trendkit

Dev-trend crawler in Rust: snapshot GitHub trending signals into SQLite, score star velocity over time windows, and rank results by personal relevance against your own repository profile.

trendkit collect                 # snapshot trending new repos + refresh profile
trendkit report --top 25         # velocity table
trendkit report --personal -md   # markdown, ranked by relevance to your repos
trendkit profile update          # rebuild profile from WyattAu/* repos
trendkit profile show

Design

• Snapshot-first. GitHub exposes no star-history API and no official trending API; honest velocity requires your own snapshots over time. Every collect run appends immutable observations; report computes Δstars/day over a window, clamped to one hour minimum.
• Dogfooded. Built on the maintainer's own published crates: fetch-kit (HTTP), retry-backoff (search rate-limit schedule), config-kit (layered settings, redacted tokens), clock-kit (injectable time for deterministic tests), plus trendkit-core as the shared seam for future sources (hn, cratesio).
• Two clients, two policies. The GitHub Search API (30 req/min authenticated) gets manual pacing plus exponential backoff on 403/429 up to 60s; the core REST API (5000 req/h) keeps fetch-kit's built-in retries.
• Tokens never print. config-kit's Sensitive<String> renders [REDACTED]; .expose() is the only way in.

Configuration

Precedence: CLI flags > TRENDKIT_* env > trendkit.toml > defaults. Unknown top-level keys in the file are rejected (strict load).

# trendkit.toml
db_path       = "trendkit.db"
github_owner  = "WyattAu"
github_pages  = 3
window_days   = 14
# github_token = "ghp_..."   # or TRENDKIT_GITHUB_TOKEN / GH_TOKEN env

Status

v0.2 — GitHub Search + HN + crates.io sources over one unified subject/observation store (schema v2, migrated from v1), trending-page HTML scrape, velocity scoring, personal relevance, Telegram digest (report --telegram, token via /etc/alertmanager/telegram-token per fleet convention). Scheduled collection: deploy/trendkit-collect.* systemd units.

The crawlkit-engine library integration is deliberately deferred: its default feature set pulls wasmtime/sqlx/redis for a single HTML GET. Revisit when crawlkit ships a lean fetch facade.

Velocity is only meaningful after ≥ 2 collection runs; until then the report shows zero velocity by design rather than fabricating spikes from a single observation.