install.sh delegates config generation to archied setup #580

Closed
opened 2026-10-27 13:36:54 +00:00 by sam · 0 comments
Owner

Remove the TOML-generating branches from install.sh and call the binary it just built. Ends the class of bug where the installer's idea of the config schema drifts from the daemon's.

Acceptance Criteria

install.sh contains no TOML templates and no sed-patching of config files. It invokes archied setup for a fresh install and skips it when a config already exists. A non-interactive install still produces a valid config or fails loudly. The forge_block helper and its awk splice are deleted.

Close reason

Already satisfied on main before this bead was picked up -- verified criterion by criterion rather than assumed, and the missing half was the test that keeps it satisfied.

AC1 (no TOML templates, no sed-patching): install.sh contains no TOML table header or key assignment at all; the only heredoc writes the systemd unit (SERVICE_FILE, INI); there is no sed on config.toml anywhere. Verified by reading and now pinned by TestInstallerDelegatesConfigGenerationToArchiedSetup.
AC2 (invokes archied setup for a fresh install, skips an existing config): install.sh guards on , runs with the answers it collected, and prints '[SKIP] ... already exists (preserving user config)' otherwise.
AC3 (non-interactive install produces a valid config or fails loudly): verified empirically, not by eye -- ran the exact invocation install.sh builds (--defaults -config ... -forge-type github -forge-secret-ref env:ARCHIE_GITHUB_TOKEN -provider openai -model "") with stdin closed: exit 0, and the file it wrote loads through the real loader (configuration.New(nil).File). The failure arm is real too: .
AC4 (forge_block and its awk splice deleted): gone; the only awk uses left are go version go1.27.0 linux/amd64 parsing and NAME ID SIZE MODIFIED
glm-5.3:cloud 8477dab3e25b - 3 days ago
glm-5.3-flash:cloud 3e780905abc0 - 3 days ago
deepseek-v4.1-flash:cloud e04da138d31e - 3 days ago
nemotron-3-ultra:cloud fc1664751465 - 2 weeks ago
gemma4:e2b-it-qat 07ea59a47401 4.3 GB 2 months ago
gemma4:e4b-it-qat ee6656371218 6.1 GB 2 months ago
gemma4:31b-cloud c382fbfbc73b - 3 months ago
minimax-m3:cloud d03a959f45c0 - 3 months ago
nemotron-3-super:cloud be3943c5a818 - 4 months ago
glm-ocr:latest 6effedd0dc8a 2.2 GB 4 months ago
embeddinggemma:300m 85462619ee72 621 MB 4 months ago
qwen3.5:0.8b f3817196d142 1.0 GB 6 months ago
embeddinggemma:latest 85462619ee72 621 MB 6 months ago .

WHAT WAS MISSING, and is now added: nothing tested any of that, which is how the generating branch could grow back after the one time it drifted (hardcoded GitHub token key + Gitea host, so choosing Gitea produced a config naming a token the installer never wrote). TestInstallerDelegatesConfigGenerationToArchiedSetup now pins all four, with the splice and redirection checks scoped per line so legitimate sed/awk elsewhere stays allowed. Red-proofed in both directions: re-adding a [forge] table fails with 'install.sh carries a TOML template ("[forge]")', and re-adding a sed splice on config.toml fails with 'install.sh:N splices config.toml'.

Filed separately rather than folded in, because it needs a decision and not a patch: archie-core-6omt -- the non-interactive cloud path pairs the ollama baseline model with -provider openai (openai/llama3), which loads but is not a model OpenAI serves. Criterion 3 is met in the letter (valid config, and the failure arm is loud), so it is not a reopening of this bead; 6omt records the three candidate resolutions.


Imported from beads archie-core-1786637491134-51-17a55e60 · GitHub #72 · type task · priority P2 · created 2026-08-08 · closed 2026-09-21

Remove the TOML-generating branches from install.sh and call the binary it just built. Ends the class of bug where the installer's idea of the config schema drifts from the daemon's. ### Acceptance Criteria install.sh contains no TOML templates and no sed-patching of config files. It invokes archied setup for a fresh install and skips it when a config already exists. A non-interactive install still produces a valid config or fails loudly. The forge_block helper and its awk splice are deleted. ### Close reason Already satisfied on main before this bead was picked up -- verified criterion by criterion rather than assumed, and the missing half was the test that keeps it satisfied. AC1 (no TOML templates, no sed-patching): install.sh contains no TOML table header or key assignment at all; the only heredoc writes the systemd unit (SERVICE_FILE, INI); there is no sed on config.toml anywhere. Verified by reading and now pinned by TestInstallerDelegatesConfigGenerationToArchiedSetup. AC2 (invokes archied setup for a fresh install, skips an existing config): install.sh guards on , runs with the answers it collected, and prints '[SKIP] ... already exists (preserving user config)' otherwise. AC3 (non-interactive install produces a valid config or fails loudly): verified empirically, not by eye -- ran the exact invocation install.sh builds (--defaults -config ... -forge-type github -forge-secret-ref env:ARCHIE_GITHUB_TOKEN -provider openai -model "") with stdin closed: exit 0, and the file it wrote loads through the real loader (configuration.New(nil).File). The failure arm is real too: . AC4 (forge_block and its awk splice deleted): gone; the only awk uses left are go version go1.27.0 linux/amd64 parsing and NAME ID SIZE MODIFIED glm-5.3:cloud 8477dab3e25b - 3 days ago glm-5.3-flash:cloud 3e780905abc0 - 3 days ago deepseek-v4.1-flash:cloud e04da138d31e - 3 days ago nemotron-3-ultra:cloud fc1664751465 - 2 weeks ago gemma4:e2b-it-qat 07ea59a47401 4.3 GB 2 months ago gemma4:e4b-it-qat ee6656371218 6.1 GB 2 months ago gemma4:31b-cloud c382fbfbc73b - 3 months ago minimax-m3:cloud d03a959f45c0 - 3 months ago nemotron-3-super:cloud be3943c5a818 - 4 months ago glm-ocr:latest 6effedd0dc8a 2.2 GB 4 months ago embeddinggemma:300m 85462619ee72 621 MB 4 months ago qwen3.5:0.8b f3817196d142 1.0 GB 6 months ago embeddinggemma:latest 85462619ee72 621 MB 6 months ago . WHAT WAS MISSING, and is now added: nothing tested any of that, which is how the generating branch could grow back after the one time it drifted (hardcoded GitHub token key + Gitea host, so choosing Gitea produced a config naming a token the installer never wrote). TestInstallerDelegatesConfigGenerationToArchiedSetup now pins all four, with the splice and redirection checks scoped per line so legitimate sed/awk elsewhere stays allowed. Red-proofed in both directions: re-adding a [forge] table fails with 'install.sh carries a TOML template ("[forge]")', and re-adding a sed splice on config.toml fails with 'install.sh:N splices config.toml'. Filed separately rather than folded in, because it needs a decision and not a patch: archie-core-6omt -- the non-interactive cloud path pairs the ollama baseline model with -provider openai (openai/llama3), which loads but is not a model OpenAI serves. Criterion 3 is met in the letter (valid config, and the failure arm is loud), so it is not a reopening of this bead; 6omt records the three candidate resolutions. --- _Imported from beads `archie-core-1786637491134-51-17a55e60` · GitHub #72 · type task · priority P2 · created 2026-08-08 · closed 2026-09-21_
sam added the mediumtask labels 2026-10-27 13:36:54 +00:00
sam self-assigned this 2026-10-27 13:36:54 +00:00
sam closed this issue 2026-10-27 13:36:54 +00:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: sam/archie#580