dev-bench
Skill0 starsChanged 3 months agoMITView on GitHub
Run tests or hands-on-test frappe_agents on the local frappe_docker stack. Use for any "run the tests", "try it on a bench", or "rebuild the bench" task.
In this repository6 other files it gives its agents.
- AGENTS.mdAGENTS.md
- docs/agents.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- releaseSkill
- semantic-commitSkill
- version-branchesSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 0 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 80
Proper (bake it): update the branch in apps.json if needed, then from fa-docker-bench/frappe_docker/, NOTHING ELSE HEAVY RUNNING:
- Very specific: line 84
AFTER any recreate of backend: docker restart fa-frontend-1 — its nginx resolves the backend hostname at startup and keeps the DEAD …
- Very specific: line 90
After ANY migrate or code swap while a human browses: bench --site <site> clear-cache AND a hard browser reload — sidebar/boot data is …
- Very specific: line 140
If it ever reaches a released tree the customer consequence is permanent: remove_orphan_entities deletes the standard icon on every …
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Dev bench (frappe_docker compose stack "fa")
The old hand-rolled bench (fa-bench + hand-run redis/serve/socketio) is RETIRED —
its uncapped bench build froze the Mac on 2026-08-16. This stack is the official
frappe_docker layered image with hard resource caps.
Everything lives in /Users/malik/Projects/fa-docker-bench/:
apps.json (canonical; frappe_docker/apps.json is a symlink), compose.yaml
(project name fa, every service has mem_limit/cpus — totals ≤6.5G/7cpu),
scripts/{setup_wizard,grant_roles,seed_agents}.py, build.log, tests*.log.
Two sites on the stack:
- test_site — the maintainer's hands-on site: UAE company (Falcon Trading LLC, AED), ERPNext demo data, users Administrator/admin + malik@leam.ae/admin. Never run the suite here — test records (13 _Test Company rows) already polluted it once.
- clean_test — pristine, allow_tests, no wizard/demo. All suite runs here.
Apps baked in the image fa-apps:v0.6.0-preview: frappe v16, payments, erpnext,
hrms v16, frappe_agents (integration/v0.6.0-preview), flow_client. Gotcha: the
flow_client REPO installs as app flow (--install-app flow). LEGAL: never
read flow_client source (AGPL vs our MIT) — install/run as black box only.
Daily driving
cd /Users/malik/Projects/fa-docker-bench
docker compose -p fa up -d # start (http://localhost:8010)
docker compose -p fa stop # stop, keeps everything
docker compose -p fa down # removes containers, KEEPS volumes — but
# also LOSES any docker-cp'd app files (see below)
docker compose -p fa logs -f backend
docker exec fa-backend-1 bash -lc 'cd /home/frappe/frappe-bench && bench --site test_site console'Login link without typing a password:
docker exec fa-backend-1 bash -lc 'cd /home/frappe/frappe-bench && bench --site test_site browse --user Administrator'
— take the sid, use http://localhost:8010/app?sid=... (ignore the broken host in
its output).
Running the suite
docker exec fa-backend-1 bash -lc 'cd /home/frappe/frappe-bench && bench --site clean_test migrate && bench --site clean_test run-tests --app frappe_agents'950 green on 2026-08-16 (v0.6.1; was 827 at v0.6.0). Check the COUNT moved as expected.
One module, and one test inside it:
docker exec fa-backend-1 bash -lc 'cd /home/frappe/frappe-bench && bench --site clean_test run-tests --module frappe_agents.tests.test_workspace_naming'
docker exec fa-backend-1 bash -lc 'cd /home/frappe/frappe-bench && bench --site clean_test run-tests --module frappe_agents.tests.test_workspace_naming --test test_sidebar_home_link_targets_the_workspace'pytest is deliberately NOT installed in the bench env — the harness suite is
*_test.py precisely so frappe's discovery ignores it. Do not pip-install pytest
here to run the harness; use CI, or a throwaway venv with pydantic+pytest only.
Syncing a worktree in: sync the TRACKED TREE, never a hand-picked file list.
An incremental docker cp of "the files I changed" left the container ahead of the
branch on 2026-08-16 and silently invalidated a whole suite run — the numbers were
for a tree that existed nowhere. Tar git ls-files 'frappe_agents/*', extract over
the app, delete any .py/.json/.txt the manifest does not list, then verify with
sha256 per file before believing any result. (macOS tar writes ._* AppleDouble
files into the archive; the delete pass sweeps them, or set COPYFILE_DISABLE=1.)
Updating frappe_agents code on the running stack
The layered image STRIPS .git from apps — you cannot fetch inside the container.
