[FEATURE] Automatic pre-assembled context package output from teamwork next

[JoshLuedeman]

Problem It Solves

Closes gap: #171

Agents start each step with no pre-assembled context. They must manually locate and read handoff artifacts, memory entries, and the state file before they can begin productive work. This is error-prone and time-consuming, and agents frequently begin work with incomplete context.

Proposed Behavior

teamwork next gains a --context flag (or always includes context for steps beyond step 1) that produces a ready-to-use context document alongside the step instruction:

$ teamwork next <workflow-id> --context

## Next Step: Step 4 — Coder
Goal: Implement the authentication endpoint per the architect's design

## Context Package
### Prior Workflow Summary
- Step 1 (Planner): Decomposed into 3 tasks. Auth endpoint, token refresh, logout. See handoff: .teamwork/handoffs/wf-042-step1-planner.md
- Step 2 (Architect): Chose JWT with RS256. Sub-claims for roles. 15-min expiry. ADR-012 filed. No DB changes required.
- Step 3 (Security Auditor): No blocking findings. Flagged: ensure token not logged at DEBUG level (LOW).

### Relevant Memory (3 of 47 entries)
- [pattern] JWT implementation: always validate aud and iss claims (source: wf-031)
- [decision] Auth errors return 401 with WWW-Authenticate header, never 403 for missing token (source: ADR-008)
- [antipattern] Do not store refresh tokens in localStorage — use httpOnly cookies (source: wf-019)

### Your Escalation Paths
- Design conflict → architect
- Test infrastructure question → tester
- Dependency vulnerability → security-auditor

The context document is sized for a single session opening. It reads prior handoffs via the handoff package, queries memory via tag matching, and pulls escalation paths from the agent file.

Dependencies

• Requires: [GAP] Agents receive no pre-assembled context at step start — manual file hunting required #171 (gap) as motivation
• Enhanced by: [MEMORY] Add semantic search over project memory and handoff artifacts #41 (semantic memory search) — better memory relevance ranking
• Enhanced by: [MEMORY] Add automatic context distillation for agent sessions #62 (context distillation) — compressed handoff summaries
• Enhanced by: [GAP] Agent project knowledge is identical across all roles — no role-specific schemas #172 (role-specific knowledge schemas) — adds role-specific project facts

Complexity Estimate

Medium — 4 files: context assembly logic, handoff summarizer, memory relevance ranker, teamwork next output extension, tests