القائمة
Run and exercise the FastAPI backend — start uvicorn, sweep the read endpoints with curl, check the diagnostic/fit/crop payloads against the live DB, and probe the admin write gate. Use when asked to run, start, test, or verify the API, a router, an endpoint, or a backend change.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Work off the Werkbank's Auftragskorb (the work_items basket) — read the open tasks, reproduce each complaint, restate it back into the row, triage it along the stage doctrine, fix the RULE rather than one spot, measure the effect and close the item with the diagnosed stage. Use when asked to work off the Korb, handle work items, process filed optimization tasks, or start an optimization round.
Take a finished change from diff to an open, green, review-clean PR — run the matching verify skills and the local CI gates first, then commit, push and open the PR, watch the pipeline AND the Copilot review, fix sensible findings and resolve the review threads. Never merges unless the user explicitly asks in this session. Use when asked to open or create a PR, ship a change, or finish up a change.
Test and smoke-run the pure-Python compute layer in core/ — run pytest, run ruff lint/format the way CI does, and call the extraction pipeline directly on a synthetic chart without DB or HTTP. Use when asked to run, test, lint, or verify core/, the pipeline, extraction, template, or fit code.
Run and visually verify the kurrentschrift app after frontend changes — start the dev servers, click through ONLY the flows the change touches (chrome-devtools MCP locally, or playwright-core against the pre-installed Chromium in the cloud), watch the console and network, check style fidelity and legibility at a desktop AND a mobile viewport, and record a performance trace when the change can move performance. Use when asked to run, start, screenshot, verify, test in the browser, or visually check the app or a frontend change.
Verify Alembic schema migrations locally against a throwaway PostgreSQL — full upgrade chain incl. seeds, model↔migration drift via alembic check, and a downgrade/upgrade roundtrip — without ever touching the shared Cloud SQL DB. Use when asked to verify, test, or check a migration, a schema change, or alembic revisions before pushing.
Conventions and checklist for writing or updating the internal design docs under docs/ — German language rules, where a new doc goes, keeping docs/index.md and the CLAUDE.md/copilot-instructions pair in sync, and what is settled (verworfen) and must not be re-litigated. Use when asked to write, add, update, or restructure documentation, concepts, or reference docs.
| name | verify-api |
| description | Run and exercise the FastAPI backend — start uvicorn, sweep the read endpoints with curl, check the diagnostic/fit/crop payloads against the live DB, and probe the admin write gate. Use when asked to run, start, test, or verify the API, a router, an endpoint, or a backend change. |
FastAPI service in api/ (routers in api/routers/), data in the
shared Cloud SQL Postgres (.env → DATABASE_URL). Local dev talks
to the SAME Cloud SQL DB as prod — there is no separate local DB.
Treat every write as touching real data. All paths relative to the
repo root; the harness is curl + python3 for JSON checks.
curl -fsS http://localhost:8000/health
If not running, start in the background (Bash run_in_background: true):
uv run uvicorn api.main:app --reload --port 8000
Expected: {"status":"healthy","database_configured":true}. If the
schema is stale: uv run alembic upgrade head — but only after the
DB-target preflight from the Gotchas below (alembic is DDL against
the shared DB). Without DATABASE_URL every endpoint except / and
/health returns 503.
OpenAPI UI: http://localhost:8000/docs (returns 200).
All of these are GET and safe against the shared DB. suetterlin-1922
is the public site's source (CONFIG.sourceId) and the most-authored
one, so sweep against it:
curl -fsS http://localhost:8000/styles | python3 -c 'import json,sys; d=json.load(sys.stdin); print([s["id"] for s in d])'
curl -fsS http://localhost:8000/sources | python3 -c 'import json,sys; d=json.load(sys.stdin); print([s["id"] for s in d])'
curl -fsS http://localhost:8000/hands
curl -fsS http://localhost:8000/quiz-words | python3 -c 'import json,sys; print(len(json.load(sys.stdin)))'
curl -fsS http://localhost:8000/sources/suetterlin-1922/bboxes/status | python3 -c 'import json,sys; d=json.load(sys.stdin); print(len(d), sum(1 for b in d if b["locked"]))'
curl -fsS http://localhost:8000/sources/suetterlin-1922/templates | python3 -c 'import json,sys; d=json.load(sys.stdin); print(len(d))'
Expected baseline: styles = ['kurrent', 'offenbacher', 'suetterlin']
(three seeded Grundvorlagen); sources = ['koch-1928', 'loth-1866', 'petzendorfer-1889', 'suetterlin-1922'] (four seeded chart sources);
hands = [] (normal at this stage); quiz-words ≈ 500. Bbox status +
templates list the authored glyph keys — the counts grow as glyphs get
authored, so check shape, not exact numbers.
