msd-core/gsd-core/workflows/discuss-phase-assumptions.md

Extract implementation decisions that downstream agents need — using codebase-first analysis and assumption surfacing instead of interview-style questioning.

You are a thinking partner, not an interviewer. Analyze the codebase deeply, surface what you believe based on evidence, and ask the user only to correct what's wrong.

<available_agent_types> Valid GSD subagent types (use exact names — do not fall back to 'general-purpose'):

• gsd-assumptions-analyzer — Analyzes codebase to surface implementation assumptions </available_agent_types>

<downstream_awareness> CONTEXT.md feeds into:

1. gsd-phase-researcher — Reads CONTEXT.md to know WHAT to research
2. gsd-planner — Reads CONTEXT.md to know WHAT decisions are locked

Your job: Capture decisions clearly enough that downstream agents can act on them without asking the user again. Output is identical to discuss mode — same CONTEXT.md format. </downstream_awareness>

**Assumptions mode philosophy:**

The user is a visionary, not a codebase archaeologist. They need enough context to evaluate whether your assumptions match their intent — not to answer questions you could figure out by reading the code.

• Read the codebase FIRST, form opinions SECOND, ask ONLY about what's genuinely unclear
• Every assumption must cite evidence (file paths, patterns found)
• Every assumption must state consequences if wrong
• Minimize user interactions: ~2-4 corrections vs ~15-20 questions

<scope_guardrail> CRITICAL: No scope creep.

The phase boundary comes from ROADMAP.md and is FIXED. Discussion clarifies HOW to implement what's scoped, never WHETHER to add new capabilities.

When user suggests scope creep: "[Feature X] would be a new capability — that's its own phase. Want me to note it for the roadmap backlog? For now, let's focus on [phase domain]."

Capture the idea in "Deferred Ideas". Don't lose it, don't act on it. </scope_guardrail>

<answer_validation> IMPORTANT: Answer validation — After every AskUserQuestion call, if the response is empty/whitespace-only:

• "Other" with empty text (the user wants to type freeform): output "What would you like to discuss?", STOP generating, wait for the user's next message, then reflect it back and continue. Do NOT retry AskUserQuestion or call any tools.
• Any other empty response: retry once with the same parameters; if still empty, present options as a plain-text numbered list. Never proceed with empty input.

Text mode (--text or workflow.text_mode: true): follow workflows/discuss-phase/modes/text.md — do not use AskUserQuestion at all. </answer_validation>

Phase number from argument (required).

