delta-cloud-208e 06057f864b fix(installers): 8 critical bugs from gpt-5.5 + glm-5.1 audit (Item 17) ...

Sub-agents reviewed all 26 installer scripts. Fixed (TDD):

BLOCKERS (install fails on platform):
1. claude/uclaude_uninstall.sh: replace `sed -i` → portable `sedi()` helper
   (BSD sed on macOS requires `-i ''`, GNU uses `-i`). Same fix style as
   codex/ucodex_install.sh:sedi.
2. claude/uclaude_install.ps1: abort if $apiKey null after fetch attempt
   (was silently completing install with broken auth env vars). Guard
   added to all 8 ps1 scripts.
3. qwen/uqwen_install.ps1 + uqwen_update.ps1: build trustedFolders.json
   via [ordered]@{} | ConvertTo-Json (PowerShell single-quoted literal
   was preserving `\"` as backslash+quote, producing INVALID JSON).
4. codex/ucodex_update.ps1: check $LASTEXITCODE after Python patcher call
   (native command non-zero exit doesn't throw under
   ErrorActionPreference='Continue' — patcher failure was silent, no
   PowerShell fallback triggered).

HIGH (wrong behavior / regressions):
5. gemini/ugemini_install.ps1 + update.ps1: read $env:UGEMINI_API_KEY
   FIRST (was only checking $env:UCLAUDE_API_KEY — claude variable).
6. gemini/ugemini_update.ps1: download gemini_config.json from PRIVATE
   unlimitedcoding-config (was downloading from public — would 404 after
   Item 14 sanitization).
7. claude/uclaude_update.ps1: drop ANTHROPIC_API_KEY assignment + dynamic
   models fetch (regression — install.ps1 was fixed earlier but update.ps1
   still set both env vars, re-introducing Auth conflict warning).
8. codex/ucodex_install.sh + update.sh: GitHub API curl needs
   `-H "User-Agent: UnlimitedCoding-Installer"` and `-f` flag (default
   curl/X.Y UA gets 403 from GitHub API + silent fail on 5xx).

Bonus fixes pulled in:
- codex/ucodex_install.ps1: switch codex_config download URL to private
  repo (consistency with update.ps1 + Item 14 sanitization)
- codex/ucodex_install.ps1: add `--all` flag to patcher invocation
  (matched between install + update)

Tests: tests/test_installer_bugs_audit.py — 9 GREEN regression guards.
Total: 186 tests GREEN.

Audit transcripts: gpt-5.5 found 24 issues (claude+gemini), glm-5.1
found 11 issues (codex+qwen). Lower-priority items (heredoc unsafe
quoting, lock files, schema validation) deferred to next iteration.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

2026-04-25 17:16:43 +00:00

3.0 KiB

Executable File

Raw Blame History

View Raw