fix: extract smart_discuss step to reference to reduce autonomous.md token count (#2227)

autonomous.md reported 11,748 tokens (over the Claude Code Read tool's 10K
limit), causing it to be read in 150-line chunks and generating a warning
on every /gsd-autonomous invocation.

Extract the 280-line smart_discuss step into a new reference file
(get-shit-done/references/autonomous-smart-discuss.md) and replace the
step body with a lean stub that directs the agent to read the reference.
This follows the established planner decomposition pattern.

autonomous.md: 38,750 → 29,411 chars (~7,350 tokens, well under 10K limit)

Fixes #2196

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Tom Boucher
2026-04-15 14:59:18 -04:00
committed by GitHub
parent 875b257c18
commit 8f1dd94495
3 changed files with 370 additions and 274 deletions

View File

@@ -0,0 +1,277 @@
# Smart Discuss — Autonomous Mode
Smart discuss is the autonomous-optimized variant of `gsd-discuss-phase`. It proposes grey area answers in batch tables — the user accepts or overrides per area — then writes an identical CONTEXT.md to what discuss-phase produces.
**Inputs:** `PHASE_NUM` from execute_phase. Run init to get phase paths:
```bash
PHASE_STATE=$(node "$HOME/.claude/get-shit-done/bin/gsd-tools.cjs" init phase-op ${PHASE_NUM})
```
Parse from JSON: `phase_dir`, `phase_slug`, `padded_phase`, `phase_name`.
---
## Sub-step 1: Load prior context
Read project-level and prior phase context to avoid re-asking decided questions.
**Read project files:**
```bash
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, must-haves vs nice-to-haves
- **STATE.md** — Current progress, decisions logged so far
**Read all prior CONTEXT.md files:**
```bash
(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", "user rejected verbose output")
**Build internal prior_decisions context** (do not write to file):
```
<prior_decisions>
## Project-Level
- [Key principle or constraint from PROJECT.md]
- [Requirement affecting this phase from REQUIREMENTS.md]
## From Prior Phases
### Phase N: [Name]
- [Decision relevant to current phase]
- [Preference that establishes a pattern]
</prior_decisions>
```
If no prior context exists, continue without — expected for early phases.
---
## Sub-step 2: Scout Codebase
Lightweight codebase scan to inform grey area identification and proposals. Keep under ~5% context.
**Check for existing codebase maps:**
```bash
ls .planning/codebase/*.md 2>/dev/null || true
```
**If codebase maps exist:** Read the most relevant ones (CONVENTIONS.md, STRUCTURE.md, STACK.md based on phase type). Extract reusable components, established patterns, integration points. Skip to building context below.
**If no codebase maps, do targeted grep:**
Extract key terms from the phase goal. Search for related files:
```bash
grep -rl "{term1}\|{term2}" src/ app/ --include="*.ts" --include="*.tsx" --include="*.js" --include="*.jsx" 2>/dev/null | head -10 || true
ls src/components/ src/hooks/ src/lib/ src/utils/ 2>/dev/null || true
```
Read the 3-5 most relevant files to understand existing patterns.
**Build internal codebase_context** (do not write to file):
- **Reusable assets** — existing components, hooks, utilities usable in this phase
- **Established patterns** — how the codebase does state management, styling, data fetching
- **Integration points** — where new code connects (routes, nav, providers)
---
## Sub-step 3: Analyze Phase and Generate Proposals
**Get phase details:**
```bash
DETAIL=$(node "$HOME/.claude/get-shit-done/bin/gsd-tools.cjs" roadmap get-phase ${PHASE_NUM})
```
Extract `goal`, `requirements`, `success_criteria` from the JSON response.
**Infrastructure detection — check FIRST before generating grey areas:**
A phase is pure infrastructure when ALL of these are true:
1. Goal keywords match: "scaffolding", "plumbing", "setup", "configuration", "migration", "refactor", "rename", "restructure", "upgrade", "infrastructure"
2. AND success criteria are all technical: "file exists", "test passes", "config valid", "command runs"
3. AND no user-facing behavior is described (no "users can", "displays", "shows", "presents")
**If infrastructure-only:** Skip Sub-step 4. Jump directly to Sub-step 5 with minimal CONTEXT.md. Display:
```
Phase ${PHASE_NUM}: Infrastructure phase — skipping discuss, writing minimal context.
```
Use these defaults for the CONTEXT.md:
- `<domain>`: Phase boundary from ROADMAP goal
- `<decisions>`: Single "### Claude's Discretion" subsection — "All implementation choices are at Claude's discretion — pure infrastructure phase"
- `<code_context>`: Whatever the codebase scout found
- `<specifics>`: "No specific requirements — infrastructure phase"
- `<deferred>`: "None"
**If NOT infrastructure — generate grey area proposals:**
Determine domain type from the phase goal:
- Something users **SEE** → visual: layout, interactions, states, density
