Build a runnable reproduce→fix→verify loop for a bug or feature in a fresh sandbox, and record it as a feedback-loop doc. Use when investigating a reported bug, validating a root cause, or proving a fix works — before and after the change.
The English summary is being prepared. Showing the original description.
Run score Not tested
See how to installReview the instructions the agent will receive before you install.
A feedback loop is a runnable document: anyone (human or agent) can re-execute
it in a fresh sandbox and observe the same failure, then the same pass after the
fix. Examples of the format live in docs/feedback-loops/ (e.g.
save-button.md, fabric-obo-second-admin-tables.md).
file.py:line references.docs/feedback-loops/<topic>.md (do NOT add new
sandbox-feedback-loop-*.md files at the repo root).The app targets Python 3.13 (3.12+ works; needs 3.12 f-string syntax; the sandbox default
python may be 3.11).
cd backend
pip install uv
uv sync --frozen --extra dev
export BOW_DATABASE_URL="sqlite:///db/app.db" # required by bow-config.dev.yaml
mkdir -p db
run_migrations fixture builds
the schema per test (tests/conftest.py). Use --db=postgres for the
testcontainers leg, --db=external when Docker is unavailable./opt/pw-browsers in cloud
sandboxes — export PLAYWRIGHT_BROWSERS_PATH=/opt/pw-browsers, never
playwright install.tools/agent/boot_stack.sh then cd backend && uv run python ../tools/agent/seed_org.py.# Feedback Loop — <symptom, quoted from the report>
One-paragraph statement of the reported behavior and the claim being validated.
## Root cause (validated)
What actually breaks, citing file:line. Distinguish validated facts from hypotheses.
## Loop A — deterministic reproduction (no external services)
Exact commands + the observed FAIL output. Stub external boundaries (LLMs,
OAuth providers, ODBC) — the loop must run in a clean sandbox.
## Loop B — live confirmation (optional, real credentials)
Only when the premise itself needs a real third party. Secrets via env vars
only — never commit them, never echo them into logs or docs.
## The fix
What changed and where. Re-run Loop A output showing the flip to PASS.
## What this proves / regression notes
What the loop demonstrates; any pre-existing unrelated failures you hit
(verify they reproduce with your change stashed before calling them unrelated).
backend/tests/AGENTS.md: assert the general invariant, not the one magic
scenario that happened to be reported.No reviews yet. Tell others what you thought.
No login needed. Share how it worked for you, ask a question, or note where it did not run.
No comments yet. Be the first to write one.
Add a new data source / connection type (e.g. a SQL Server-like database) end to end — client with get_schemas/execute_query, config + credentials schemas, registry entry, icon, tests — plus the verification steps. Also covers adding or updating an MCP connector (a preset, not a new type). Use when adding or significantly extending a connector.
The English summary is being prepared. Showing the original description.
Add a new LLM model to the preset catalog, or a whole new LLM provider — with the mandatory pre-flight verification of model id, pricing, and context window against the provider's official docs. Use when asked to add/update LLM models, providers, or pricing.
The English summary is being prepared. Showing the original description.
Query, visualize, and analyze data in a Bag of Words (BOW) workspace via the bagofwords MCP tools. Use for creating reports, running tracked queries, and building dashboards.
The English summary is being prepared. Showing the original description.
Update the product docs at docs.bagofwords.com (Mintlify) with text and fresh screenshots after a user-facing change ships. Use when a merged change alters user-visible behavior, adds a feature, or when asked to update/refresh documentation.
The English summary is being prepared. Showing the original description.
The locale/i18n architecture of bagofwords — catalogs, resolution order, RTL, backend contracts — and the procedures for adding strings, adding a locale, or translating UI. Use when adding/changing user-facing strings, working on RTL, emails, localized errors, or anything locale-related.
The English summary is being prepared. Showing the original description.
Run a live QA pass over the app — first map all user-facing functionality, then boot the full stack and manually exercise flows with Playwright, recording pass/fail evidence and filing a QA report. Use for release QA, post-merge smoke passes, or "QA the app / this area" requests.
The English summary is being prepared. Showing the original description.