Skip to content

ARCHITECTURE.md

Latest commit

 

History

History
135 lines (95 loc) · 4.96 KB

ARCHITECTURE.md

File metadata and controls

135 lines (95 loc) · 4.96 KB
Raw

Architecture

Rolester is organized around skills plus deterministic scripts.

Flow

candidate profile  ──────────────────────────────────────────────────────┐
  -> search setup                                                         │
  -> sourced scan                                                         │
  -> gated intake  <── research loop (research-company / -comp / -boards)│
  -> body-read evaluation                                                 │
  -> tailoring                                                            │
  -> communication tracking                                               │
  -> application tracking                                                 │
  -> interview prep                                                       │
  -> outcome tracking ──> reevaluation loop (reevaluate-strategy) ───────┘

The research loop feeds evaluation and interview prep with cited company intel and comp benchmarks. The reevaluation loop reads the full funnel and recommends targeting, fit-calibration, or channel-mix changes when outcome thresholds trip.

Intent Router

Root AGENTS.md and CLAUDE.md instruct agents how to map user intent to the 16 skills. Routes are grouped below by cluster.

Onboarding

  • New workspace or incomplete profile → ingest-profile
  • "Scan my projects folder / repo" → ingest-profile (projects-scan mode)
  • "Sync / import email, pull recruiter replies" (Apple Mail or opted-in Gmail/Outlook webmail) → ingest-mail
  • "Sync / check LinkedIn or Wellfound messages/DMs" (opt-in browser) → ingest-messages

Apply Cycle

  • "Apply", "submit", or a JD URL with apply intent → apply-job (apply-job must run or verify evaluate-job as step zero; LinkedIn Easy Apply postings can use the opt-in authenticated one-click path gated by the one_click_apply capability — no new skill, stays within apply-job)
  • "Gate", "should I apply", or a JD URL without apply intent → evaluate-job
  • "Find jobs", "source", "search", "refresh queue" → search-jobs
  • "Set up searches", "build search config" → setup-searches
  • "Tailor résumé / cover letter / short answer" (after gate passes) → tailor-application

Research Loop

  • "Research this company" / company URL or name pasted → research-company
  • "What's market comp", "benchmark my salary ask" → research-comp
  • "Find more boards", "what sources should I add", stale queue → research-boards

Communications

  • Email or recruiter message: draft, reply, follow-up, thank-you, scheduling, written negotiation counter → email-comms
  • Live / verbal offer call, real-time negotiation coaching, rehearsal → interview-prep

Interview & Story Bank

  • Interview invite, screen, panel, assessment → interview-prep
  • Building behavioral / STAR stories, story bank → interview-prep

Outcomes & Strategy

  • Rejection, offer, status change, outcome update → track-outcomes
  • "Check my statuses", "sync my pipeline", poll ATS dashboards → sync-status (opt-in browser automation; reads portals, hands transitions to track-outcomes)
  • "Why am I getting filtered", strategy review, re-rank, or when outcome thresholds trip → reevaluate-strategy

Settings

  • "Change a setting", comp floor/target, exclusions, writing style, form defaults, search sources, usage mode, application mode, browser automation, or session browser → configure

apply-job must run or verify evaluate-job as step zero.

Layers

Skill Layer

Skills make judgment calls:

  • what to ask during onboarding
  • whether a job passes the body-read gate
  • how to rate fit
  • which evidence should support an application
  • how to draft and summarize candidate communications
  • what to include in an interview packet

Script Layer

Scripts should be deterministic:

  • validate setup
  • build source URLs from config
  • parse saved jobs
  • dedupe sourced roles
  • check link liveness
  • render documents
  • validate tracker state

Communication Layer

Tracker state stores concise communication thread metadata. Longer message bodies and summaries live in workspace/comms/.

email-comms should read both before drafting so the user does not need to re-provide thread history.

Source Layer

Search sources are provider adapters under src/core/providers/.

  • URL-query sources build stable URLs from config before Playwright opens them.
  • RSS/Atom sources should poll feeds before browser capture.
  • ATS sources should use stable public endpoints where possible.
  • Browser-rendered sources should preserve the generated URL, raw capture, and exact recency cutoff.

See SOURCES.md.

User Layer

Candidate facts, generated artifacts, and tracker state stay local.

System Layer

Reusable skills, scripts, templates, and schemas are public-safe.

See ../DATA_CONTRACT.md.