- Something users **CALL** → interface: contracts, responses, errors, auth
- Something users **RUN** → execution: invocation, output, behavior modes, flags
- Something users **READ** → content: structure, tone, depth, flow
- Something being **ORGANIZED** → organization: criteria, grouping, exceptions, naming
Check prior_decisions — skip grey areas already decided in prior phases.
Generate **3-4 grey areas** with **~4 questions each**. For each question:
- **Pre-select a recommended answer** based on: prior decisions (consistency), codebase patterns (reuse), domain conventions (standard approaches), ROADMAP success criteria
- Generate **1-2 alternatives** per question
- **Annotate** with prior decision context ("You decided X in Phase N") and code context ("Component Y exists with Z variants") where relevant
---
## Sub-step 4: Present Proposals Per Area
Present grey areas **one at a time**. For each area (M of N):
Display a table:
```
### Grey Area {M}/{N}: {Area Name}
| # | Question | ✅ Recommended | Alternative(s) |
|---|----------|---------------|-----------------|
| 1 | {question} | {answer} — {rationale} | {alt1}; {alt2} |
| 2 | {question} | {answer} — {rationale} | {alt1} |
| 3 | {question} | {answer} — {rationale} | {alt1}; {alt2} |
| 4 | {question} | {answer} — {rationale} | {alt1} |
```
Then prompt the user via **AskUserQuestion**:
- **header:** "Area {M}/{N}"
- **question:** "Accept these answers for {Area Name}?"
- **options:** Build dynamically — always "Accept all" first, then "Change Q1" through "Change QN" for each question (up to 4), then "Discuss deeper" last. Cap at 6 explicit options max (AskUserQuestion adds "Other" automatically).
**On "Accept all":** Record all recommended answers for this area. Move to next area.
**On "Change QN":** Use AskUserQuestion with the alternatives for that specific question:
- **header:** "{Area Name}"
- **question:** "Q{N}: {question text}"
- **options:** List the 1-2 alternatives plus "You decide" (maps to Claude's Discretion)
Record the user's choice. Re-display the updated table with the change reflected. Re-present the full acceptance prompt so the user can make additional changes or accept.
**On "Discuss deeper":** Switch to interactive mode for this area only — ask questions one at a time using AskUserQuestion with 2-3 concrete options per question plus "You decide". After 4 questions, prompt:
- **header:** "{Area Name}"
- **question:** "More questions about {area name}, or move to next?"
- **options:** "More questions" / "Next area"
If "More questions", ask 4 more. If "Next area", display final summary table of captured answers for this area and move on.
**On "Other" (free text):** Interpret as either a specific change request or general feedback. Incorporate into the area's decisions, re-display updated table, re-present acceptance prompt.
**Scope creep handling:** If user mentions something outside the phase domain:
```
"{Feature} sounds like a new capability — that belongs in its own phase.
I'll note it as a deferred idea.
Back to {current area}: {return to current question}"
```
Track deferred ideas internally for inclusion in CONTEXT.md.
---
## Sub-step 5: Write CONTEXT.md
After all areas are resolved (or infrastructure skip), write the CONTEXT.md file.
**File path:** `${phase_dir}/${padded_phase}-CONTEXT.md`
Use **exactly** this structure (identical to discuss-phase output):
```markdown
# Phase {PHASE_NUM}: {Phase Name} - Context
**Gathered:** {date}
**Status:** Ready for planning
<domain>
## Phase Boundary
{Domain boundary statement from analysis — what this phase delivers}
</domain>
<decisions>
## Implementation Decisions
### {Area 1 Name}
- {Accepted/chosen answer for Q1}
- {Accepted/chosen answer for Q2}
- {Accepted/chosen answer for Q3}
- {Accepted/chosen answer for Q4}
### {Area 2 Name}
- {Accepted/chosen answer for Q1}
- {Accepted/chosen answer for Q2}
...
### Claude's Discretion
{Any "You decide" answers collected — note Claude has flexibility here}
</decisions>
<code_context>
## Existing Code Insights
### Reusable Assets
- {From codebase scout — components, hooks, utilities}
### Established Patterns
- {From codebase scout — state management, styling, data fetching}
### Integration Points
- {From codebase scout — where new code connects}
</code_context>
<specifics>
## Specific Ideas
{Any specific references or "I want it like X" from discussion}
{If none: "No specific requirements — open to standard approaches"}
</specifics>
<deferred>
## Deferred Ideas
{Ideas captured but out of scope for this phase}
{If none: "None — discussion stayed within phase scope"}
</deferred>
```
Write the file.
**Commit:**
```bash
node "$HOME/.claude/get-shit-done/bin/gsd-tools.cjs" commit "docs(${PADDED_PHASE}): smart discuss context" --files "${phase_dir}/${padded_phase}-CONTEXT.md"
```
Display confirmation:
```
Created: {path}
Decisions captured: {count} across {area_count} areas
```

