Record the site serve and view commands and what building them taught #70

Merged

vnprc merged 1 commit from agent/site-serve-notes into master

2026-09-28 16:07:43 +01:00

allod-agent commented

2026-09-28 15:48:55 +01:00

Member

Memory now lists the two new site commands and records what building them taught. Nothing changes on any machine: these are notes agents read.

What changes: allod.md lists allod site serve and allod site view, and says that allod change submit does not push. Six gotchas and one naming rule go to the topic files that own them; two sightings go to the ledger.
What stays identical: memory.md apart from one index description, every skill, and every existing entry.
What proves it: .hooks/memory-budget --worktree exits 0. Each gotcha was measured in the session that implemented allod/tools#240; the two version numbers in nix.md were measured again before writing.
If it is wrong: revert the commit, or correct the one entry.

Refs allod/tools#240

What was added, and where

File	Entry
`allod.md`	`allod site serve` and `allod site view` in Workspace Tools. `allod change record` pushes even with nothing new to commit; `allod change submit` does not push and answers an unpushed branch with `HTTP 404: The target couldn't be found`
`cli-design.md`	Command names: the verb for what the command does where it runs, taken from the domain's own tools when they agree, which is why `site preview` became `site serve`
`shell.md`	`systemctl` exit codes that read as absence: `stop` exits 5 for a unit that is not loaded; `is-active` exits 0, 3, 4 or 1; `list-units ... \|\| true` hides a failed probe
`shell.md`	`websocat` in line mode does not show a zero-length frame, which is live-server's whole reload signal
`nix.md`	`nixpkgs#<package>` answers for the machine's registry, not the repository's pin; `--inputs-from <checkout>` asks the pin
`nix.md`	`nix run` writes a `flake.lock` into a checkout that has none, `builtins.getFlake` does not; `nix eval --expr` does not call a function with `--argstr`
`git-workflow.md`	After merging `master` into a branch, run the checks: a name one side deleted and the other began to use merges cleanly and does not compile
`ledger.md`	Two sightings: a typed field in a shared decoder fails the whole file for every reader; a command that finds a repository by its registry checkout has to be walked from a worktree
`memory.md`	The index line for `cli-design.md` now mentions command names

Risk

R0. Notes only.

Validation

.hooks/memory-budget --worktree: exit 0. nix.md is 21,980 bytes against its 24 KB cap.
nix eval --raw nixpkgs#zola.version printed 0.22.1 and nix eval --raw --inputs-from <allod/tools checkout> nixpkgs#zola.version printed 0.23.6, run again for this change.
The systemctl codes are the ones measured for allod/tools#242, on systemd 260.4.

No external review pass: R0.

Memory now lists the two new site commands and records what building them taught. Nothing changes on any machine: these are notes agents read. - **What changes:** `allod.md` lists `allod site serve` and `allod site view`, and says that `allod change submit` does not push. Six gotchas and one naming rule go to the topic files that own them; two sightings go to the ledger. - **What stays identical:** `memory.md` apart from one index description, every skill, and every existing entry. - **What proves it:** `.hooks/memory-budget --worktree` exits 0. Each gotcha was measured in the session that implemented allod/tools#240; the two version numbers in `nix.md` were measured again before writing. - **If it is wrong:** revert the commit, or correct the one entry. Refs allod/tools#240 ## What was added, and where | File | Entry | | --- | --- | | `allod.md` | `allod site serve` and `allod site view` in Workspace Tools. `allod change record` pushes even with nothing new to commit; `allod change submit` does not push and answers an unpushed branch with `HTTP 404: The target couldn't be found` | | `cli-design.md` | Command names: the verb for what the command does where it runs, taken from the domain's own tools when they agree, which is why `site preview` became `site serve` | | `shell.md` | `systemctl` exit codes that read as absence: `stop` exits 5 for a unit that is not loaded; `is-active` exits 0, 3, 4 or 1; `list-units ... \|\| true` hides a failed probe | | `shell.md` | `websocat` in line mode does not show a zero-length frame, which is live-server's whole reload signal | | `nix.md` | `nixpkgs#<package>` answers for the machine's registry, not the repository's pin; `--inputs-from <checkout>` asks the pin | | `nix.md` | `nix run` writes a `flake.lock` into a checkout that has none, `builtins.getFlake` does not; `nix eval --expr` does not call a function with `--argstr` | | `git-workflow.md` | After merging `master` into a branch, run the checks: a name one side deleted and the other began to use merges cleanly and does not compile | | `ledger.md` | Two sightings: a typed field in a shared decoder fails the whole file for every reader; a command that finds a repository by its registry checkout has to be walked from a worktree | | `memory.md` | The index line for `cli-design.md` now mentions command names | ## Risk R0. Notes only. ## Validation - `.hooks/memory-budget --worktree`: exit 0. `nix.md` is 21,980 bytes against its 24 KB cap. - `nix eval --raw nixpkgs#zola.version` printed `0.22.1` and `nix eval --raw --inputs-from <allod/tools checkout> nixpkgs#zola.version` printed `0.23.6`, run again for this change. - The `systemctl` codes are the ones measured for allod/tools#242, on systemd 260.4. No external review pass: R0.

allod-agent added 1 commit

2026-09-28 15:48:55 +01:00

Record the site serve and view commands and what building them taught 06b09c05de

allod.md lists the two commands and says which of record and submit
pushes. The gotchas go to the topic files that own them, the naming rule
to cli-design.md, and two sightings to the ledger.

Refs allod/tools#240