The public write/render endpoints (the engine behind Federprobe, Tafel
and the quiz — a 200 with items/glyphs verifies core + DB + chart
file together, and both carry Cache-Control):
curl -fsS 'http://localhost:8000/sources/suetterlin-1922/write/word?text=lesen' | python3 -c 'import json,sys; d=json.load(sys.stdin); print(len(d["items"]), d["missing"])'
curl -fsS 'http://localhost:8000/sources/suetterlin-1922/write/glyphs?keys=n-medial' | python3 -c 'import json,sys; d=json.load(sys.stdin); print([g["glyph_key"] for g in d["glyphs"]], d["missing"])'
curl -fsS http://localhost:8000/sources/suetterlin-1922/templates/n-medial/diagnostic | python3 -c 'import json,sys; print(sorted(json.load(sys.stdin).keys())[:6])'
curl -fsS -o /dev/null -w '%{content_type} %{http_code}\n' http://localhost:8000/sources/suetterlin-1922/bboxes/n-medial/crop
Expected: word compose returns draw items (missing = [] for a fully
authored word); diagnostic keys include anchors_template,
outline_polygons, template_guides; crop returns image/png 200.
Write endpoints (PUT/DELETE bboxes, POST …/trace,
POST …/trace-preview, POST …/resample, DELETE templates) AND the
compute-heavy diagnostics GET …/fit + GET …/quality are gated by
require_admin (the two GETs cost seconds of CPU per call, so they are
gated like the writes — an unauthenticated /fit returning 401 is
correct, not a regression). The safe probe is the unauthorized request:
curl -s -o /dev/null -w '%{http_code}\n' -X PUT -H 'Content-Type: application/json' -d '{}' http://localhost:8000/sources/suetterlin-1922/bboxes/n-medial
curl -s -o /dev/null -w '%{http_code}\n' http://localhost:8000/sources/suetterlin-1922/templates/n-medial/fit
Expected: 401 for both with ADMIN_TOKEN configured — or 503 when neither
ADMIN_TOKEN nor Cloudflare Access is configured (require_admin
fails closed, api/auth.py); a 503 here means the env is missing the
token, not that the gate is broken. Do not send authorized writes
as part of routine verification — with ADMIN_TOKEN/X-Admin-Token they mutate the
shared Cloud SQL data that the admin UI authored. Only exercise an
authorized write when the task explicitly is about that endpoint, and
prefer a glyph key the user designates as scratch.
The API/compute test suite (synthetic fixtures, no DB or network needed) is the same one CI runs:
uv run --extra test pytest
Expected: all pass (a handful of tool-fixture tests skip when the
local-only bench fixtures are absent — that's normal). Check "all
pass", not an exact count; the suite grows continuously. CI
(.github/workflows/ci.yml) runs this plus ruff on every push/PR —
see the verify-core skill for the lint commands. The HTTP-level API
suites (tests/test_api_http.py, tests/test_api_admin_writes.py,
tests/test_api_auth.py) already cover the gate, the authorized write
paths and the CF-Access branch against in-memory SQLite — no live DB
needed for those.
The DB is shared. A locally "tested" PUT/POST/DELETE lands in the same Cloud SQL instance prod reads. The 401 probe is the only write-shaped request that is always safe.
Check which DB .env points at before ANY alembic/DDL command —
the local .env has pointed at the anyplot database before, and a
destructive migration against the shared DB once took prod down.
Preflight (must print user=kurrentschrift db=kurrentschrift,
anything else → stop):
python3 -c "
import re
url = next(l.split('=',1)[1].strip() for l in open('.env') if l.startswith('DATABASE_URL='))
m = re.search(r'//([^:]+):[^@]*@[^/]+/([^?\s]+)', url)
print(f'user={m.group(1)} db={m.group(2)}')"
Prod schema changes go through the kurrentschrift-migrate Cloud
Run job that the deploy pipeline executes (api/cloudbuild.yaml),
never ad-hoc DDL.
hands being empty is normal at the current MVP stage — don't
read it as a broken endpoint.
/diagnostic is slow-ish by design (it runs skeletonisation on
the chart crop per call); a sub-second JSON response is still the
norm locally. /fit + /quality are admin-gated for the same
reason — probe them for 401, don't expect public 200s.
Router prefixes nest under sources: bboxes and templates live at
/sources/{source_id}/…, not top-level. Seeded sources:
loth-1866 + petzendorfer-1889 (Kurrent), suetterlin-1922
(the public default), koch-1928 (Offenbacher).
uv run uvicorn … → "address already in use": the server is
already running (that's why §1 checks first); just use it./, /health) return 503: DATABASE_URL not
loaded — check .env exists in the repo root.