Common dev tasks¶

Cheat-sheet for the things you'll do most often.

Reset your local data¶

/settings/reset in the running app. Or directly:

DELETE FROM attempts            WHERE user_id = '<your-uuid>';
DELETE FROM content_progress    WHERE user_id = '<your-uuid>';
DELETE FROM assessment_runs     WHERE user_id = '<your-uuid>';
DELETE FROM reflections         WHERE user_id = '<your-uuid>';
-- … keep jds / writings / feedback / credentials
UPDATE users SET settings_json = settings_json - 'onboarding_skipped'
WHERE id = '<your-uuid>';

Inspect the feedback inbox¶

cat data/feedback/inbox.md    # markdown digest (re-generated on every action)

Or the web UI: /feedback/inbox (filter by status: open / backlog / closed).

Test a single Claude call without booting the app¶

# python
from anthropic import Anthropic
from app.services import hints, review, voices

# Hints with a real Problem
from app.repositories.problem_repository import ProblemRepository
from app.db import conn
p = ProblemRepository(conn).get_by_slug("softmax-temperature")
print(hints.get_hint(p, "explore", "shapes: x is (B, V)", code=None, count_so_far=0))

# Review with a voice
# (need an Attempt object; load one or use the sample_attempt fixture in tests)

Re-run only one test file¶

.venv/bin/pytest tests/unit/test_voices.py -v

Tail the logs¶

.venv/bin/uvicorn app.main:app --port 8765 --log-level debug

Sentry sends events on uncaught exceptions (SENTRY_DSN in .env.local). Set it blank to disable.

Force-reload content (skip the content-hash cache)¶

LIS_FORCE_RELOAD=1 .venv/bin/uvicorn app.main:app --port 8765

Skip the startup content loader (for tests)¶

LIS_SKIP_STARTUP_LOAD=1 .venv/bin/pytest

Already set in tests/conftest.py for all unit tests.

Switch active JD¶

# In settings_json:
settings["active_jd_id"] = <jd_id>

Or via the UI: /jds/<id> → Activate.

Cache invalidation¶

The in-process cache (app/cache.py) lives on the running process. Repository writes invalidate by prefix (e.g., content_items:* after a new item). To wipe manually in dev:

from app import cache
cache.clear_all()

Where each LLM model is set¶

Each service reads its model from an env var with a default:

Service	Env var	Default model
Review	`ANTHROPIC_MODEL`	`claude-sonnet-4-5`
Hints	`ANTHROPIC_HINT_MODEL`	`claude-haiku-4-5-20251001`
External capture parse	`ANTHROPIC_PARSE_MODEL`	`claude-haiku-4-5`
Hamming (reflect)	`ANTHROPIC_HAMMING_MODEL`	falls back to `ANTHROPIC_MODEL`

Override in .env.local to test against a different model.