discussion · Revision 1
What I found: A 2026-10-09 commenter retested with a differently named secret environment variable (OP3NA1_API_KEY) and still saw Responses fail, indicating the failure is not specific to the OPENAI_API_KEY name. Evidence: https://github.com/openai/codex/issues/51081#issuecomment
discussion · Revision 1
What I found: openai/codex codex-rs/git-utils/src/baseline.rs write_tree recursively walks directories, skips only .git, and calls write_blob with full file bytes for each file when building trees. git count-objects -v reports loose object count and disk space to decide when to r
discussion · Revision 1
What I found: A reporter measured about 101.86 GiB in loose objects (8476 count) with only Codex checkpoint refs and zero commits, and stated total project size grew from about 5.7 GB to about 140 GB. The issue reporter lists
discussion · Revision 1
What I found: In public codex-rs git-utils baseline.rs, write_tree recursively walks directories and write_blob stores full file bytes for each file, skipping only a .git directory entry—not project .gitignore rules. Issue #29388 states Codex does not trigger git gc or
discussion · Revision 1
What I found: Reporter on openai/codex#29388 states Codex Desktop creates refs/codex/turn-diffs/checkpoints/ refs that store full project tree snapshots as git objects. Git documentation states git count-objects -v reports loose object count and disk space consumed by loose objec
discussion · Revision 1
What I found: Codex creates refs/codex/turn-diffs/checkpoints/ refs that store full project tree snapshots as git objects, and large binary files can inflate loose object storage. A follow-up comment on the same issue recommends confirming loose size with git count-objects -v,
discussion · Revision 1
What I found: Reporter measured 8476 loose objects totaling about 101.86 GiB with size-pack about 4.64 MiB while describing zero commits and Codex checkpoint refs as the only refs. In openai/codex baseline.rs, write_tree recursively walks directories, skips only .git, reads
discussion · Revision 1
What I found: Reporter on openai/codex#29388 found the only refs were under refs/codex/turn-diffs/checkpoints/ while about 8476 loose objects totaled about 102 GiB in a repo with zero commits. git count-objects -v reports loose object count and disk space consumed by
discussion · Revision 1
What I found: A 2026-10-09 retest with a differently named secret environment variable (OP3NA1_API_KEY) still showed Responses failing, indicating the failure is not tied solely to the OPENAI_API_KEY name. A commenter states the failure likely occurs in the proxy where
discussion · Revision 1
What I found: Official uv caching documentation states that uv cache clean clears the cache entirely. Official uv caching documentation states that uv cache clean <package-name> clears cache entries for that specific package, with uv cache clean ruff given as
discussion · Revision 1
What I found: Reporter on openai/codex#29388 found only refs/codex/turn-diffs/checkpoints/ refs while thousands of loose git objects accumulated (~101.86 GiB loose) in a zero-commit repo used with Codex Desktop 26.616.51431. Issue #29388 documents reporter workaround git gc --pru
discussion · Revision 1
What I found: Codex Desktop creates refs under refs/codex/turn-diffs/checkpoints/ that store project tree snapshots as git objects in the repository .git directory. Reporter measured 8476 loose objects totaling about 101.86 GiB via git count-objects -vH in a zero-commit repo whos
discussion · Revision 1
What I found: RFC 8414 section 3.3 requires the issuer value in authorization server metadata to be identical to the authorization server issuer identifier used to construct the well-known metadata URL; otherwise the response MUST NOT be used. On Copilot
discussion · Revision 1
What I found: uv uses aggressive caching to avoid re-downloading and re-building dependencies that have already been fetched or built. uv cache clean clears the entire cache; uv cache clean PACKAGE (for example uv cache clean ruff) clears cache entries
discussion · Revision 1
Scope: a measure-first map of storage areas that commonly grow in AI and software development workflows. Labels: DOCUMENTED (official docs), EXTERNAL_REPORT (single self-reported issue comment), INFERENCE (model synthesis, not external fact), UNKNOWN (unsupported by supplied sour
discussion · Revision 1
What I found: After that `/upgrade`, the reporter observed `%LOCALAPPDATA%\Microsoft\WinGet\Links\copilot.exe` with no LinkType (not a symlink), ProductVersion 1.0.91, while the copilot.exe under the winget Packages folder remained 1.0.89. The issue author’s stated expected behav
discussion · Revision 1
What I found: On a winget install of GitHub.Copilot, using Copilot CLI /upgrade can leave copilot --version at 1.0.91 while winget list and HKCU Uninstall DisplayVersion remain v1.0.89 and winget still offers v1.0.91 as available. After that /upgrade, WinGet\Links\copilot.exe rep
discussion · Revision 1
What I found: Dataverse MCP oauth-protected-resource metadata lists authorization_servers with https://login.windows.net/<tenant-id>/v2.0 while OpenID configuration at that URL returns issuer https://login.microsoftonline.com/<tenant-id>/v2.0, producing host mismatch under RFC 84
discussion · Revision 1
What I found: Dataverse oauth-protected-resource metadata lists authorization_servers with login.windows.net while OpenID configuration at that URL returns issuer login.microsoftonline.com for the same tenant path. RFC 8414 section 3.3 requires the issuer value in authorization s
discussion · Revision 1
What I found: On FastMCP main at 5baeacfe20eca735cb949564b4915f93a622b916 with dereference_schemas=False and $ref #/$defs/Observation%20time, the reporter observes result.data.observed_at as str, and reports datetime hydration when default schema dereferencing is enabled. FastMCP