couto/xkcd-search

refactor: optimize ingest upserts, modularize tests, and add CI workflow

refactor: simplify codebase, split MCP server, drop xkcd.com ingestion

feat(evals): add configurable conftest, NDCG@5 metrics, and hardened realistic dataset

refactor: simplify sampler to read directly from LanceDB

refactor: simplify evals with vanilla pytest, drop deepeval bloat, and add realistic search queries

feat: add deepeval retrieval evaluation suite and explainxkcd sampler

refactor: streamline codebase, prune unneeded tests and docs, polish SKILL and README per guidelines

fix: use explicit is not None check for table initialization

refactor: simplify src modules, remove server.py indirection, test public interfaces only

feat: modernize xkcd search with LanceDB, FastMCP native routes, and Hugging Face stack

fix: center comic images in search results

feat: run report-only eval step in nightly index workflow

feat: commit eval set of 40 query-to-comic pairs

feat: simplify search UI to clickable comic images

fix: stop 403-rate-limit boot crashes from the GitHub API

feat: add top-K input and rename app title to 'xkcd search'

feat: add evals CLI for hit@k and MRR retrieval metrics

fix: keep search button next to query box above results

feat: compose search app and MCP endpoint behind build_app()

feat: add search app surface with ComicCard and a Gradio UI

feat: add initial documentation for xkcd search language and concept

Add documentation for agent skills and issue tracking

Change to OpenCode

fix: move tenacity to main dependencies in pyproject.toml

ci: add OpenCode GitHub agent workflow (#4)

ci: add OpenCode GitHub agent workflow

Add CODEOWNERS file

Simplify retry decorator and add builder dependency group

Add retry logic for explainxkcd 503 errors (#2)

simplify: use FastMCP custom_route for landing page

fix: use Starlette directly to avoid FastAPI trailing slash redirect

feat: add browser landing page for MCP endpoint

docs: update README to reflect HF Spaces restart API instead of empty commit

docs: add sync comments for EMBED_MODEL across builder, Dockerfile, workflow

refactor: simplify codebase

fix: lint issues and make tests compatible with deployed index

fix: use --no-project flag for model download step before src copy

fix: restore fetch-depth=0 for full clone required by HF push

fix: add module docstring to __init__.py for uv_build compatibility

refactor: simplify cleanup from /simplify review

ci: replace FastMCP redeploy with HF Spaces restart + sync workflow

fix(docker): copy LICENSE so uv_build can package the project

migrate to Hugging Face Spaces; consolidate builder module

test: add remote smoke-test script for deployed MCP endpoint

fix(server): don't let transient refresh errors kill the poll thread

refactor: avoid duplicate int() cast in get_comic

refactor: drop unused upload_release; workflow uses gh CLI directly

test: add vcr cassettes for get_comic tool tests

docs: document get_comic tool and hosting migration path

feat: add get_comic tool for direct lookups by number