Repository navigation
install: a guided, coloured install that leaves codeaf ready to use - #1635
Merged
Merged
Conversation
The installer printed a bare version line and a PATH line. It now shows checked steps with a spinner, links the command into a folder already on PATH so it works in the same terminal, prints a short Get started guide with agentfield.ai docs links, and offers to start codeaf right away. Plain text for pipes, NO_COLOR, CI and --verbose; --no-start skips the question. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LzgNn3hAVmEVnHvXHLJ1E1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LzgNn3hAVmEVnHvXHLJ1E1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LzgNn3hAVmEVnHvXHLJ1E1
…er services go through /connect Four things from the review of #1635, three of them found against the v0.4.1 binary that main ships: - link_into_path replaced ANY symlink named codeaf in ~/.local/bin, ~/bin or /usr/local/bin, so a developer's source build linked there was silently swapped for the stable install, against the page's own promise. It now replaces a link only when it already lands on this install's file (`-ef`, a shell test, so the installer still needs no readlink); a link to anything else stays and the paste line is printed. - `Start codeaf in ~ now? [Y/n]` defaulted to yes in the home folder, where a piped install usually runs, which opened codeaf over the whole home folder. The offer is skipped in the home folder, at / and in a deleted folder; the guide already says to cd into a project first. - The guide said DeepSeek, Qwen, GLM and Kimi take a key and Ollama needs none right after "Connect a model when it asks", but the first-run screen offers only OpenRouter. It now says those go through /connect, which is true on v0.4.1 and on dev. - The spinner sliced a string of braille frames one character at a time, which needs the named locale to be installed: with a UTF-8 LANG that ssh forwarded to a machine without it (bash 5 and bash 3.2 alike) every frame printed as a broken byte. The frames are an array now. GUIDE also said NO_COLOR, CI and --verbose all meant plain text and no question; NO_COLOR is plain text and still asks, CI and --verbose skip the question. The sentence now says which is which. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
AbirAbbas
pushed a commit
that referenced
this pull request
Sep 28, 2026
…1635) The installer on main, on top of v0.4.1 and nothing else from dev: checked steps with a spinner, a link into a folder already on PATH so codeaf works in the same terminal, a Get started guide, and on a terminal the question `Start codeaf in <folder> now? [Y/n]`, read from /dev/tty. The binary is untouched; only the script, its tests and the pages that describe it move. scripts/install.sh, test/installer-telemetry.sh and internal/release/install_test.go are byte-identical to #1635's head, which also carries the installer halves of two dev changes it was written on: the installer prints no telemetry notice (#1511, with the matching docs/TELEMETRY.md paragraph; v0.4.1 still prints the full notice itself before any count is sent), and an install under another name puts that name first in its receipt (#1519). docs/GUIDE.md and the two manual pages take #1635's paragraphs in place of the ones that described the three-line notice. It also carries the review fixes made on the pull request before it landed: only the installer's own link on PATH is ever replaced, the start question is not asked in the home folder or at /, the spinner's frames survive a UTF-8 locale that is not installed, and the guide sends DeepSeek, Qwen, GLM, Kimi and Ollama through /connect, since v0.4.1's first run offers OpenRouter alone. This commit is merged into staging and dev so main stays an ancestor of staging and staging of dev, and every later promotion is still a fast-forward. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
AbirAbbas
added a commit
that referenced
this pull request
Sep 28, 2026
Brings e446507 (the guided installer, on top of v0.4.1) into staging, so main can fast-forward onto it and the rc it publishes passes the "already on staging" check. Staging already carried #1511's half of that commit; the conflicts in scripts/install.sh and test/installer-telemetry.sh resolve to #1635's head, byte for byte. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
AbirAbbas
added a commit
that referenced
this pull request
Sep 28, 2026
#1635 reached dev as 88d7e71. The same change reached main as e446507, on top of v0.4.1 and nothing else, and staging as the merge 80c23bc. This merge changes no file; it makes 80c23bc an ancestor of dev, so the staging build passes release.yml's "already on dev" check, the rc passes "already on staging", and the next promotion of dev to staging and of staging to main is still a fast-forward. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Collaborator
|
Reviewed and landed. On top of your three commits I pushed a6061a7 with four review fixes (the first three found against v0.4.1, which main ships):
Tested: the installer suite under bash 5 and a from-source bash 3.2, the release tests, and live tmux installs against the real v0.4.1 release (home folder, a foreign link, a clean project folder where Enter opens the first-run screen). Where it landed:
|
santoshkumarradha
added
the
area:build
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What changes
The installer's output was a bare
installed codeaf …line plus a rawexport PATH=…, and the livemainscript still prints a three-line telemetry notice. This replaces it with:Downloaded,Installed codeaf <tag>(split to fit 80 columns),PATH,Linked.~/.local/bin,~/binor/usr/local/binis on PATH and writable, it links the command there socodeafworks in the same terminal. A file there that is not its own link is never replaced; if anothercodeafwould still win on PATH, the paste line stays.cd your-project/codeaf, how to connect a model, handy commands, and links to agentfield.ai/docs/codeaf, playbooks, connections, Discord.Start codeaf in <folder> now? [Y/n](read from/dev/tty, not the piped script); Enter opens codeaf on its first-run connect screen. Skipped for pipes,CI,--verbose,--no-start/CODEAF_NO_START=1.NO_COLORand non-tty.Bash 3.2 (macOS
/bin/bash) compatible.Proof
go test ./internal/release/ -run 'Install|H10|V1|V2|Pinned|Documented|Custom': ok (newTestInstallerLinksIntoAFolderAlreadyOnPath)bash test/installer-telemetry.sh: 33 passedgo test ./internal/manual/ ./internal/namelaw/: ok🤖 Generated with Claude Code
https://claude.ai/code/session_01LzgNn3hAVmEVnHvXHLJ1E1