CCE v2.4 Consolidation — dashboard refresh + documentation sweep + offline-first verification (SPEC)
The CLOSING task of the v2.4 milestone. CCE MCP (v2.4.0) and CCE Sync (v2.3.0) are both merged on main. This is a version bump to 2.4.1: additive dashboard/metrics work + a verified, gapless documentation sweep. Keep SYNC_FORMAT_VERSION = "2.3" UNCHANGED (decoupled from the app version) so the cross-engine sync golden 581cbd0f… stays byte-identical; bump only the app/crate version to 2.4.1. Single-repo conformance.json stays byte-identical. Offline-first is the most important guarantee.
The closing task of the v2.4 milestone (after CCE MCP (v2.4.0, merged) merges): a full sweep of the documentation so that — with everything from language packs → secret-scrubbing → workspaces → CCE Sync → CCE MCP now shipped — the docs are current, coherent, and gapless. A stranger must be able to install, set up (including CCE Sync and CCE MCP), and use CCE end-to-end from the docs alone, and — the most important thing — offline-first must work and be verified.
This is not "tidy the README." It is (1) a dashboard refresh so it surfaces the information that is now valuable, and (2) a verified, fresh-eyes audit across the whole documentation surface — both gated by a cold-start run (online and offline).
The dashboard was built at v1.1 and still shows only savings + retrieval quality from search metrics. Several capabilities have landed since; the dashboard must be brought up to date with what users actually want to see:
- Per-package / per-member breakdown (workspace, v2.2): savings, searches, and quality per member (the
by_packagesection) — where in the ecosystem is CCE helping most. - Agent vs human usage (MCP, v2.4): split CLI searches vs MCP/agent searches — how much is my agent actually leaning on CCE.
- Index freshness / sync status (Sync, v2.3): the indexed
sha, source (local vs pulled), and whether behind the remote — is my context current. - Secret-safety reassurance (v2.1): the sensitive-files-skipped count — the redaction is working.
- Review existing panels for continued value; drop or fix anything stale.
Enabling changes (spec them additively so old logs still parse):
- Extend the metrics event schema with the fields these panels need (e.g.
source: "cli" | "mcp"on search events;packageis already present; index events carrysha/source). Unknown/absent fields degrade gracefully. - Keep the dashboard loopback-only, read-only, self-contained (unchanged posture).
- Cross-language parity: identical
/api/metricsshape and panels in cce-ruby and cce-rust; refresh the committed dashboard screenshot.
README.md(single-repo · workspace · sync · MCP), install, quickstart, usage examplesdocs/: getting-started, architecture, workspace,sync.md, MCP/agents, how-to, dashboard, adding-a-language, DECISIONS, BENCHMARKS, VERIFIEDSECURITY.md,CONTRIBUTING.md,CHANGELOG.md,CITATION.cff,llms.txt,AGENTS.md
- Everything up to date. Every command, flag, and output example reflects shipped v2.4 behaviour — no stale references to removed/renamed things. Cross-file consistency: versions, repo URLs, and command names agree everywhere (the "consistency across files is the real work" lesson).
- Usage examples set. Worked, copy-pasteable examples with real captured output for each of: single repo · workspace/ecosystem · CCE Sync (
init/push/pull+ the CI recipe) · CCE MCP (init+ editor wiring + confirming the agent used it) · the dashboard. - Setup without gaps. Install + environment setup for macOS and Ubuntu, prerequisites explicit (toolchain, C compiler, git, git-LFS), each verified from a cold start. No "obvious" step left implicit.
- Easy install. The simplest install path front-and-centre, plus a one-command quickstart.
- Best practices for CCE Sync and CCE MCP. A dedicated "best practices" section: one sync repo per access boundary, CI as the canonical pusher,
.gitignore .cce/, when to use a workspace vs a single repo, wiring MCP + confirming usage via the dashboard, and the secret-safe-by-default posture. - Offline-first — THE most important. A dedicated, VERIFIED section proving every core workflow runs with no network and no remote:
index,search,stats,dashboard,workspace, and MCP against the local index. Explicitly document the only things that touch the network — the optional Ollama embedder,cce sync push/pull, and installing the binary/gem — and state plainly that everything else works fully offline.
- Fresh-eyes cold start (online): a reader who has never used CCE installs → sets up (incl. Sync + MCP) → uses it, from the docs alone, with zero friction.
- Offline cold start (mandatory): with the network disabled and no sync remote configured,
index+search+stats+dashboard+workspace+cce mcp(serving the local index) all work exactly as documented. Recorded indocs/VERIFIED.md. - Every documented code example runs verbatim (a doc example that doesn't run is a bug — fix the doc or the code).
- Cross-file consistency pass (versions, URLs, command names, feature lists).
llms.txtandAGENTS.mdreflect the full v2.4 surface.
- Dashboard refreshed: per-package breakdown, agent-vs-CLI split, index freshness/sync status, and sensitive-skipped count are shown; stale panels fixed/removed; metrics schema extended additively; screenshots +
/api/metricsdocs updated; cross-language parity. - All docs current and internally consistent (version/URL/command/feature checks pass).
- Worked, output-backed examples for single-repo, workspace, Sync, MCP, dashboard.
- macOS + Ubuntu install/setup verified from cold start; git-LFS covered.
- Best-practices section for Sync + MCP.
- Offline-first section, with a recorded offline cold-start run proving index/search/stats/dashboard/workspace/MCP work with no network.
-
docs/VERIFIED.mdupdated with both the online and offline cold-start transcripts.
- Runs after CCE MCP ( (v2.4.0, merged)) and CCE Sync merge; it is the final gate of the v2.4 milestone.
- Sibling repo (same sweep, same bar): davidslv/cce-ruby