_GSD_SHIM_NAME="gsd-tools.cjs"; _GSD_RUNTIME_ROOT="${RUNTIME_DIR:-$(git rev-parse --show-toplevel 2>/dev/null || pwd)}"; GSD_TOOLS="${_GSD_RUNTIME_ROOT}/gsd-core/bin/${_GSD_SHIM_NAME}"; if [ -f "$GSD_TOOLS" ]; then gsd_run() { node "$GSD_TOOLS" "$@"; }; elif [ -f "${_GSD_RUNTIME_ROOT}/.claude/gsd-core/bin/${_GSD_SHIM_NAME}" ]; then GSD_TOOLS="${_GSD_RUNTIME_ROOT}/.claude/gsd-core/bin/${_GSD_SHIM_NAME}"; gsd_run() { node "$GSD_TOOLS" "$@"; }; elif [ -f "${_GSD_RUNTIME_ROOT}/.codex/gsd-core/bin/${_GSD_SHIM_NAME}" ]; then GSD_TOOLS="${_GSD_RUNTIME_ROOT}/.codex/gsd-core/bin/${_GSD_SHIM_NAME}"; gsd_run() { node "$GSD_TOOLS" "$@"; }; elif command -v gsd-tools >/dev/null 2>&1; then GSD_TOOLS="$(command -v gsd-tools)"; gsd_run() { "$GSD_TOOLS" "$@"; }; elif [ -f "${CLAUDE_CONFIG_DIR:-$HOME/.claude}/gsd-core/bin/${_GSD_SHIM_NAME}" ]; then GSD_TOOLS="${CLAUDE_CONFIG_DIR:-$HOME/.claude}/gsd-core/bin/${_GSD_SHIM_NAME}"; gsd_run() { node "$GSD_TOOLS" "$@"; }; elif [ -f "${HERMES_HOME:-$HOME/.hermes}/gsd-core/bin/${_GSD_SHIM_NAME}" ]; then GSD_TOOLS="${HERMES_HOME:-$HOME/.hermes}/gsd-core/bin/${_GSD_SHIM_NAME}"; gsd_run() { node "$GSD_TOOLS" "$@"; }; elif [ -f "${CURSOR_CONFIG_DIR:-$HOME/.cursor}/gsd-core/bin/${_GSD_SHIM_NAME}" ]; then GSD_TOOLS="${CURSOR_CONFIG_DIR:-$HOME/.cursor}/gsd-core/bin/${_GSD_SHIM_NAME}"; gsd_run() { node "$GSD_TOOLS" "$@"; }; elif [ -f "${CODEX_HOME:-$HOME/.codex}/gsd-core/bin/${_GSD_SHIM_NAME}" ]; then GSD_TOOLS="${CODEX_HOME:-$HOME/.codex}/gsd-core/bin/${_GSD_SHIM_NAME}"; gsd_run() { node "$GSD_TOOLS" "$@"; }; elif [ -f "${GEMINI_CONFIG_DIR:-$HOME/.gemini}/gsd-core/bin/${_GSD_SHIM_NAME}" ]; then GSD_TOOLS="${GEMINI_CONFIG_DIR:-$HOME/.gemini}/gsd-core/bin/${_GSD_SHIM_NAME}"; gsd_run() { node "$GSD_TOOLS" "$@"; }; elif [ -f "${COPILOT_CONFIG_DIR:-$HOME/.copilot}/gsd-core/bin/${_GSD_SHIM_NAME}" ]; then GSD_TOOLS="${COPILOT_CONFIG_DIR:-$HOME/.copilot}/gsd-core/bin/${_GSD_SHIM_NAME}"; gsd_run() { node "$GSD_TOOLS" "$@"; }; elif [ -f "${WINDSURF_CONFIG_DIR:-$HOME/.codeium/windsurf}/gsd-core/bin/${_GSD_SHIM_NAME}" ]; then GSD_TOOLS="${WINDSURF_CONFIG_DIR:-$HOME/.codeium/windsurf}/gsd-core/bin/${_GSD_SHIM_NAME}"; gsd_run() { node "$GSD_TOOLS" "$@"; }; elif [ -f "${AUGMENT_CONFIG_DIR:-$HOME/.augment}/gsd-core/bin/${_GSD_SHIM_NAME}" ]; then GSD_TOOLS="${AUGMENT_CONFIG_DIR:-$HOME/.augment}/gsd-core/bin/${_GSD_SHIM_NAME}"; gsd_run() { node "$GSD_TOOLS" "$@"; }; elif [ -f "${TRAE_CONFIG_DIR:-$HOME/.trae}/gsd-core/bin/${_GSD_SHIM_NAME}" ]; then GSD_TOOLS="${TRAE_CONFIG_DIR:-$HOME/.trae}/gsd-core/bin/${_GSD_SHIM_NAME}"; gsd_run() { node "$GSD_TOOLS" "$@"; }; elif [ -f "${QWEN_CONFIG_DIR:-$HOME/.qwen}/gsd-core/bin/${_GSD_SHIM_NAME}" ]; then GSD_TOOLS="${QWEN_CONFIG_DIR:-$HOME/.qwen}/gsd-core/bin/${_GSD_SHIM_NAME}"; gsd_run() { node "$GSD_TOOLS" "$@"; }; elif [ -f "${CODEBUDDY_CONFIG_DIR:-$HOME/.codebuddy}/gsd-core/bin/${_GSD_SHIM_NAME}" ]; then GSD_TOOLS="${CODEBUDDY_CONFIG_DIR:-$HOME/.codebuddy}/gsd-core/bin/${_GSD_SHIM_NAME}"; gsd_run() { node "$GSD_TOOLS" "$@"; }; elif [ -f "${CLINE_CONFIG_DIR:-$HOME/.cline}/gsd-core/bin/${_GSD_SHIM_NAME}" ]; then GSD_TOOLS="${CLINE_CONFIG_DIR:-$HOME/.cline}/gsd-core/bin/${_GSD_SHIM_NAME}"; gsd_run() { node "$GSD_TOOLS" "$@"; }; elif [ -f "${GROK_AGENTS_HOME:-$HOME/.agents}/gsd-core/bin/${_GSD_SHIM_NAME}" ]; then GSD_TOOLS="${GROK_AGENTS_HOME:-$HOME/.agents}/gsd-core/bin/${_GSD_SHIM_NAME}"; gsd_run() { node "$GSD_TOOLS" "$@"; }; elif [ -f "${ANTIGRAVITY_CONFIG_DIR:-$HOME/.gemini/antigravity}/gsd-core/bin/${_GSD_SHIM_NAME}" ]; then GSD_TOOLS="${ANTIGRAVITY_CONFIG_DIR:-$HOME/.gemini/antigravity}/gsd-core/bin/${_GSD_SHIM_NAME}"; gsd_run() { node "$GSD_TOOLS" "$@"; }; elif [ -f "${OPENCODE_CONFIG_DIR:-${XDG_CONFIG_HOME:-$HOME/.config}/opencode}/gsd-core/bin/${_GSD_SHIM_NAME}" ]; then GSD_TOOLS="${OPENCODE_CONFIG_DIR:-${XDG_CONFIG_HOME:-$HOME/.config}/opencode}/gsd-core/bin/${_GSD_SHIM_NAME}"; gsd_run() { node "$GSD_TOOLS" "$@"; }; elif [ -f "${KILO_CONFIG_DIR:-${XDG_CONFIG_HOME:-$HOME/.config}/kilo}/gsd-core/bin/${_GSD_SHIM_NAME}" ]; then GSD_TOOLS="${KILO_CONFIG_DIR:-${XDG_CONFIG_HOME:-$HOME/.config}/kilo}/gsd-core/bin/${_GSD_SHIM_NAME}"; gsd_run() { node "$GSD_TOOLS" "$@"; }; else echo "ERROR: gsd-tools.cjs not found at $GSD_TOOLS and gsd-tools is not on PATH. Run: npx -y @opengsd/gsd-core@latest --claude --local" >&2; exit 1; fi; if [ -n "${CLAUDE_ENV_FILE:-}" ] && [ -n "${GSD_TOOLS:-}" ]; then printf "export PATH='%s':\"\$PATH\"\n" "${GSD_TOOLS%/*}" >> "$CLAUDE_ENV_FILE" 2>/dev/null || true; fi
RESPONSE_LANGUAGE=$(gsd_run query config-get response_language --default "" 2>/dev/null || echo "")
AUTO_PARAM=""; if [[ "$ARGUMENTS" =~ (^|[[:space:]])--auto([[:space:]]|$) ]]; then AUTO_PARAM="--auto"; fi
INIT=$(gsd_run query init.discuss-phase-assumptions "${PHASE}" $AUTO_PARAM)
if [[ "$INIT" == @file:* ]]; then INIT=$(cat "${INIT#@file:}"); fi
AGENT_SKILLS_ANALYZER=$(gsd_run query agent-skills gsd-assumptions-analyzer)
# #2072: resolve the routed model so model_overrides / models.discuss are honored
# (the resolver maps gsd-assumptions-analyzer → phaseType "discuss"); thread it below.
ANALYZER_MODEL=$(gsd_run query resolve-model gsd-assumptions-analyzer --raw)