View File

@@ -511,283 +511,13 @@ Display the review result summary (score from UI-REVIEW.md if produced). Continu
## Smart Discuss
Run smart discuss for the current phase. Proposes grey area answers in batch tables — the user accepts or overrides per area. Produces identical CONTEXT.md output to regular discuss-phase.
> Full instructions are in `get-shit-done/references/autonomous-smart-discuss.md`. Read that file now and follow it exactly.
> **Note:** Smart discuss is an autonomous-optimized variant of the `gsd-discuss-phase` skill. It produces identical CONTEXT.md output but uses batch table proposals instead of sequential questioning. The original `gsd-discuss-phase` skill remains unchanged (per CTRL-03). Future milestones may extract this to a separate skill file.
Smart discuss is an autonomous-optimized variant of `gsd-discuss-phase`. It proposes grey area answers in batch tables — the user accepts or overrides per area — and writes an identical CONTEXT.md to what discuss-phase produces.
**Inputs:** `PHASE_NUM` from execute_phase. Run init to get phase paths:
**Inputs:** `PHASE_NUM` from execute_phase.
```bash
PHASE_STATE=$(node "$HOME/.claude/get-shit-done/bin/gsd-tools.cjs" init phase-op ${PHASE_NUM})
```
Parse from JSON: `phase_dir`, `phase_slug`, `padded_phase`, `phase_name`.
---
### Sub-step 1: Load prior context
Read project-level and prior phase context to avoid re-asking decided questions.
**Read project files:**
```bash
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, must-haves vs nice-to-haves
- **STATE.md** — Current progress, decisions logged so far
**Read all prior CONTEXT.md files:**
```bash
(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", "user rejected verbose output")
**Build internal prior_decisions context** (do not write to file):
```
<prior_decisions>
## Project-Level
- [Key principle or constraint from PROJECT.md]
- [Requirement affecting this phase from REQUIREMENTS.md]
## From Prior Phases
### Phase N: [Name]
- [Decision relevant to current phase]
- [Preference that establishes a pattern]
</prior_decisions>
```
If no prior context exists, continue without — expected for early phases.
---
### Sub-step 2: Scout Codebase
Lightweight codebase scan to inform grey area identification and proposals. Keep under ~5% context.
**Check for existing codebase maps:**
```bash
ls .planning/codebase/*.md 2>/dev/null || true
```
**If codebase maps exist:** Read the most relevant ones (CONVENTIONS.md, STRUCTURE.md, STACK.md based on phase type). Extract reusable components, established patterns, integration points. Skip to building context below.
**If no codebase maps, do targeted grep:**
Extract key terms from the phase goal. Search for related files:
```bash
grep -rl "{term1}\|{term2}" src/ app/ --include="*.ts" --include="*.tsx" --include="*.js" --include="*.jsx" 2>/dev/null | head -10 || true
ls src/components/ src/hooks/ src/lib/ src/utils/ 2>/dev/null || true
```
Read the 3-5 most relevant files to understand existing patterns.
**Build internal codebase_context** (do not write to file):
- **Reusable assets** — existing components, hooks, utilities usable in this phase
- **Established patterns** — how the codebase does state management, styling, data fetching
