Skip to content

ROADMAP.md

Latest commit

 

History

History
170 lines (104 loc) · 7.16 KB

ROADMAP.md

File metadata and controls

170 lines (104 loc) · 7.16 KB
Raw

CareerOps roadmap

Phase 1 (live on careerops.telivity.app): bullet memory with provenance, portfolio, career advisor, ranked Generate, board-pack v2 — see DOCTRINE_MEMORY.md.

Phases 2–4 turn the search pipeline into a durable operating system without becoming another spammy AI job tool.

Doctrine (unchanged): no invented experience; no auto-apply / no auto-send; AI polish only with accept; labeled market/salary judgment; enrichment lands as inbox candidates. See DOCTRINE_MEMORY.md.

Reuse Phase 1: mt_reports kinds, resume sync, board pack, accomplishment inbox/accept, skill modes (interview, followup, outcome, advise).

Prerequisite — durability (shipped)

Move browser-local career truth into Supabase so analytics and multi-device work. Without this, Phase 2 analytics lie.

Ship (done):

  • Outcomes, story bank, and Sent / version display metadata synced to Supabase with RLS
  • Local fallback when offline / unauthenticated
  • One-time migrate from existing localStorage data
  • Board pack schema_version 3 — durable outcomes, Sent markers, version display names, stories

Acceptance: refresh and multi-device see the same Sent flags, outcomes, and story bank; pack export/import round-trips them.

Phase 2 — Decide / Interview / Offer depth

Goal: make the kanban’s right half as serious as Find → Write.

Order: durability (done) → interview events → offer schema/UI → version analytics (analytics needs Sent + outcomes in DB).

2.1 Resume version analytics

Today: append-only mt_reports resume/cover/match + Sent freeze; no outcome join in the builder.

Ship:

  • Sent markers + version display names already in DB / pack (durability)
  • Per-role version timeline: match score over time, which version was Sent, linked outcome
  • One “What shipped” panel in builder/drawer — no ML, no A/B magic

Out of scope: A/B testing frameworks, auto-picking a “winning” resume version.

Acceptance: user can answer “which resume version did I send, and did that role convert?”

2.2 Interview tracking

Today: interview stage, ~14d follow-up strip, ephemeral prep drafts, story bank, skill interview / followup.

Ship:

  • Table mt_interview_events: role_id, round, scheduled_at, type (screen / onsite / loop / …), notes, optional interviewer name
  • Drawer: add/edit rounds; persist prep drafts as mt_reports.kind='interview'
  • Follow-up due strip reads real event dates when present
  • Board pack carries interview events (further schema bump after v3)

Out of scope: calendar OAuth, auto-send thank-you mail, invented interviewer bios.

Acceptance: rounds survive refresh/devices; prep is recoverable; still copy-only drafts.

2.3 Offer comparison

Today: offer stage + unstructured outcome notes.

Ship:

  • Structured outcome/offer fields: base, bonus, equity_notes, remote, deadline, currency, free-text note
  • Side-by-side compare for 2–3 roles in Offer (and saved outcomes)
  • Skill outcome reads/writes structured fields via board pack

Out of scope: invented comp numbers, third-party salary APIs.

Acceptance: compare two offers without a spreadsheet; no invented comp numbers.

Phase 2 done when: track interview rounds, compare offers, see which resume version was Sent and what happened.

Phase 3 — Network, ATS I/O, honest comp

Goal: expand surface without violating no-send / no-invent.

Order: contacts table → comp_range on roles + pack import → target band compare UI.

3.1 Recruiter CRM (no integrations that send)

Today: Conversation stage, outreach/email drafts only.

Ship:

  • mt_contacts: name, channel, company, linked role_id(s), last_touch_at, notes
  • Log touches when user copies outreach/email (explicit “Log touch”)
  • Simple contact list filtered by role/company

Out of scope: Gmail/Outlook OAuth, auto-send, sequences. “Integrations” here means draft + log, not mailbox sync.

3.2 ATS import/export

Today: strong Find scrape + CSV/JSON/board-pack export; weak structured re-import.

Ship:

  • Board-pack import upserts roles/materials/accomplishments/portfolio (keys never imported) — schema migrator already started in v2+
  • Persist optional comp_range / comp_raw on mt_roles from boards that expose it
  • Document export formats for self-hosters

Out of scope: applying into employer ATS; application-status webhooks.

3.3 Salary intelligence (posted truth, not a market fantasy DB)

Today: materials-only negotiation draft; salary noise stripped from match keywords.

Ship:

  • Show stored posted band on cards/drawer when present
  • User target band on profile
  • Compare posted vs target vs structured offer — label gaps as data, not “market average”
  • Advisor may discuss negotiation only from these numbers + user notes (grounded follow-ups inside Advise — no freeform chat)

Out of scope: third-party salary API in v1 (paid, noisy, invents confidence). Revisit later as optional plugin.

Phase 3 done when: remember recruiters, import/export full career pack, see posted pay vs your target.

Phase 4 — Enrichment, extensibility, multi-agent

Goal: OSS-power-user leverage without silent claim invention.

Order: enrichment inbox (reuses Phase 1 accept UX) → pack/plugin registry → mode chains.

4.1 GitHub / LinkedIn enrichment (user-confirmed)

Pattern: same as polish/accept and accomplishment inbox.

Ship:

  • Paste GitHub repo or LinkedIn profile/job URL → fetch public metadata only where ToS-safe (or client-side paste of README/About text if fetch blocked)
  • Propose mt_portfolio_items / mt_accomplishments candidates with body_original = source excerpt; status inbox until Accept
  • Never auto-promote into resume_struct

Out of scope: silent scrape into resume; auto-promote enrichment.

4.2 Plugin system (hooks, not a browser extension)

Today: swappable ats_boards; npm skill init.

Ship:

  • Documented extension points: board source adapters, mt_reports kind handlers, board-pack schema plugins
  • Small registry in SPA/edge (function map + manifest JSON) — bump pack schema_version as needed
  • Example plugin: extra board pack or custom report renderer

Out of scope: Chrome extension, arbitrary code upload in hosted demo.

4.3 Multi-agent workflows (mode chains)

Today: discrete skill modes + grounded Advise follow-ups (materials-ranked; no freeform chat).

Ship:

  • Declared mode chains (e.g. evaluate → rank → interview) over one board pack
  • Each step writes mt_reports; human confirm gate between steps
  • CLI / skill surface: careerops run-chain <name> reading/writing pack only

Out of scope: autonomous apply, unsupervised parallel agents inventing bullets.

Phase 4 done when: confirm GitHub/LinkedIn suggestions into memory/portfolio; extend boards/reports; run gated mode pipelines.

Explicitly deferred forever (unless doctrine changes)

  • Auto-apply
  • Auto-send email
  • Silent LinkedIn (or similar) scrape into resume
  • Invented salary bands
  • Opaque / ungated multi-agent that writes experience without accept