If response_language is set: All user-facing questions, prompts, and explanations in this workflow MUST be presented in {response_language}. Technical terms, code, file paths, and subagent prompts stay in English — only user-facing output is translated.

Parse JSON for: commit_docs, phase_found, phase_dir, phase_number, phase_name, phase_slug, padded_phase, has_research, has_context, has_plans, has_verification, plan_count, roadmap_exists, planning_exists.

If phase_found is false:

Phase [X] not found in roadmap.

Use /gsd:progress to see available phases.

Exit workflow.

If phase_found is true: Continue to check_existing.

Auto mode — If --auto is present in ARGUMENTS:

• In check_existing: auto-select "Update it" (if context exists) or continue without prompting
• In present_assumptions: skip confirmation gate, proceed directly to write CONTEXT.md
• In correct_assumptions: auto-select recommended option for each correction
• Log each auto-selected choice inline
• After completion, auto-advance to plan-phase

Check if CONTEXT.md already exists using `has_context` from init.

ls ${phase_dir}/*-CONTEXT.md 2>/dev/null || true

If exists:

If --auto: Auto-select "Update it". Log: [auto] Context exists — updating with assumption-based analysis.

Otherwise: Use AskUserQuestion:

• header: "Context"
• question: "Phase [X] already has context. What do you want to do?"
• options:
  • "Update it" — Re-analyze codebase and refresh assumptions
  • "View it" — Show me what's there
  • "Skip" — Use existing context as-is

If "Update": Load existing, continue to load_prior_context If "View": Display CONTEXT.md, then offer update/skip If "Skip": Exit workflow

If doesn't exist:

Check has_plans and plan_count from init. If has_plans is true:

If --auto: Auto-select "Continue and replan after". Log: [auto] Plans exist — continuing with assumption analysis, will replan after.

Otherwise: Use AskUserQuestion:

• header: "Plans exist"
• question: "Phase [X] already has {plan_count} plan(s) created without user context. Your decisions here won't affect existing plans unless you replan."
• options:
  • "Continue and replan after"
  • "View existing plans"
  • "Cancel"

If "Continue and replan after": Continue to load_prior_context. If "View existing plans": Display plan files, then offer "Continue" / "Cancel". If "Cancel": Exit workflow.

If has_plans is false: Continue to load_prior_context.

Read project-level and prior phase context to avoid re-asking decided questions.

Step 1: Read project-level files

cat .planning/PROJECT.md 2>/dev/null || true
cat .planning/REQUIREMENTS.md 2>/dev/null || true
cat .planning/STATE.md 2>/dev/null || true

Extract from these:

• PROJECT.md — Vision, principles, non-negotiables, user preferences
• REQUIREMENTS.md — Acceptance criteria, constraints
• STATE.md — Current progress, any flags

Step 2: Read all prior CONTEXT.md files

(find .planning/phases -name "*-CONTEXT.md" 2>/dev/null || true) | sort

For each CONTEXT.md where phase number < current phase:

• Read the <decisions> section — these are locked preferences
• Read <specifics> — particular references or "I want it like X" moments
• Note patterns (e.g., "user consistently prefers minimal UI")

Step 3: Build internal <prior_decisions> context

Structure the extracted information for use in assumption generation.

If no prior context exists: Continue without — expected for early phases.

Check if any pending todos are relevant to this phase's scope.

TODO_MATCHES=$(gsd_run query todo.match-phase "${PHASE_NUMBER}")

Parse JSON for: todo_count, matches[].

If todo_count is 0: Skip silently.

If matches found: Present matched todos, use AskUserQuestion (multiSelect) to fold relevant ones into scope.

For selected (folded) todos: Store as <folded_todos> for CONTEXT.md <decisions> section. For unselected: Store as <reviewed_todos> for CONTEXT.md <deferred> section.

Auto mode (--auto): Fold all todos with score >= 0.4 automatically. Log the selection.

Read the project-level methodology file if it exists. This must happen before assumption analysis so that active lenses shape how assumptions are generated and evaluated.

cat .planning/METHODOLOGY.md 2>/dev/null || true

If METHODOLOGY.md exists:

• Parse each named lens: its diagnoses, recommendations, and triggering conditions
• Store as internal <active_lenses> for use in deep_codebase_analysis and present_assumptions
• When spawning the gsd-assumptions-analyzer, pass the lens list so it can flag which lenses apply
• When presenting assumptions, append a "Methodology" section showing which lenses were applied and what they flagged (if anything)

If METHODOLOGY.md does not exist: Skip silently. This artifact is optional.

Lightweight scan of existing code to inform assumption generation.

Step 1: Check for existing codebase maps

ls .planning/codebase/*.md 2>/dev/null || true

If codebase maps exist: Read relevant ones (CONVENTIONS.md, STRUCTURE.md, STACK.md). Extract reusable components, patterns, integration points. Skip to Step 3.

Step 2: If no codebase maps, do targeted grep

Extract key terms from phase goal, search for related files.

grep -rl "{term1}\|{term2}" src/ app/ --include="*.ts" --include="*.tsx" 2>/dev/null | head -10

Read the 3-5 most relevant files.

Step 3: Build internal <codebase_context>

Identify reusable assets, established patterns, integration points, and creative options. Store internally for use in deep_codebase_analysis.

Spawn a `gsd-assumptions-analyzer` agent to deeply analyze the codebase for this phase. This keeps raw file contents out of the main context window, protecting token budget.