Two ways:
- Quick (container-local, lost on
compose down): from a local worktree,git diff --name-only <baked>..<new>thendocker cpeach file intofa-backend-1:/home/frappe/frappe-bench/apps/frappe_agents/<path>(+ chown frappe:frappe as root), then migrate + clear-cache both sites anddocker compose -p fa restart backend queue-short queue-long scheduler. - Proper (bake it): update the branch in
apps.jsonif needed, then fromfa-docker-bench/frappe_docker/, NOTHING ELSE HEAVY RUNNING:docker build --build-arg=FRAPPE_PATH=https://github.com/frappe/frappe --build-arg=FRAPPE_BRANCH=version-16 --secret=id=apps_json,src=apps.json --build-arg=CACHE_BUST=$(date +%s) --tag=fa-apps:v0.6.0-preview --file=images/layered/Containerfile .(~28 min) thendocker compose -p fa up -drecreates onto the new image. AFTER any recreate of backend:docker restart fa-frontend-1— its nginx resolves the backend hostname at startup and keeps the DEAD container's IP, serving 502/504 until restarted (bit us 2026-08-16). NOTE: current frappe_docker uses the secret mount, not APPS_JSON_BASE64 — the base64 build-arg is dead, docs forbid it.
After ANY migrate or code swap while a human browses: bench --site <site>
clear-cache AND a hard browser reload — sidebar/boot data is site-cached AND
client-cached (auto_generate_sidebar_from_module is @site_cache).
Hard-won rules
- REALTIME on frappe_docker: v16's websocket auth requires the request's Host
and Origin hostnames to be EQUAL, and then uses the Origin as the URL for its
session-validation fetch — stock frappe_docker's nginx template breaks both
for localhost setups ("Invalid origin" in the browser console; NOTHING live —
no streaming, no events). Fixed persistently here: patched
fa-docker-bench/frappe.conf.template(socket.io location pins Host=test_site, Origin=http://test_site:8080) bind-mounted over /templates/nginx/frappe.conf.template, plus atest_sitenetwork ALIAS on the frontend service so the auth fetch resolves. Alsosites/currentsite.txt= test_site. VERIFY REALTIME WITH A BROWSER, NOT CURL — curl passes the transport handshake without ever hitting the namespace auth that fails. - HOT-COPIES MUST REACH EVERY PYTHON CONTAINER: backend, queue-long, queue-short, scheduler each have their OWN container-local app tree. A docker cp to fa-backend-1 alone leaves chat runs (queue workers!) on old code — symptom: console-driven runs use new code, browser sends use old (burned us on the Responses wire). Copy to all four + restart, or rebuild.
- Fast JS iteration without an image rebuild: node lives at
/home/frappe/.nvm/versions/node/v24.12.0/bin in the image — docker cp the
source, then
export PATH=<that>:$PATH && bench build --app frappe_agents; the assets volume is shared, frontend serves the new bundle immediately. - One heavy docker operation at a time. Never
docker system prune. Never touch techmaze-/openconstructionerp- containers (other projects). - compose YAML: don't use folded scalars (
>) for multi-flag bench commands — the fold silently splitnew-siteflags into separate commands. Andset-config -p allow_tests truecrashes (literal_eval wantsTrue); plainset-config allow_tests trueper site is fine. - pkill in containers: bracket the pattern
[b]enchAND make sure no other unbracketed copy of the pattern appears in the SAME command line (an unbracketed pgrep later in the line self-killed the shell once). - Sidebar looks wrong after app work? Check
tabWorkspace Sidebar.app is "frappe_agents" — list views filter sidebars by app and fall back to the auto module sidebar ("Frappe Agents", hammer icon) when it's empty. And check the row EXISTS AT ALL: a migrate-only site had none until 0.6.1, because the builder only ever ran inafter_install. Symptom is the desk tile throwing "Icon is not correctly configured" while/app/agentsand every list view work. frappe.rename_docon a Desktop Icon DELETES the app's shipped JSON file.DesktopIcon.after_renamecallsdelete_desktop_icon_fileunconditionally, while the matchingexport_desktop_iconis gated ondeveloper_mode— so with developer_mode 0 the file is removed and nothing writes it back. This is the whole explanation for "the icon file vanished 4 times": our owntest_rename_patch_converges_a_half_renamed_sitewas doing it, and an IntegrationTestCase rollback cannot undo a filesystem unlink. Any test that renames a Desktop Icon must neutralisedelete_desktop_icon_fileor restore the file. If it ever reaches a released tree the customer consequence is permanent:remove_orphan_entitiesdeletes the standard icon on every migrate, andrestrict_removalis NOT enforced server-side (check_for_restrict_removalis defined in frappe and called from nowhere).
Last verified: 2026-08-16 — v0.6.0 RELEASED; 827 tests green on clean_test; running image = the released tree; apps.json now pins tag v0.6.0 for rebuilds.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

