GitHub Dash

Local GitHub repository dashboard. Pass an org/repo on the CLI; the app opens a web UI for pull requests, commits, releases, and GitHub Actions metrics.

Stack

• pnpm · TypeScript · ESM
• React + MUI frontend
• Express API backed by Octokit
• CLI entry via Commander

Setup

pnpm install

Set a GitHub token (recommended — unauthenticated calls are rate-limited):

export GITHUB_TOKEN=ghp_...

The token needs repo (private) or public read access for public repos. Workflow metrics need Actions read permission.

Run

Build and start for a repository:

pnpm build
pnpm start -- facebook/react

Or:

pnpm start -- owner/repo --port 3847

Options:

Flag	Description
<repo>	Required. owner/repo or a github.com URL
-p, --port	HTTP port (default 3847)
-t, --token	GitHub token (else GITHUB_TOKEN)
-r, --release-workflow	Actions workflow file that creates releases (default release.yml, or RELEASE_WORKFLOW)
--no-open	Skip opening the browser
--host	Bind address (default 127.0.0.1)

Development

export GITHUB_REPO=owner/repo   # used by the API in dev
export GITHUB_TOKEN=...
pnpm dev

Vite serves the UI on :5173 and proxies /api to the Express server on :3847.

Features

• Overview — repo snapshot, attention strip (oldest PRs, flaky/slow workflows, release gap, orchestration), merge timing, pinned workflow cards, and charts
• Pull requests — list with merge timing (avg / median / p90), drill into description, reviewers, timeline, and commit checks (polls while pending)
• Commits — recent history, verification, author stats, file-level drill-down
• Releases — publish cadence timing; release notes parsed for #123 / PR URLs; associated PRs listed on the detail page; creating-run stats from the configured release workflow (default release.yml), matched by release tag; cached release-train / orchestration stage health
• Workflows — Actions success rate and duration stats with per-workflow flake streaks, conclusion strips, and duration sparklines; pin summary cards; click a workflow to see its past 100 runs, then open a run for job/step timings with longest job and step highlighted
• Failure analysis — on failed runs, Analyze failure uses the Claude Agent SDK to analyze each failed job and orchestration stage (with GitHub Actions job logs looked up via the API), plus release notes and stage TestReport data, returning per-failure and overall ranked likely PRs/authors; the UI streams progress/chat logs while the agent runs and collapses the log panel when finished (auth via ANTHROPIC_API_KEY or apiKeyHelper / env in ~/.claude/settings.json). Results are cached in-memory per run so revisiting a workflow run restores the prior analysis without re-calling Claude; Re-analyze forces a fresh run (?refresh=1)

GitHub responses are cached in-memory (CACHE_TTL_MS, default 5 minutes). Overview and orchestration health are warmed on server start. Claude failure analyses use a separate no-expiry in-memory entry per run (cleared on process restart or Re-analyze).

API

Endpoint	Description
GET /api/meta	Configured owner/repo
GET /api/overview	Aggregated stats + insights
GET /api/prs	PR list + stats
GET /api/prs/:number	PR detail
GET /api/prs/:number/checks	Check runs + commit statuses for the PR head (for polling)
GET /api/commits	Commit list + stats
GET /api/commits/:sha	Commit detail
GET /api/releases	Release list + stats
GET /api/releases/:id	Release detail + linked PRs
GET /api/workflows	Workflow runs + metrics (includes per-workflow ids for drill-down)
GET /api/workflows/by/:workflowId	Past 100 runs for a selected workflow
GET /api/workflows/:id	Run detail with jobs/steps and durations
GET /api/workflows/:id/analyze	Cached Claude failure analysis for a run (404 when none stored)
POST /api/workflows/:id/analyze	Claude Agent SDK failure analysis (tag → release notes + stage TestReports → likely PR/author). Add ?stream=1 (or Accept: application/x-ndjson) for NDJSON progress/chat events. Add ?refresh=1 to bypass the cache and re-run Claude
GET /api/orchestration/health	Cached release-workflow orchestration stage rollups