Resolve calibration tier (if USER-PROFILE.md exists):

PROFILE_PATH="$HOME/.claude/gsd-core/USER-PROFILE.md"

If file exists at PROFILE_PATH:

• Priority 1: Read config.json > preferences.vendor_philosophy (project-level override)
• Priority 2: Read USER-PROFILE.md Vendor Choices/Philosophy rating (global)
• Priority 3: Default to "standard"

Map to calibration tier:

• conservative OR thorough-evaluator → full_maturity (more alternatives, detailed evidence)
• opinionated → minimal_decisive (fewer alternatives, decisive recommendations)
• pragmatic-fast OR any other value → standard

If no USER-PROFILE.md: calibration_tier = "standard"

Spawn Explore subagent (runs in a subagent — no output until it returns, ~1–5 min; expected, not a freeze):

Runtime-aware dispatch (#2508 Phase 4). GSD workflows dispatch specialized subagents by role. Before dispatching on a built-in-only runtime (kimi-code — three built-ins only), resolve the role to a built-in via gsd_run query resolve-dispatch-type --requested <role> --raw. On named-dispatch runtimes (Claude/OpenCode/…) the role is returned unchanged; on kimi-code it maps to coder/explore/plan by role-suffix. The persona rides ${AGENT_SKILLS_<ROLE>} (Phase 3) regardless. See @gsd-core/references/runtime-aware-dispatch.md.

Model omission (#2517). Omit the model parameter entirely when the value it would carry (ANALYZER_MODEL) is "inherit" or empty. An empty value 404s on runtimes without native tier aliases — the default on non-Claude runtimes. Omitting it inherits the orchestrator's model. See @gsd-core/references/model-profile-resolution.md.

Agent(subagent_type="gsd-assumptions-analyzer", model="{ANALYZER_MODEL}", prompt="""
Analyze the codebase for Phase {PHASE}: {phase_name}.

Phase goal: {roadmap_description}
Prior decisions: {prior_decisions_summary}
Codebase scout hints: {codebase_context_summary}
Calibration: {calibration_tier}

Your job:
1. Read ROADMAP.md phase {PHASE} description
2. Read any prior CONTEXT.md files from earlier phases
3. Glob/Grep for files related to: {phase_relevant_terms}
4. Read 5-15 most relevant source files
5. Return structured assumptions

## Output Format

Return EXACTLY this structure:

## Assumptions

### [Area Name] (e.g., "Technical Approach")
- **Assumption:** [Decision statement]
  - **Why this way:** [Evidence from codebase — cite file paths]
  - **If wrong:** [Concrete consequence of this being wrong]
  - **Confidence:** Confident | Likely | Unclear

(3-5 areas, calibrated by tier:
- full_maturity: 3-5 areas, 2-3 alternatives per Likely/Unclear item
- standard: 3-4 areas, 2 alternatives per Likely/Unclear item
- minimal_decisive: 2-3 areas, decisive single recommendation per item)

## Needs External Research
[Topics where codebase alone is insufficient — library version compatibility,
ecosystem best practices, etc. Leave empty if codebase provides enough evidence.]

${AGENT_SKILLS_ANALYZER}
""")

ORCHESTRATOR RULE — CODEX RUNTIME: After calling Agent() above, stop working on this task immediately. Do not read more files, analyze the codebase, or process assumptions while the subagent is active. Wait for the subagent to return its result. This prevents duplicate work, conflicting edits, and wasted context. Only resume when the subagent result is available.

Parse the subagent's response. Extract:

• assumptions[] — each with area, statement, evidence, consequence, confidence
• needs_research[] — topics requiring external research (may be empty)

Initialize canonical refs accumulator:

• Source 1: Copy Canonical refs: from ROADMAP.md for this phase, expand to full paths
• Source 2: Check REQUIREMENTS.md and PROJECT.md for specs/ADRs referenced
• Source 3: Add any docs referenced in codebase scout results

**Skip if:** `needs_research` from deep_codebase_analysis is empty.

If research topics were flagged, spawn a general-purpose research agent (runs in a subagent — no output until it returns, ~1–5 min; expected, not a freeze):

Agent(subagent_type="general-purpose", prompt="""
Research the following topics for Phase {PHASE}: {phase_name}.

Topics needing research:
{needs_research_content}

For each topic, return:
- **Finding:** [What you learned]
- **Source:** [URL or library docs reference]
- **Confidence impact:** [Which assumption this resolves and to what confidence level]

Use Context7 (resolve-library-id then query-docs) for library-specific questions.
Use WebSearch for ecosystem/best-practice questions.
""")

> **ORCHESTRATOR RULE — CODEX RUNTIME**: After calling Agent() above, stop working on this task immediately. Do not independently research any of these topics while the subagent is active. Wait for the subagent to return its result. This prevents duplicate work and wasted context. Only resume when the subagent result is available.

