v0.7.0 — context all-rounder + cross-run insight

The context all-rounder + cross-run insight. This release adds a layer of context-engineering tooling that keeps the agent's window lean on long, autonomous work, plus two read-only views into how your runs are going.

Highlights

• /ss-context — a standing-context budget cockpit. It measures your always-loaded footprint (CLAUDE.md, skill descriptions, STATE.md / CONTEXT.md) against a budget, verdicts OK / WARN / OVER, and warns you before you blow it — automatically, at session start. (context all-rounder, Front 1)

• ss-ctx runtime-output sandbox — two halves over one store:

  • a PostToolUse hook that transparently shrinks oversized clean Bash output, offloading the full text to a retrievable store;
  • a dependency-free Node MCP server exposing ctx_execute / ctx_batch_execute (run a command, keep its verbose output out of context), ctx_search / ctx_show (over the shared store), and ctx_fetch_and_index (fetch a URL, store the text, return a preview).

  So noisy command runs and fetched pages never flood the context window. (Front 2)

• /ss-stats — DORA-style cross-run analytics: a per-run table plus a rollup (gate-fail rate, skips, and an improving / worsening / flat trend).

• /ss-trace — a change's full provenance, joining its spec/plan docs, ledger gate events, and git commits into one chronological lineage.

30 skills · 1 MCP server. Everything ships bash + PowerShell twins (the MCP server is a single cross-platform Node process) and is covered by the self-test suite + CI.

Full detail in the Changelog. MIT licensed.