exact commit README URL isn't indexed, access fails repeatedly, limiting retrieval options for
  commit-specific content.

• Searched

• Since the URL isn't in the search results causing a cache miss, I'm planning to fetch the content from the README on master by opening and searching it for the desired terms instead.

• Searched 'free-threaded' in https://github.com/Z3Prover/z3

• The master README search shows no free-threaded mention, so next I’ll open the full README file from the repo page directly to check if the section appears there despite search misses.

  Checking README file view for free-threaded section

  I'm opening the README file view directly to perform an internal search for "free-threaded" in case relevant sections load dynamically or aren’t in the repo page HTML.

• Searched https://github.com/Z3Prover/z3/blob/master/README.md

• The README content appears truncated or dynamically loaded on GitHub, preventing direct access; traditional raw URL access isn't feasible here due to restrictions and dynamic loading.

  Planning raw README retrieval via page link

  I'm targeting the README page's "Raw" button by searching for "Raw" in the page DOM to locate and follow its link for fetching the raw README content.

• Searched 'Raw' in https://github.com/Z3Prover/z3/blob/master/README.md

• The current tools restrict opening URLs only from prior search results or user messages, so I can't directly access alternate raw URLs like the README; I'll need to ask the user to provide content or confirm permission to use specific
  URLs.

  Reporting fetch failure and requesting content

  I tried to open the exact README URL but the fetch failed due to cache restrictions (website). I accessed the general repository page but couldn't find the requested "free-threaded" section (website). Please paste the relevant content
  or provide an accessible raw URL, and let me know how you'd like me to assist further.

─ Worked for 1m 30s ─────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────

• I tried to open the exact README link you provided, but the fetch failed due to a cache restriction in the browsing tool. ( ()) I could only reach the general GitHub repository pages, and they did not expose the “free-threaded” section
  or even the README content in a way I could read. (github.com (https://github.com/Z3Prover/z3/blob/master/README.md))

  If you paste the relevant section (or provide an alternate accessible URL), I can read it and summarize or apply it to your codebase right away.

  Summary of what I found and why it was insufficient: I could not access the specific README at the commit you linked due to a tool cache limitation, and the accessible GitHub pages did not show the needed section, so I wasn’t able to
  read the free-threaded Python 3.13t documentation. ( ())

What is the expected behavior?

The the documentation be read, either from the original link or by reading the raw files from GH, for example in this case:
https://raw.githubusercontent.com/Z3Prover/z3/38d6107de47a7997c4798525e1a0aaf993b36ca3/README.md

Additional information

Claude Code can do it:

⏺ I'll start by understanding the free-threaded Z3 documentation, then capture the golden truth output, and plan the migration from multiprocessing to threading.

⏺ Fetch(https://raw.githubusercontent.com/Z3Prover/z3/38d6107de47a7997c4798525e1a0aaf993b36ca3/README.md)
  ⎿  Received 12.1KB (200 OK)

⏺ 2 Task agents finished (ctrl+o to expand)
   ├─ Explore Z3 Context API · 51 tool uses · 71.7k tokens
   │  ⎿  Done
   └─ Read all source files · 6 tool uses · 23.0k tokens
      ⎿  Done