- **Integration points** — where new code connects (routes, nav, providers)
---
### Sub-step 3: Analyze Phase and Generate Proposals
**Get phase details:**
```bash
DETAIL=$(node "$HOME/.claude/get-shit-done/bin/gsd-tools.cjs" roadmap get-phase ${PHASE_NUM})
```
Extract `goal`, `requirements`, `success_criteria` from the JSON response.
**Infrastructure detection — check FIRST before generating grey areas:**
A phase is pure infrastructure when ALL of these are true:
1. Goal keywords match: "scaffolding", "plumbing", "setup", "configuration", "migration", "refactor", "rename", "restructure", "upgrade", "infrastructure"
2. AND success criteria are all technical: "file exists", "test passes", "config valid", "command runs"
3. AND no user-facing behavior is described (no "users can", "displays", "shows", "presents")
**If infrastructure-only:** Skip Sub-step 4. Jump directly to Sub-step 5 with minimal CONTEXT.md. Display:
```
Phase ${PHASE_NUM}: Infrastructure phase — skipping discuss, writing minimal context.
```
Use these defaults for the CONTEXT.md:
- `<domain>`: Phase boundary from ROADMAP goal
- `<decisions>`: Single "### Claude's Discretion" subsection — "All implementation choices are at Claude's discretion — pure infrastructure phase"
- `<code_context>`: Whatever the codebase scout found
- `<specifics>`: "No specific requirements — infrastructure phase"
- `<deferred>`: "None"
**If NOT infrastructure — generate grey area proposals:**
Determine domain type from the phase goal:
- Something users **SEE** → visual: layout, interactions, states, density
- Something users **CALL** → interface: contracts, responses, errors, auth
- Something users **RUN** → execution: invocation, output, behavior modes, flags
- Something users **READ** → content: structure, tone, depth, flow
- Something being **ORGANIZED** → organization: criteria, grouping, exceptions, naming
Check prior_decisions — skip grey areas already decided in prior phases.
Generate **3-4 grey areas** with **~4 questions each**. For each question:
- **Pre-select a recommended answer** based on: prior decisions (consistency), codebase patterns (reuse), domain conventions (standard approaches), ROADMAP success criteria
- Generate **1-2 alternatives** per question
- **Annotate** with prior decision context ("You decided X in Phase N") and code context ("Component Y exists with Z variants") where relevant
---
### Sub-step 4: Present Proposals Per Area
Present grey areas **one at a time**. For each area (M of N):
Display a table:
```
### Grey Area {M}/{N}: {Area Name}
| # | Question | ✅ Recommended | Alternative(s) |
|---|----------|---------------|-----------------|
| 1 | {question} | {answer} — {rationale} | {alt1}; {alt2} |
| 2 | {question} | {answer} — {rationale} | {alt1} |
| 3 | {question} | {answer} — {rationale} | {alt1}; {alt2} |
| 4 | {question} | {answer} — {rationale} | {alt1} |
```
Then prompt the user via **AskUserQuestion**:
- **header:** "Area {M}/{N}"
- **question:** "Accept these answers for {Area Name}?"
- **options:** Build dynamically — always "Accept all" first, then "Change Q1" through "Change QN" for each question (up to 4), then "Discuss deeper" last. Cap at 6 explicit options max (AskUserQuestion adds "Other" automatically).
