Data Model & the Workspace

Directory layout

rolester/
  candidate/          ← private, gitignored — your config and evidence
  workspace/          ← private, gitignored — runtime data and artifacts
  config/             ← public schemas and example files
  .agents/skills/     ← the 21 skill definitions (public)
  src/                ← core scripts and CLI (public)
  bin/                ← the CLI entry point

ROLESTER_HOME overrides where candidate/ and workspace/ are created. Set it to keep your data outside the repo tree.

candidate/ — private config

These files are gitignored and never shipped:

File	Contents
`profile.yml`	Identity, location, compensation floor and targets, domain/toolchain
`targeting.yml`	Role buckets, keep/cut signals, excluded companies, degree policy
`evidence.yml`	Accomplishment claims that feed tailored artifacts
`honesty.yml`	Tools confirmed, do-not-claim list, fabrication boundaries
`form-defaults.yml`	Applicant facts and expected-base for portal forms
`modes.yml`	Usage and application posture switches (optional)
`automation.yml`	Browser capability and platform consent switches (optional)
`writing-style.md`	Voice calibration for outbound artifacts
`stories.yml`	STAR+R behavioral story bank
`learnings/<family>.md`	Per-role-family durable lessons (compounding)
`AGENTS.md`	Personalized router generated by `ingest-profile`

workspace/ — runtime data

Also gitignored:

Path	Contents
`tracker.json`	The single source of truth for the dashboard
`tracker.html`	Latest static dashboard snapshot
`dashboard-data.js`	Browser-side data adapter (re-generated with each render)
`activity.jsonl`	Append-only Activity Pulse feed
`jobs/`	Saved JD bodies (`<slug>.md`)
`tailored/`	Tailored résumés, cover letters, short answers
`comms/`	Full recruiter / hiring thread bodies
`interview-prep/`	Interview packets and debriefs
`research/`	Company intel, comp benchmarks, board-discovery log
`intake/`	Sourced and triaged posting queue
`scan-results/`	Raw board / ATS scan output
`writing-samples/`	Voice-calibration writing samples
`captures/`	Throwaway browser screenshots (never committed)

tracker.json structure

tracker.json is the primary data contract between the agent and the dashboard. Its top-level shape:

{
  "meta": {
    "lastUpdatedAt": "2026-06-01T12:00:00Z",
    "lastSweepAt": "...",
    "version": 42
  },
  "applications": [...],
  "sourced": [...],
  "communications": [...],
  "strategyReview": {...}
}

applications[]

Each application row contains:

company, role, url, status — identity and pipeline position
fitScore, fitBucket, roleFit — body-read gate verdict and fit details
compEstimate, compNote — comp intel
interviewAt, nextInterviewAt, interviewNote — scheduled interview state
followUp — follow-up due date and draft
conversations[] — array of calls, interviews, and debriefs (typed by kind)
artifacts — paths to tailored documents, JD body, interview dossier

sourced[]

Roles discovered by search-jobs before the gate runs. Each entry has fitScore, fitBucket, fitBasis, and the saved JD body path. Promoted to applications[] after a KEEP gate verdict.

communications[]

Recruiter and hiring threads. Each has status, messages[], draft, and nextActionDue. The dashboard's Next Steps and CTA queue derives from these fields, not from the Activity Pulse.

The tracker is the authority

The dashboard renders exclusively from workspace/tracker.json. There is no separate database. If the dashboard shows stale data, re-render:

rolester tracker        # re-render from current tracker.json

Or run rolester tracker-dev for a live-reloading session.

Data Model & the Workspace

On this page