Merge findings back into assumptions:

• Update confidence levels where research resolves ambiguity
• Add source attribution to affected assumptions
• Store research findings for DISCUSSION-LOG.md

If no gaps flagged: Skip entirely. Most phases will skip this step.

Display all assumptions grouped by area with confidence badges.

Format for display:

## Phase {PHASE}: {phase_name} — Assumptions

Based on codebase analysis, here's what I'd go with:

### {Area Name}
{Confidence badge} **{Assumption statement}**
↳ Evidence: {file paths cited}
↳ If wrong: {consequence}

### {Area Name 2}
...

[If external research was done:]
### External Research Applied
- {Topic}: {Finding} (Source: {URL})

If --auto:

• If all assumptions are Confident or Likely: log assumptions, skip to write_context. Log: [auto] All assumptions Confident/Likely — proceeding to context capture.
• If any assumptions are Unclear: log a warning, auto-select recommended alternative for each Unclear item. Log: [auto] {N} Unclear assumptions auto-resolved with recommended defaults. Proceed to write_context.

Otherwise: Use AskUserQuestion:

• header: "Assumptions"
• question: "These all look right?"
• options:
  • "Yes, proceed" — Write CONTEXT.md with these assumptions as decisions
  • "Let me correct some" — Select which assumptions to change

If "Yes, proceed": Skip to write_context. If "Let me correct some": Continue to correct_assumptions.

The assumptions are already displayed above from present_assumptions.

Present a multiSelect where each option's label is the assumption statement and description is the "If wrong" consequence:

Use AskUserQuestion (multiSelect):

• header: "Corrections"
• question: "Which assumptions need correcting?"
• options: [one per assumption, label = assumption statement, description = "If wrong: {consequence}"]

For each selected correction, ask ONE focused question:

Use AskUserQuestion:

• header: "{Area Name}"
• question: "What should we do instead for: {assumption statement}?"
• options: [2-3 concrete alternatives describing user-visible outcomes, recommended option first]

Record each correction:

• Original assumption
• User's chosen alternative
• Reason (if provided via "Other" free text)

After all corrections processed, continue to write_context with updated assumptions.

Auto mode: Should not reach this step (--auto skips from present_assumptions).

Create phase directory if needed. Write CONTEXT.md using the standard 6-section format.

File: ${phase_dir}/${padded_phase}-CONTEXT.md

Map assumptions to CONTEXT.md sections:

• Assumptions → <decisions> (each assumption becomes a locked decision: D-01, D-02, etc.)
• Corrections → override the original assumption in <decisions>
• Areas where all assumptions were Confident → marked as locked decisions
• Areas with corrections → include user's chosen alternative as the decision
• Folded todos → included in <decisions> under "### Folded Todos"

# Phase {PHASE}: {phase_name} - Context

**Gathered:** {date} (assumptions mode)
**Status:** Ready for planning

<domain>
## Phase Boundary

{Domain boundary from ROADMAP.md — clear statement of scope anchor}
</domain>

<decisions>
## Implementation Decisions

### {Area Name 1}
- **D-01:** {Decision — from assumption or correction}
- **D-02:** {Decision}

### {Area Name 2}
- **D-03:** {Decision}

### Claude's Discretion
{Any assumptions where the user confirmed "you decide" or left as-is with Likely confidence}

### Folded Todos
{If any todos were folded into scope}
</decisions>

<canonical_refs>
## Canonical References

**Downstream agents MUST read these before planning or implementing.**

{Accumulated canonical refs from analyze step — full relative paths}

[If no external specs: "No external specs — requirements fully captured in decisions above"]
</canonical_refs>

<code_context>
## Existing Code Insights

### Reusable Assets
{From codebase scout + Explore subagent findings}

### Established Patterns
{Patterns that constrain/enable this phase}

### Integration Points
{Where new code connects to existing system}
</code_context>

<specifics>
## Specific Ideas

{Any particular references from corrections or user input}

[If none: "No specific requirements — open to standard approaches"]
</specifics>

<deferred>
## Deferred Ideas

{Ideas mentioned during corrections that are out of scope}

### Reviewed Todos (not folded)
{Todos reviewed but not folded — with reason}

[If none: "None — analysis stayed within phase scope"]
</deferred>

Write file.

Write audit trail of assumptions and corrections.

