Merge pull request 'feat(claude): WSP local build memory, and structural terseness rules for the Soviet Engineer style' (#90) from feat/claude-wsp-build-memory-and-terseness into main
CI / flake (push) Successful in 4m23s

Reviewed-on: #90
This commit was merged in pull request #90.
This commit is contained in:
2026-08-13 13:59:37 +01:00
3 changed files with 96 additions and 0 deletions
+1
View File
@@ -14,3 +14,4 @@
- [Sandbox prompts](feedback_sandbox_prompts.md) — don't prompt for sandbox-disable or routine read-only shell ops; broaden permissions instead
- [Dev clusters disposable](dev_clusters_disposable.md) — Lyra's dev clusters are recreatable; mutate/break freely, no confirmation needed
- [Nix shell tooling](nix_shell_tooling.md) — any nixpkgs tool runs ad hoc via `nix run`/`nix shell nixpkgs#<pkg>`; a missing command is never a dead end
- [WSP local build and test](wsp_local_build_and_test.md) — core-services-cloud on this box: dotnet via nix, artifactory creds from `~/.artifactoryenv` sourced per command, how to tell auth failure from a code failure
@@ -0,0 +1,76 @@
---
name: wsp-local-build-and-test
description: "How to compile and test core-services-cloud locally on Lyra's NixOS/WSL box: dotnet via nix, artifactory creds from ~/.artifactoryenv, sourced per command"
metadata:
node_type: memory
type: reference
---
Canonical build/test commands for `core-services-cloud` live in the repo at
`.ai/agents.md` and `.ai/component-tests.md` — read those rather than guessing.
The repo docs assume Windows/PowerShell paths; this box is NixOS under WSL, so
the environment deltas below are what actually make them run.
**dotnet is not on PATH.** Get it from nixpkgs — see [[nix-shell-tooling]]:
```sh
nix shell nixpkgs#dotnet-sdk_8 --command dotnet build
```
`global.json` pins SDK 8 with `rollForward: minor`, so `dotnet-sdk_8` is the
right attribute.
**Every restore needs artifactory credentials.** They live in
`~/.artifactoryenv` (mode 0600) as `ARTIFACTORY_READ_ACCESS_USER` and
`ARTIFACTORY_READ_ACCESS_TOKEN`, consumed by `nuget.config`. Shell state does
not persist between tool calls, so source them inside each command:
```sh
set -a; . ~/.artifactoryenv; set +a
```
**Check the credentials before blaming the code.** A failed restore reports
`NU1301: Unable to load the service index`, which looks like a network fault but
is usually auth. Confirm which it is:
```sh
curl -s -o /dev/null -w '%{http_code}\n' \
-u "$ARTIFACTORY_READ_ACCESS_USER:$ARTIFACTORY_READ_ACCESS_TOKEN" \
https://repo.citrite.net/api/nuget/v3/stf-virtual-nuget/index.json
```
200 means the credentials are good. 401 means the token is the problem, not the
change under test. `https://repo.citrite.net/api/system/ping` returning `OK`
proves reachability independently of auth.
**Component tests** need Docker plus the same credentials, and are driven by
`./service.ps1` — PowerShell, so `nix shell nixpkgs#powershell` if `pwsh` is
missing. Log in to the image registry first:
```sh
echo "$ARTIFACTORY_READ_ACCESS_TOKEN" | docker login stf-virtual-docker.repo.citrite.net \
--username "$ARTIFACTORY_READ_ACCESS_USER" --password-stdin
```
Two Docker Desktop leftovers break this box, both fatal and both easy to miss:
1. `/usr/bin/docker` is a dangling symlink into an absent Docker Desktop WSL
mount, and it shadows the working NixOS docker inside `pwsh`. The script dies
with `Program 'docker' failed to run ... No such file`.
2. `~/.docker/config.json` sets `"credsStore": "desktop.exe"`, a helper that does
not exist. `docker login` reports success while storing nothing, then pulls
fail with `error getting credentials - err: exit status 1`. Remove the
`credsStore` key and log in again; docker then writes the auth into
`config.json` itself.
Put the real docker first when invoking anything that shells out to it, and note
`$PATH` must expand _inside_ the nix shell or dotnet drops off the path:
```sh
nix shell nixpkgs#dotnet-sdk_8 --command sh -c \
'export PATH="/run/current-system/sw/bin:$PATH"; dotnet test ...'
```
A feature canary used by a component test must also be registered in
`Automation/Component/ComponentTests/src/Citrix.Wsp.Test.Mocks/WspComprehensive/__files/unleash/unleash-test-environment.json`,
or `SetFeatureFlag` fails the test as inconclusive rather than failing loudly.
@@ -20,6 +20,25 @@ report? If the latter, rewrite. Retain all software-engineering capability and t
- Refer to the user as "comrade Lyra" when it reads naturally; do not force it into every line.
- No emojis.
## Length and form (the voice fails here first)
Terseness is structural, not just tonal. A dry register wrapped in report furniture —
headers, tables, a full status recap every turn — is the failure mode, and it passes a
tone-only self-check. Enforce:
- Default ceiling around 150 words. Longer only when the content genuinely needs it:
a real analysis, a comparison of options, a requested writeup.
- Headers and tables only for four or more distinct items. Two facts are two sentences.
- Report the delta since the last message, never the accumulated state. Assume Lyra
remembers what she was told.
- State each caveat once per session. Repeating a settled limitation is filler.
- Do the obvious next action and report it. Do not present a menu of options for a
decision that has an obvious answer.
- Do not restate the request, or narrate what is about to be done.
Self-check before sending: is this the delta, at the shortest length that stays accurate?
If it reads like a status report, cut it to the three facts that changed.
## Scope
The persona lives in PROSE ONLY — explanations, summaries, status, discussion. It must NEVER