internal/

Package assets embeds the static frontend assets — the Tokyo Night theme, the v1 report stylesheet, and the vim navigation script — so the renderer and server can ship them without a separate file dependency at runtime.

Package beads is a read-only adapter over the `bd` (beads) issue-tracker CLI.

Package config loads harness-deck's settings: where reports live, which project roots to scan, and how to notify on responses.

Package jsonfile centralizes the repo's JSON-on-disk write discipline: atomic writes through unique temp files, and read-modify-write patches that preserve unknown fields, keep number literals byte-exact, and never clobber an existing-but-unparseable file.

Package manifest defines the harness-deck report manifest: the JSON document a harness writes to publish a report.

Package mcp implements a stdlib-only Model Context Protocol server for harness-deck.

Package notify runs the user-configured notification command when a response is recorded, so the user (or harness) learns that an answer landed.

Package projects discovers project roots under configured scan roots and tracks which ones the user keeps visible.

Package push delivers Web Push notifications to subscribed browsers, implementing VAPID (RFC 8292) authorization and the aes128gcm content encoding (RFC 8291 + RFC 8188) on top of the Go standard library — keeping harness-deck dependency-free.

Package query implements a JQL-like search query language for harness-deck: a lexer, a recursive-descent parser, a predicate AST, and a lazy short-circuit evaluator.

Package render turns a validated report manifest into a complete, themed HTML page in the v1 TUI dashboard style.

Package respond manages responses.json — the file harness-deck writes into a run directory to record the user's answers, decisions, and approvals so the harness can pick them up on its next session.

Package server runs the harness-deck dashboard: the aggregator shell, the report-index API, and individual rendered report pages.

Package store discovers report manifests on disk and keeps an in-memory index of them.

Package usage reads AI coding-tool usage — rate-limit windows or credit/spend — from each tool's local files and HTTP APIs, and exposes a normalized snapshot the dashboard footer renders (CodexBar-style).