File: ${phase_dir}/${padded_phase}-DISCUSSION-LOG.md

# Phase {PHASE}: {phase_name} - Discussion Log (Assumptions Mode)

> **Audit trail only.** Do not use as input to planning, research, or execution agents.
> Decisions captured in CONTEXT.md — this log preserves the analysis.

**Date:** {ISO date}
**Phase:** {padded_phase}-{phase_name}
**Mode:** assumptions
**Areas analyzed:** {comma-separated area names}

## Assumptions Presented

### {Area Name}
| Assumption | Confidence | Evidence |
|------------|-----------|----------|
| {Statement} | {Confident/Likely/Unclear} | {file paths} |

{Repeat for each area}

## Corrections Made

{If corrections were made:}

### {Area Name}
- **Original assumption:** {what Claude assumed}
- **User correction:** {what the user chose instead}
- **Reason:** {user's rationale, if provided}

{If no corrections: "No corrections — all assumptions confirmed."}

## Auto-Resolved

{If --auto and Unclear items existed:}
- {Assumption}: auto-selected {recommended option}

{If not applicable: omit this section}

## External Research

{If research was performed:}
- {Topic}: {Finding} (Source: {URL})

{If no research: omit this section}

Write file.

Commit phase context and discussion log:

gsd_run query commit "docs(${padded_phase}): capture phase context (assumptions mode)" --files "${phase_dir}/${padded_phase}-CONTEXT.md" "${phase_dir}/${padded_phase}-DISCUSSION-LOG.md"

Confirm: "Committed: docs(${padded_phase}): capture phase context (assumptions mode)"

Update STATE.md with session info:

gsd_run query state.record-session \
  --stopped-at "Phase ${PHASE} context gathered (assumptions mode)" \
  --resume-file "${phase_dir}/${padded_phase}-CONTEXT.md"

Commit STATE.md:

gsd_run query commit "docs(state): record phase ${PHASE} context session" --files .planning/STATE.md

Present summary and next steps:

Created: .planning/phases/${PADDED_PHASE}-${SLUG}/${PADDED_PHASE}-CONTEXT.md

## Decisions Captured (Assumptions Mode)

### {Area Name}
- {Key decision} (from assumption / corrected)

{Repeat per area}

[If corrections were made:]
## Corrections Applied
- {Area}: {original} → {corrected}

[If deferred ideas exist:]
## Noted for Later
- {Deferred idea} — future phase

---

## ▶ Next Up — [${PROJECT_CODE}] ${PROJECT_TITLE}

**Phase ${PHASE}: {phase_name}** — {Goal from ROADMAP.md}

`/clear` then:

`/gsd:plan-phase ${PHASE}`

---

**Also available:**
- `/gsd:plan-phase ${PHASE} --skip-research` — plan without research
- `/gsd:ui-phase ${PHASE}` — generate UI design contract (if frontend work)
- Review/edit CONTEXT.md before continuing

---

Check for auto-advance trigger:

1. Parse --auto flag from $ARGUMENTS
2. Sync chain flag:

   if [[ ! "$ARGUMENTS" =~ --auto ]]; then
     gsd_run query config-set workflow._auto_chain_active false || true
   fi
   

3. Read consolidated auto-mode (active = chain flag OR user preference):

   AUTO_MODE=$(gsd_run query check auto-mode --pick active 2>/dev/null || echo "false")
   

If --auto flag present AND AUTO_MODE is not true:

gsd_run query config-set workflow._auto_chain_active true

If section_manifest is null or "auto-advance-dispatch" is in its included list: read and execute gsd-core/workflows/discuss-phase-assumptions/steps/auto-advance-dispatch.md. Otherwise skip — do not read the file.

If neither --auto nor config enabled: End here — confirm_creation already ran; do not route back to it.

<success_criteria>

• Phase validated against roadmap
• Prior context loaded (no re-asking decided questions)
• Codebase deeply analyzed via Explore subagent (5-15 files read)
• Assumptions surfaced with evidence and confidence levels
• User confirmed or corrected assumptions (~2-4 interactions max)
• Scope creep redirected to deferred ideas
• CONTEXT.md captures actual decisions (identical format to discuss mode)
• CONTEXT.md includes canonical_refs with full file paths (MANDATORY)
• CONTEXT.md includes code_context from codebase analysis
• DISCUSSION-LOG.md records assumptions and corrections as audit trail
• STATE.md updated with session info
• User knows next steps </success_criteria>