agent-cv

agent-cv — Your code tells a story

Most profiles stop at what GitHub shows publicly. Yours also lives on disk: branches that never got pushed, work behind NDA, weekend spikes in odd stacks, years of commits under addresses you forgot. agent-cv scans a folder tree (and optionally augments from GitHub), reads git history, READMEs, and dependency markers, then asks Claude Code, Codex, Cursor, or any OpenAI-compatible API you already run to describe each project. That work runs on your machine until you publish; you choose what becomes a public page at agent-cv.dev or a markdown file you own.

Try it locally:

npx agent-cv generate ~/Projects

Or publish it to the web:

npx agent-cv publish

Example profile →

How it works

1. Try locally

No account, no sign-up. npx downloads agent-cv, finds your projects, asks which to include, and generates AI descriptions. Everything stays on your machine.

npx agent-cv generate ~/Projects

2. Like it? Publish.

One more command to put it on the web. GitHub login for your username. You review everything before it goes live.

npx agent-cv publish

3. Customize

Edit your bio, reorder projects, hide what you don't want to show.

→ agent-cv.dev/yourusername/edit

Privacy

| Published | Never published | |---------------|---------------------| | Project names you included, AI-written descriptions, stacks, coarse activity | Paths on your machine | | Commit counts and dates | Source code or file contents | | GitHub links (public repos only) | Private repo URLs, secrets, .env, credentials | | | Raw email addresses |

Repo and CLI behavior are inspectable on GitHub. Legal: Terms of Service.

Features

Descriptions from real repos, not boilerplate
The model sees your tree and metadata; output is specific to that project.

Sees what GitHub cannot
Git history and folders are local first — private and internal work counts.

Nothing ships without you
Pick emails and projects before analysis; publish is explicit.

Re-run when life moves
New quarter, new job, new side project — run again; the profile catches up.

Also

Stack over time — languages, frameworks, activity patterns.

npx agent-cv stats ~/Projects

Markdown out — job forms, README “about me”, anywhere plain text wins.

npx agent-cv generate ~/Projects --output cv.md

Try it

No account. Runs locally.

npx agent-cv generate ~/Projects

npx agent-cv@latest … pins the newest publish if your cache is stale. Prefer a global install? npm install -g agent-cv, then agent-cv … instead of npx.

Commands

| Command | What it does | |---------|----------------| | agent-cv generate [dir] | Discover → analyze → markdown (stdout or --output) | | agent-cv publish [dir] | Full pipeline → agent-cv.dev (device login when needed) | | agent-cv login | Save GitHub credentials locally, no publish | | agent-cv config | Display name, bio, socials, telemetry prefs, email privacy | | agent-cv stats [dir] | Tech evolution from inventory + scan | | agent-cv diff <dir> | New / updated / removed since last scan | | agent-cv unpublish | Remove your page from agent-cv.dev |

All flags and AI: docs/CLI.md.

Inside one run

When you pass a directory (or resolve one from saved scan paths), generate and publish both run the same Pipeline component: scan, interactive pickers, per-project analysis, then a finishing step (significance tiers + profile insights) before the command-specific ending.

flowchart TD
  P["Directory on disk ± optional GitHub enrich"] --> S["Scan + merge saved inventory"]
  S --> K["Emails · projects · AI"]
  K --> A["Analyze projects · cache · privacy-aware context"]
  A --> F["Finish · tiers · profile insights"]
  F --> C{Command}
  C -->|generate| M[Render Markdown]
  C -->|publish| U["Confirm if needed → sync to agent-cv.dev"]

publish with no directory skips the pipeline: GitHub login, load the last saved inventory from local storage, confirm (unless -y), then sync — intended when you already ran a full scan/analyze (e.g. via generate) and only want to push.

generate may offer optional GitHub sign-in up front (cloud scan / later publish); you can skip it and stay local.

Defaults stick across runs so repeat visits feel like “update my profile”, not a full wizard. Persistent data lives under ~/.agent-cv/: pglite-data/ (projects and synced profile via PGlite) and state.json (scan paths, last agent, email picks). Override the base dir with AGENT_CV_DATA_DIR.

Local development: CONTRIBUTING.md.

License

Proprietary. Source available for reference. Use is subject to the Terms of Service.