**On "Accept all":** Record all recommended answers for this area. Move to next area.
**On "Change QN":** Use AskUserQuestion with the alternatives for that specific question:
- **header:** "{Area Name}"
- **question:** "Q{N}: {question text}"
- **options:** List the 1-2 alternatives plus "You decide" (maps to Claude's Discretion)
Record the user's choice. Re-display the updated table with the change reflected. Re-present the full acceptance prompt so the user can make additional changes or accept.
**On "Discuss deeper":** Switch to interactive mode for this area only — ask questions one at a time using AskUserQuestion with 2-3 concrete options per question plus "You decide". After 4 questions, prompt:
- **header:** "{Area Name}"
- **question:** "More questions about {area name}, or move to next?"
- **options:** "More questions" / "Next area"
If "More questions", ask 4 more. If "Next area", display final summary table of captured answers for this area and move on.
**On "Other" (free text):** Interpret as either a specific change request or general feedback. Incorporate into the area's decisions, re-display updated table, re-present acceptance prompt.
**Scope creep handling:** If user mentions something outside the phase domain:
```
"{Feature} sounds like a new capability — that belongs in its own phase.
I'll note it as a deferred idea.
Back to {current area}: {return to current question}"
```
Track deferred ideas internally for inclusion in CONTEXT.md.
---
### Sub-step 5: Write CONTEXT.md
After all areas are resolved (or infrastructure skip), write the CONTEXT.md file.
**File path:** `${phase_dir}/${padded_phase}-CONTEXT.md`
Use **exactly** this structure (identical to discuss-phase output):
```markdown
# Phase {PHASE_NUM}: {Phase Name} - Context
**Gathered:** {date}
**Status:** Ready for planning
<domain>
## Phase Boundary
{Domain boundary statement from analysis — what this phase delivers}
</domain>
<decisions>
## Implementation Decisions
### {Area 1 Name}
- {Accepted/chosen answer for Q1}
- {Accepted/chosen answer for Q2}
- {Accepted/chosen answer for Q3}
- {Accepted/chosen answer for Q4}
### {Area 2 Name}
- {Accepted/chosen answer for Q1}
- {Accepted/chosen answer for Q2}
...
### Claude's Discretion
{Any "You decide" answers collected — note Claude has flexibility here}
</decisions>
<code_context>
## Existing Code Insights
### Reusable Assets
- {From codebase scout — components, hooks, utilities}
### Established Patterns
- {From codebase scout — state management, styling, data fetching}
### Integration Points
- {From codebase scout — where new code connects}
</code_context>
<specifics>
## Specific Ideas
{Any specific references or "I want it like X" from discussion}
{If none: "No specific requirements — open to standard approaches"}
</specifics>
<deferred>
## Deferred Ideas
{Ideas captured but out of scope for this phase}
{If none: "None — discussion stayed within phase scope"}
</deferred>
```
Write the file.
**Commit:**
```bash
node "$HOME/.claude/get-shit-done/bin/gsd-tools.cjs" commit "docs(${PADDED_PHASE}): smart discuss context" --files "${phase_dir}/${padded_phase}-CONTEXT.md"
```
Display confirmation:
```
Created: {path}
Decisions captured: {count} across {area_count} areas
```
Read and execute: `$HOME/.claude/get-shit-done/references/autonomous-smart-discuss.md`
</step>

