feat(product): guide next actions and compose local career skills

[avabbbb]

Goal

This PR upgrades two product boundaries without adding another Agent loop:

1. Guided interaction — Today should tell a non-technical user what to do next instead of making them discover features/modes.
2. Local Skill ecosystem compatibility — OfferU should compose with resume / recruiting / interview / career Skills already installed in the user's local Agent host.

Related product direction: #18.

What changes

1. Today: Next Best Actions

Today now derives at most three guided actions from existing canonical projections:

• pending ApplicationProgressCandidates;
• active / upcoming interviews;
• pending external job-search signals;
• Pipeline next_action.

Each card includes:

• what to do;
• why it matters now;
• one direct CTA.

No new task store or backend truth source is introduced.

The intent is:

Career State
→ rank actionable work
→ show <= 3 actions
→ user acts
→ state changes
→ recalculate

instead of:

open dashboard
→ figure out which feature/mode to use

2. OfferU Skill becomes explicitly composable

The generated OfferU external-agent projections now state that other installed career Skills may be used for:

• resume methodology / critique;
• recruiting workflows;
• interview coaching;
• portfolio / storytelling;
• negotiation / outreach;
• other specialized career workflows.

OfferU remains authoritative for:

• Profile / Evidence;
• Job / Application / Interview state;
• Operation Registry;
• Proposal / HITL;
• audit and side effects.

Third-party Skill output is treated as draft / analysis / Candidate input. It cannot:

• write OfferU DB directly;
• bypass Registry / confirmation;
• silently promote inferred facts;
• auto-submit applications;
• send email/recruiter messages;
• weaken OfferU's secret or permission rules.

This composition rule is projected to the existing generic Skill, Claude, Codex operator and Copilot surfaces.

3. Design contract

Adds:

docs/design/2026-09-22-guided-career-coach-skill-ecosystem.md

The document defines:

• setup burden + decision burden as separate product problems;
• Guided Mode / Next Best Action principles;
• progressive profiling;
• OfferU's role in the open Agent Skills ecosystem;
• the third-party Skill composition/trust contract;
• follow-up slices (guided onboarding, metadata-only Skill inventory, proactive follow-up, outcome learning).

Why this approach

Agent Skills is an open, composable format based on progressive disclosure, so OfferU does not need to reimplement every resume/interview/recruiting methodology:

https://agentskills.io/

https://www.anthropic.com/engineering/equipping-agents-for-the-real-world-with-agent-skills

career-ops is a useful example of a broad career workflow Skill:

https://github.com/career-ops-hq/career-ops

Its onboarding/readiness, auto-pipeline and follow-up concepts are useful, but the beginner UX should not require users to know commands like scan, followup, reply-watch, etc. OfferU should surface those conditions as proactive actions.

Deliberate non-goals

This PR does not:

• create a Skill marketplace;
• scan arbitrary local Skill folders;
• execute third-party Skill scripts;
• add a parallel Agent runtime;
• add another Career Truth store;
• change Operation Registry permissions;
• auto-submit applications or send external messages.

A future metadata-only Skill Inventory can be a separate, explicit-permission PR.

Tests

Added/updated contracts for:

• generated OfferU Skill projections containing the composable career-Skill contract;
• Today rendering proactive next actions from existing Career State.

CI should validate frontend typecheck/tests/build and backend projection drift/contracts.