← back to Cli Printing Press
feat(skills): auto-brainstorm features before absorb gate (#87)
d289a89237f2629c2907b6291608ae929f27e385 · 2026-03-30 20:41:51 -0700 · Matt Van Horn
Step 1.5c.5 now self-brainstorms by answering 3 questions from research
context (power user workflows, pain points, killer feature) in addition
to the existing 5-category gap analysis. Features are scored and added
to the transcendence table before the gate fires.
The user sees the full manifest including brainstormed ideas at first
glance. "Brainstorm more features" becomes "Add your own feature ideas"
with reworded questions targeting personal knowledge research can't
surface.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Files touched
A docs/plans/2026-03-30-009-feat-auto-brainstorm-before-absorb-gate-plan.mdM skills/printing-press/SKILL.md
Diff
commit d289a89237f2629c2907b6291608ae929f27e385
Author: Matt Van Horn <mvanhorn@users.noreply.github.com>
Date: Mon Mar 30 20:41:51 2026 -0700
feat(skills): auto-brainstorm features before absorb gate (#87)
Step 1.5c.5 now self-brainstorms by answering 3 questions from research
context (power user workflows, pain points, killer feature) in addition
to the existing 5-category gap analysis. Features are scored and added
to the transcendence table before the gate fires.
The user sees the full manifest including brainstormed ideas at first
glance. "Brainstorm more features" becomes "Add your own feature ideas"
with reworded questions targeting personal knowledge research can't
surface.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---
...feat-auto-brainstorm-before-absorb-gate-plan.md | 109 +++++++++++++++++++++
skills/printing-press/SKILL.md | 25 +++--
2 files changed, 124 insertions(+), 10 deletions(-)
diff --git a/docs/plans/2026-03-30-009-feat-auto-brainstorm-before-absorb-gate-plan.md b/docs/plans/2026-03-30-009-feat-auto-brainstorm-before-absorb-gate-plan.md
new file mode 100644
index 00000000..a87bf773
--- /dev/null
+++ b/docs/plans/2026-03-30-009-feat-auto-brainstorm-before-absorb-gate-plan.md
@@ -0,0 +1,109 @@
+---
+title: "feat: Auto-brainstorm features before absorb gate"
+type: feat
+status: active
+date: 2026-03-30
+---
+
+# feat: Auto-brainstorm features before absorb gate
+
+## Overview
+
+Run the brainstorm questions internally (LLM answers them using research context) and pre-populate the transcendence table with brainstormed features BEFORE presenting the absorb gate. The user sees the full feature set including brainstormed ideas at first glance. "Brainstorm more features" remains as an option for users who want to add their own ideas on top.
+
+## Problem Frame
+
+The absorb gate currently presents the manifest, then offers "Brainstorm more features" as option 2. When the user picks it, the skill asks 3 smart questions:
+1. "What workflows do you personally use `<API>` for that aren't covered?"
+2. "What's annoying about existing tools that you wish someone would fix?"
+3. "If this CLI could do one magical thing, what would make you say 'I need this'?"
+
+These questions are great - but the LLM already has enough research context to answer them itself. In the Redfin run, the auto-suggested features (Step 1.5c.5) scored 7-9/10 and the brainstorm-generated features (portfolio tracker, smart scoring, export engine) scored 8-10/10. The user just wanted to say YES to everything. The brainstorm dialogue was friction that delayed approval without adding insight the LLM didn't already have.
+
+The fix: run the brainstorm questions as an internal LLM exercise during Step 1.5c.5 (which already runs automatically), merge the results into the transcendence table, and present the combined set at the gate. The user still sees "Brainstorm more features" for adding their own ideas - but the default path already includes smart brainstormed features.
+
+## Requirements Trace
+
+- R1. The 3 brainstorm questions are answered by the LLM using research context during Step 1.5c.5
+- R2. Brainstormed features are scored with the same 4-dimension scoring system and added to the transcendence table
+- R3. The absorb gate presents the COMBINED set (auto-suggested + auto-brainstormed) as the default manifest
+- R4. "Brainstorm more features" remains available for users who want to add their own ideas on top
+- R5. No new user-facing prompts are added - the brainstorm happens silently as part of the existing auto-suggest step
+
+## Scope Boundaries
+
+- Only modifying `skills/printing-press/SKILL.md`
+- Not changing the scoring system or thresholds
+- Not removing the "Brainstorm more features" option - just making it additive rather than the first pass
+- Not changing Phase 1 research or Phase 2 generation
+
+## Key Technical Decisions
+
+- **Merge into Step 1.5c.5, not a new step**: The auto-suggest step already analyzes research and generates scored features. Adding the brainstorm lens (workflows, pain points, one magical thing) is a natural extension, not a separate step.