View File

@@ -0,0 +1,89 @@
/**
* Regression test for #2196
*
* autonomous.md exceeded the Claude Code Read tool's 10K token limit
* (reported as 11,748 tokens), causing it to be read in 150-line chunks.
*
* Fix: extract the smart_discuss step into a separate reference file so
* autonomous.md stays under the token limit.
*
* At ~4 chars/token, 10K tokens ≈ 40K chars. We target < 38K to stay
* comfortably under the limit with room for future additions.
*/
'use strict';
const { describe, test } = require('node:test');
const assert = require('node:assert/strict');
const fs = require('fs');
const path = require('path');
const PROJECT_ROOT = path.join(__dirname, '..');
// ─── Size threshold ──────────────────────────────────────────────────────────
// 38K chars ≈ 9,500 tokens — stays below 10K with margin
const AUTONOMOUS_SIZE_LIMIT = 38 * 1024;
// ─── File paths ──────────────────────────────────────────────────────────────
const AUTONOMOUS_PATH = path.join(PROJECT_ROOT, 'get-shit-done', 'workflows', 'autonomous.md');
const SMART_DISCUSS_REF = path.join(PROJECT_ROOT, 'get-shit-done', 'references', 'autonomous-smart-discuss.md');
// ─── autonomous.md size ──────────────────────────────────────────────────────
describe('autonomous.md size constraints (#2196)', () => {
test('autonomous.md file exists', () => {
assert.ok(fs.existsSync(AUTONOMOUS_PATH), `Missing: ${AUTONOMOUS_PATH}`);
});
test('autonomous.md is under 38K chars (below Claude Code 10K-token Read limit)', () => {
const raw = fs.readFileSync(AUTONOMOUS_PATH, 'utf-8');
const content = raw.replace(/\r\n/g, '\n').replace(/\r/g, '\n');
assert.ok(
content.length < AUTONOMOUS_SIZE_LIMIT,
`autonomous.md is ${content.length} chars, expected < ${AUTONOMOUS_SIZE_LIMIT} to stay under ` +
`the 10K token Read limit. Extract smart_discuss step to autonomous-smart-discuss.md reference.`
);
});
});
// ─── Reference file exists ───────────────────────────────────────────────────
describe('autonomous-smart-discuss.md reference file', () => {
test('autonomous-smart-discuss.md exists', () => {
assert.ok(
fs.existsSync(SMART_DISCUSS_REF),
`Missing: ${SMART_DISCUSS_REF} — smart_discuss step must be extracted to this file`
);
});
test('autonomous.md references autonomous-smart-discuss.md', () => {
const content = fs.readFileSync(AUTONOMOUS_PATH, 'utf-8');
assert.ok(
content.includes('autonomous-smart-discuss.md'),
'autonomous.md must reference autonomous-smart-discuss.md after extraction'
);
});
});
// ─── Reference file contains key content ────────────────────────────────────
describe('autonomous-smart-discuss.md contains key smart_discuss content', () => {
test('reference file contains smart_discuss step instructions', () => {
const content = fs.readFileSync(SMART_DISCUSS_REF, 'utf-8');
const hasSmartDiscuss =
content.includes('smart_discuss') ||
content.includes('Smart Discuss') ||
content.includes('grey area') ||
content.includes('grey_area');
assert.ok(hasSmartDiscuss, 'autonomous-smart-discuss.md must contain smart discuss step content');
});
test('reference file contains CONTEXT.md writing instructions', () => {
const content = fs.readFileSync(SMART_DISCUSS_REF, 'utf-8');
assert.ok(
content.includes('CONTEXT.md'),
'autonomous-smart-discuss.md must contain instructions for writing CONTEXT.md'
);
});
});