feat: add workflow.discuss_mode assumptions config (#637)

Add codebase-first assumption-driven alternative to the interview-style
discuss-phase. New `workflow.discuss_mode: "assumptions"` config routes
to a separate workflow that spawns a gsd-assumptions-analyzer agent to
read 5-15 codebase files, surface assumptions with evidence, and ask
only for corrections (~2-4 interactions vs ~15-20).

- New gsd-assumptions-analyzer agent for deep codebase analysis
- New discuss-phase-assumptions.md workflow (15 steps)
- Command-level routing via dual @reference + process gate
- Identical CONTEXT.md output — downstream agents unaffected
- Existing discuss-phase.md workflow untouched (zero diff)
- Mode-aware plan-phase gate and progress display
- User documentation and integration tests
- Update agent count and list in copilot-install tests (17 → 18)

Closes #637

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

docs/workflow-discuss-mode.md

@@ -0,0 +1,68 @@
+# Discuss Mode: Assumptions vs Interview
+
+GSD's discuss-phase has two modes for gathering implementation context before planning.
+
+## Modes
+
+### `discuss` (default)
+
+The original interview-style flow. Claude identifies gray areas in the phase, presents them
+for selection, then asks ~4 questions per area. Good for:
+
+- Early phases where the codebase is new
+- Phases where the user has strong opinions they want to express proactively
+- Users who prefer guided, conversational context gathering
+
+### `assumptions`
+
+A codebase-first flow. Claude deeply analyzes the codebase via a subagent (reading 5-15
+relevant files), forms assumptions with evidence, and presents them for confirmation or
+correction. Good for:
+
+- Established codebases with clear patterns
+- Users who find the interview questions obvious
+- Faster context gathering (~2-4 interactions vs ~15-20)
+
+## Configuration
+
+```bash
+# Enable assumptions mode
+gsd-tools config-set workflow.discuss_mode assumptions
+
+# Switch back to interview mode
+gsd-tools config-set workflow.discuss_mode discuss
+```
+
+The setting is per-project (stored in `.planning/config.json`).
+
+## How Assumptions Mode Works
+
+1. **Init** — Same as discuss mode (load prior context, scout codebase, check todos)
+2. **Deep analysis** — Explore subagent reads 5-15 codebase files related to the phase
+3. **Surface assumptions** — Each assumption includes:
+   - What Claude would do and why (citing file paths)
+   - What goes wrong if the assumption is incorrect
+   - Confidence level (Confident / Likely / Unclear)
+4. **Confirm or correct** — User reviews assumptions, selects any that need changing
+5. **Write CONTEXT.md** — Identical output format to discuss mode
+
+## Flag Compatibility
+
+| Flag | `discuss` mode | `assumptions` mode |
+|------|----------------|-------------------|
+| `--auto` | Auto-selects recommended answers | Skips confirm gate, auto-resolves Unclear items |
+| `--batch` | Groups questions in batches | N/A (corrections already batched) |
+| `--text` | Plain-text questions (remote sessions) | Plain-text questions (remote sessions) |
+| `--analyze` | Shows trade-off tables per question | N/A (assumptions include evidence) |
+
+## Output
+
+Both modes produce identical CONTEXT.md with the same 6 sections:
+- `<domain>` — Phase boundary
+- `<decisions>` — Locked implementation decisions
+- `<canonical_refs>` — Specs/docs downstream agents must read
+- `<code_context>` — Reusable assets, patterns, integration points
+- `<specifics>` — User references and preferences
+- `<deferred>` — Ideas noted for future phases
+
+Downstream agents (researcher, planner, checker) consume this identically regardless of mode.