Rolester

Agentic job-search workspace for finding, vetting, tailoring, tracking, and preparing for the right roles.

A local, skill-driven workspace: define what you actually want, turn it into searches, gate jobs against the real posting, tailor honest application artifacts, track outcomes, and prep for interviews — all from your own data, which never leaves your machine.

Tagline:

Find, vet, and advance the right roles.

Quickstart

Prerequisites:

• Node.js ≥ 18 — check with node -v
• A coding-agent CLI on your PATH — Claude Code or Codex:
  • Claude Code: npm install -g @anthropic-ai/claude-code (claude.com/claude-code)
  • Codex: npm install -g @openai/codex (github.com/openai/codex)

Get it running:

npm install -g rolester
rolester start claude    # or: rolester start codex

Developing from a source checkout is the same command shape; link the local binary once:

git clone https://github.com/CodesWhat/rolester
cd rolester
npm install
npm link
rolester start claude    # or: rolester start codex

That scaffolds your workspace, installs the skills, opens the dashboard at http://localhost:7777, and hands off to your agent. Then paste a job posting and say "evaluate this" — or try the bundled sample under examples/sample-jobs/. Run rolester next any time you want the terse next agent task.

Update later:

rolester update     # fetches the latest published code; your data is untouched

Nothing you enter leaves your machine. No account, no server, no build step.

Rolester is an agent runtime: the CLI sets up the workspace and dashboard, but the actual job-search work happens inside your agent reading the skills.

Useful flags: --no-agent (scaffold + dashboard only), --no-dashboard, --agent <name> (any CLI on your PATH, e.g. cursor), --port <n> (default 7777).

A 5-minute path to confirm the whole loop works

Once your agent is up, use it the way a real candidate would:

1. Let it onboard you. It builds a profile by asking a few questions. To kick the tires without using real details, just say: "set me up with a quick sample profile so I can test the flow."
2. Paste a job posting (a JD copied from anywhere, a posting URL, or the bundled sample in examples/sample-jobs/) and say "evaluate this." You should get a GATE / FIT / COMP / ACTION verdict from an actual read of the posting — not a keyword match.
3. Ask it to tailor: "write a resume and cover letter for this." It builds honest artifacts from the profile's evidence bank and refuses to invent facts.
4. Paste a recruiter email and say "draft a reply" — it routes to the comms flow and tracks the thread.
5. Open http://localhost:7777 and watch the application appear, move through the funnel, and accumulate activity. The dashboard is read-only — the agent does the work; the dashboard just shows it.

If those five land, the whole stack — CLI plumbing, skills, agent reasoning, and the live dashboard — is working end to end.

Prefer to let the agent run the test itself? Open claude or codex in this folder and say:

read the README and AGENTS.md, set yourself up, and walk me through testing Rolester end to end

What's normal, not a bug: until you onboard, rolester doctor reports that local candidate setup is incomplete and lists candidate/*.yml to create — that's the expected pre-setup state, and your agent fills it in during onboarding.

What It Is

Rolester is built around a simple rule: do not tailor or apply until the job has passed a real body-read gate. Title and keyword matches are triage, not truth.

The intended workflow:

1. Ingest the candidate profile, resume, evidence bank, and constraints.
2. Generate searches and source jobs into the sourced queue.
3. Read the JD body and emit a KEEP/CUT/REVIEW decision.
4. Tailor honest application artifacts from the evidence bank.
5. Draft and track recruiter/hiring communications.
6. Track applications and outcomes.
7. Build interview packets when a role advances.

What's Built

The full apply-cycle is shipped and working end-to-end:

• Guided onboarding (ingest-profile) — conversational interview that produces candidate/ config: targets, comp floor, evidence bank, honesty boundaries, writing-style calibration.
• Search & intake (setup-searches, research-boards, discover-companies, search-jobs) — build searches from your targets, find boards and employer ATS sources, dedupe, liveness-check, and triage the sourced queue.
• Body-read gate (evaluate-job) — reads the full posting; emits GATE / FIT / COMP / ACTION against your config before any tailoring.
• Honest tailoring (tailor-application) — résumé, cover letter, and short-answer artifacts built only from your evidence bank, with placeholder lint that blocks unresolved tokens.
• Apply assistant (apply-job) — portal form-fill recipes, manual-submit default, CAPTCHA pause.
• Communication memory (email-comms) — draft and track recruiter threads, follow-ups, scheduling, and negotiation.
• Interview prep (interview-prep) — audience-segmented packets grounded in your evidence; STAR+R story bank; live and written negotiation coaching.
• Outcome tracking & strategy (track-outcomes, reevaluate-strategy) — records results; triggers strategy review when rejection or advance thresholds trip.
• Research loop (research-company, research-comp, research-boards) — company intel, comp benchmarking, and board discovery with a citation-tier firewall so web findings never launder into résumé claims.
• Tracker dashboard — stat cards, funnel, Active Pipeline (Sourced → Offer), All-Jobs table, per-job detail, follow-up reminders; Table ⇄ Board ⇄ Calendar views; Tokyo Night and Gruvbox theme families.
• Learning memory — per-role-family lessons compound across loops; tailoring and evaluation get sharper each time.

Health check:

rolester doctor
rolester next

The dashboard

rolester start brings the dashboard up for you. To run it on its own:

rolester tracker        # render workspace/tracker.html once — a static snapshot
rolester tracker-dev    # serve http://localhost:7777 with live reload

rolester start [agent] starts the live server as a separate local process and records its PID/log at .internal/tracker-dev.pid and .internal/tracker-dev.log, so the dashboard keeps serving while the agent does the work.

rolester tracker publishes the dashboard shell to workspace/tracker.html and its browser data adapter to workspace/dashboard-data.js. Use rolester tracker-dev for the live command center: it serves the shell, the adapter, and workspace/tracker.json, watches tracker data plus dashboard source, re-renders through the same path as rolester tracker, and refreshes the open page over Server-Sent Events.

Product Notes

The roadmap lives in docs/ROADMAP.md. The source strategy lives in docs/SOURCES.md.

Naming decision: Rolester.

Why: it has the Napster/Friendster feel, it is role-specific, and it avoids the generic "jobster" lane while still making sense for a job-search workspace.

Prior candidates now treated as alternates: Prospectr, Prospector, RoleRelay, CareerRelay, RoleProxy, See-V, and RoleScopeGo.