+- **LLM self-brainstorm, not user Q&A**: The 3 questions become internal prompts the LLM answers using the research brief and absorb manifest as context. This works because the research already covers user workflows, pain points, and competitor gaps.
+- **Label brainstormed features distinctly**: In the transcendence table, mark auto-brainstormed features with a different evidence source (e.g., "brainstorm: domain workflow analysis") so the user can distinguish them from data-driven suggestions if they want.
+
+## Implementation Units
+
+- [ ] **Unit 1: Expand Step 1.5c.5 to include self-brainstorm**
+
+ **Goal:** Add a self-brainstorm section to the auto-suggest step that answers the 3 brainstorm questions internally and generates additional features.
+
+ **Requirements:** R1, R2, R3, R5
+
+ **Files:**
+ - Modify: `skills/printing-press/SKILL.md`
+
+ **Approach:**
+ - In Step 1.5c.5 (Auto-Suggest Novel Features), after the existing 5-category gap analysis, add a 6th category: "Self-brainstorm"
+ - The self-brainstorm section instructs the LLM to answer 3 questions using research context:
+ 1. Based on the research brief's top workflows and user profiles, what workflows does the typical power user of this API do that aren't covered in the absorbed features?
+ 2. Based on competitor repo issues, community pain points, and ecosystem gaps found in Phase 1/1.5, what are the most annoying limitations that a CLI with SQLite could fix?
+ 3. Based on the NOI and domain archetype, what single "killer feature" would make a power user install this CLI over any alternative?
+ - Each answer that produces a concrete feature gets scored with the same 4-dimension system
+ - Features scoring >= 5/10 are added to the transcendence table with evidence citing which research findings informed them
+ - Update the gate message to say "auto-suggested and brainstormed" instead of just "auto-suggested"
+
+ **Patterns to follow:**
+ - The existing Step 1.5c.5 gap analysis categories and scoring system
+ - The existing "Evidence" column requirement for transcendence features
+
+ **Verification:**
+ - Step 1.5c.5 now includes a self-brainstorm section with 3 internal questions
+ - The absorb gate message references both auto-suggested and brainstormed features
+ - "Brainstorm more features" option still exists at the gate for user additions
+ - No new user-facing prompts are added before the gate
+
+- [ ] **Unit 2: Update gate option 2 to be additive with structured questions**
+
+ **Goal:** Reword option 2 so it's clear the user is adding their own ideas on top of the auto-brainstorm. Keep the 3 structured questions but reword them to ask for personal knowledge the research couldn't surface.
+
+ **Requirements:** R4
+
+ **Dependencies:** Unit 1
+
+ **Files:**
+ - Modify: `skills/printing-press/SKILL.md`
+
+ **Approach:**
+ - Change option 2 label from "Brainstorm more features" to "Add your own feature ideas"
+ - Change option 2 description from "Interactive dialogue to explore your own feature ideas before building" to "Add features from your personal experience that research couldn't surface"
+ - Keep the 3 structured questions but reword them to target user-specific knowledge:
+ 1. "What workflows do YOU personally use `<API>` for that we might have missed?"
+ 2. "What frustrates YOU about this API that the research didn't surface?"
+ 3. "What's YOUR killer feature - something only you'd think of?"
