feat(ideation): capture-first default — lean brainstorm, idea import, techniques on demand#11
Open
ydmitry wants to merge 3 commits into
Open
Conversation
… techniques on demand Invert the default from frame-first machinery to capture-first: - DEFAULT (no clear signal): one plain `generate.brainstorm` call — no framing checkpoint, no persona fan-out, no compare report. Backed by a blind A/B: plain beats the old 2-persona `starter` on substance (4.09 vs 3.88), relevance (4.5 vs 3.88) and mechanism diversity (13 vs 11.5); 4/4 judges prefer plain. (The win is architectural — one self-deduping agent vs two blind parallel personas — not a claim that techniques are bad.) - INGEST: new `generate.import` operator stores the user's own ideas; being a generate.* op it rides the Step-5B emit loop, so imports populate the live wall at ingest (closes the previously-validated FEED gap). - Light-frame precondition (`frame.light`) resolves the frame FK without a framing pause; real framing deferred (frame-on-first-iterate). - `starter` demoted to opt-in (`--techniques`); all operators/playbooks kept and surfaced via the Step 6 technique menu (discoverability). New operators: generate.brainstorm, generate.import. Planner Step 3 routing reworked (3a ingest / 3b opt-in / 3c lean default). Stress-tested via a multi-agent workflow; all findings fixed (min_cohort linter break, finished starter demotion in playbook sources, --import wiring, stale avoid_when pointers, regenerated operator reference). Note: live/serve.py and live/view.html (separate in-progress viz work) are intentionally left uncommitted.
…analysis The working harness and reports behind the capture-first redesign: - UX stress test + dynamic-wall prototype (STRESS-TEST-REPORT.md, view.dynamic.html, replay harness, screenshots) - quality loops Tracks A–E (idea-substance interventions: all null/marginal; QUALITY-LOOP-REPORT.md + *-result.json + workflow scripts) - plain-vs-starter gate (the A/B that justifies the lean default) - import→iterate→visualize journey validation (VALIDATION-import-iterate-visualize.md) - capture-first stress test + punch-list (STRESS-TEST-capture-first.md) and the new-default specs/operators (new-default/) Evidence and analysis only; not loaded by the skill at runtime.
Follow-on quality-loop experiments on whether classical techniques
(Reverse Brainstorming, SCAMPER, Six Thinking Hats) beat a plain call —
evidence for the redesign's "techniques stay opt-in" decision:
- Track F (as generators): none beat plain (4 tech-wins / 7 plain); plain
is the most mechanically diverse arm. Idea substance is model-bound.
- Track G (as transforms on captured ideas): technique-choice is noise,
but developing a seed beats the raw seed 8/8 — developing is the lever.
- Track H (4 frames + why-analysis): confirmed and explained — the model
already makes the high-value moves; most ideas have one obvious fix.
Adds track-{f,g,h} results + reports, QUALITY-LOOP-REPORT addenda, and the
experiment extracts. Evidence only; not loaded by the skill at runtime.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What & why
Inverts the
ideationskill's default from frame-first machinery to capture-first:generate.brainstormcall — no framing checkpoint, no persona fan-out, no compare report. A blind A/B showed a single plain call matches or beats the old 2-personastarterdefault on substance (4.09 vs 3.88), relevance (4.5 vs 3.88) and mechanism diversity (13 vs 11.5); 4/4 blind judges preferred plain.generate.importoperator stores the user's own ideas; being agenerate.*op it rides the Step-5B emit loop, so imports populate the live wall (closes a validated "can't feed my own ideas" gap).frame.light) resolves the not-null frame FK without a framing pause; real framing is deferred (frame-on-first-iterate).starterdemoted to opt-in (--techniques); every operator/playbook is kept and surfaced via the Step 6 technique menu.New operators:
generate.brainstorm,generate.import. Planner Step 3 routing reworked (3a ingest / 3b opt-in / 3c lean default).Evidence (
stress-test/)Backed by a multi-track measurement harness (blind, position-swapped A/B throughout):
min_cohortlinter break that broke the operator catalog, plus a half-donestarterdemotion)Validation status (honest)
Validated by spec review + CLI plumbing (
list-operators --format jsonloads all 28 operators), a reproduced linter break, and an adversarial multi-agent re-check. Not yet a full interactive orchestrator→planner→subagent run.Notes
live/serve.pyandlive/view.html(separate in-progress visualization WIP) are intentionally excluded from this PR.stress-test/analysis is bundled as evidence — happy to drop it for a code-only PR if preferred.