+ - The structured questions still generate scored features added to the transcendence table
+ - After the brainstorm round, return to the gate with the updated manifest
+
+ **Verification:**
+ - Option 2 label and description updated
+ - The 3 structured questions still fire when the user picks option 2
+ - Questions are reworded to target personal/experiential knowledge, not research-derivable knowledge
+ - Features from user brainstorm are still scored and added to transcendence table
+
+## Sources & References
+
+- Current absorb gate: `skills/printing-press/SKILL.md` Phase 1.5, lines 869-1023
+- Step 1.5c.5 (Auto-Suggest): lines 950-1002
+- Phase Gate 1.5 (brainstorm option): lines 1004-1023
+- Redfin run transcript showing brainstorm generated portfolio tracker (10/10), smart scoring (9/10), export engine (8/10) - all features the LLM could have generated from research alone
diff --git a/skills/printing-press/SKILL.md b/skills/printing-press/SKILL.md
index fa6699c8..7a589155 100644
--- a/skills/printing-press/SKILL.md
+++ b/skills/printing-press/SKILL.md
@@ -970,9 +970,14 @@ Analyze these 5 categories using data already gathered — do NOT run new search
5. **Agent workflow gaps** — What would an AI agent using this CLI wish it could do in one command instead of multiple? (e.g., "show me everything about X" commands, bulk operations, pre-flight checks)
+6. **Self-brainstorm** — Answer these 3 questions using the research context gathered so far. Do NOT ask the user — answer them yourself from the research brief, absorb manifest, and ecosystem findings:
+ - Based on the research brief's top workflows and user profiles, what workflows does the typical power user of this API do that aren't covered in the absorbed features?
+ - Based on competitor repo issues, community pain points, and ecosystem gaps found in Phase 1/1.5, what are the most annoying limitations that a CLI with SQLite could fix?
+ - Based on the NOI and domain archetype, what single "killer feature" would make a power user install this CLI over any alternative?
+
#### Generate and Score Candidates
-Generate 3-5 novel feature ideas. For each, score on 4 dimensions:
+Generate 3-8 novel feature ideas (across all 6 categories). For each, score on 4 dimensions:
| Dimension | Points | Scoring |
|-----------|--------|---------|
@@ -999,27 +1004,27 @@ The "Evidence" column MUST cite specific findings from Phase 1 or Phase 1.5 rese
Write to `$RESEARCH_DIR/<stamp>-feat-<api>-pp-cli-absorb-manifest.md`
-The manifest now includes both compound use cases (Step 1.5c) and auto-suggested features (Step 1.5c.5) in the transcendence table.
+The manifest now includes compound use cases (Step 1.5c) and auto-suggested + auto-brainstormed features (Step 1.5c.5) in the transcendence table.
### Phase Gate 1.5
**STOP.** Present the absorb manifest to the user via `AskUserQuestion`:
-"Found [N] features across [X] tools (MCPs, skills, CLIs, scripts). Our CLI will absorb all [N] and add [M] transcendence features ([K] auto-suggested with scores). Total: [N+M] features. This is [Z]% more than the best existing tool."
+"Found [N] features across [X] tools (MCPs, skills, CLIs, scripts). Our CLI will absorb all [N] and add [M] transcendence features (auto-suggested and brainstormed with scores). Total: [N+M] features. This is [Z]% more than the best existing tool."
Options:
-1. **Approve — generate now** — Start CLI generation with the full manifest
-2. **Brainstorm more features** — Interactive dialogue to explore your own feature ideas before building
+1. **Approve - generate now** — Start CLI generation with the full manifest
+2. **Add your own feature ideas** — Add features from your personal experience that research couldn't surface
3. **Review the research** — Show me the full brief and manifest before deciding
4. **Trim scope** — The feature count is too ambitious, let's focus on a subset
-If user selects **"Brainstorm more features"**, run a lightweight feature brainstorm:
+If user selects **"Add your own feature ideas"**, ask 3 structured questions targeting personal knowledge the research couldn't surface:
-1. "What workflows do you personally use `<API>` for that aren't covered in the manifest?"
-2. "What's annoying about existing tools for `<API>` that you wish someone would fix?"
-3. "If this CLI could do one magical thing, what would make you say 'I need this'?"
+1. "What workflows do YOU personally use `<API>` for that we might have missed?"
+2. "What frustrates YOU about this API that the research didn't surface?"
+3. "What's YOUR killer feature - something only you'd think of?"
-Each answer that produces a concrete feature → add to the transcendence table. After the brainstorm, return to this gate with the updated manifest.
+Each answer that produces a concrete feature → score and add to the transcendence table. After the brainstorm, return to this gate with the updated manifest.
WAIT for approval. Do NOT generate until approved.
---
← 5bc0b6b8 fix(skills): add mandatory sniff gate checkpoint before abso
·
back to Cli Printing Press
·
fix(skills): add mandatory publish checkpoint after archive dfc20ac0 →