dare · diff

git:20260411.0b6d343 to git:20261023.e14c1d7

46 added, 113 removed. Audit A to A.

---

- name: /dare

- description: DARE research engine entry point — orchestrates the full research pipeline

- type: meta-strategy

- layer: meta-strategy

- calls: [intake, lit-survey, gap-analysis, insight, round]

+ name: dare

+ description: "DARE v4 is the product-level orchestrator for a research run. It turns a user request into a controlled sequence of North Star crystallization, executable specification, and phase-by-phase execution."

---

- # /dare — De-Anthropocentric Research Engine

- ## Overview

- Entry point for the full DARE research pipeline. Transforms a user's research topic

- into validated research artifacts (survey, gaps, ideas, experiment designs, results).

- ## Layer Rules

- - **Layer**: meta-strategy — top of the hierarchy

- - **Called by**: user (via `/dare` command)

- - **Calls**: strategies only (intake, lit-survey, gap-analysis, insight, round)

- - **NEVER**: calls tactics, SOPs, or MCP tools directly

- ## Pipeline

- ### Phase 0: Brainstorming (superpowers integration)

- - MUST invoke `superpowers:brainstorming` for structured requirement clarification

- - Truncate after clarify questions — DARE's research loop replaces brainstorming's design/spec flow

- - Output: `userTopic` (refined research topic)

- ### Phase 0.5: Topic-Size Classification

- <HARD-GATE>

- Before entering the research loop, classify the research topic into one of three size tiers.

- This classification determines budget targets for ALL downstream strategies.

- | Tier | Description | Example |

- |------|-------------|---------|

- | **Small (S)** | Narrow sub-problem, specific technique, single benchmark | "Improving GNN message-passing efficiency on heterogeneous graphs" |

- | **Medium (M)** | Standard research topic spanning multiple sub-areas | "Graph neural networks for drug discovery" |

- | **Large (L)** | Broad survey, cross-disciplinary, or emerging field | "Foundation models for scientific discovery" |

- **Classification heuristic:**

- - S: topic can be described in one sentence with a single technical focus

- - M: topic spans 2-4 distinct sub-areas or methodologies

- - L: topic spans 5+ sub-areas, multiple disciplines, or is a survey/position paper

- Store the tier in state: `topicSize: "S" | "M" | "L"`

- This tier is passed to every strategy and determines their budget targets.

- </HARD-GATE>

+ # dare

- ### Phase 1: Intake

- - Call **intake strategy** with `userTopic`

- - Output: `researchBrief` (structured topic + initial queries + field context)

+ ## Purpose

- ### Phase 2: Research Loop (Stages 1-3)

+ DARE v4 is the product-level orchestrator for a research run. It turns a user request into a controlled sequence of North Star crystallization, executable specification, and phase-by-phase execution. It is not a scientific graph node and must not be counted among the 267 tactic/SOP nodes.

- #### Research Budget Overview (by topic size)

+ ## Input Contract

- The following aggregate targets guide the entire research loop. Individual strategy budgets sum to these totals.

+ ```yaml

+ required: [user_request]

+ optional: [existing_context, north_star, research_brief, constraints]

+ constraints:

+ - a run may start with an existing North Star only when it is explicit and user-confirmed

+ - execution requires a current SpecView reconstructed from checkpoint events

+ ```

- |--------|-------|--------|-------|

- | Papers read (total) | 30 | 60 | 100+ |

- | Web pages fetched (total) | 30 | 80 | 200+ |

- | Search queries (total) | 20 | 50 | 100+ |

- | Iteration rounds (total) | 6 | 10 | 15+ |

- | Cross-domain papers (ideation) | 5 | 15 | 25+ |

+ ## Execution Protocol

- - **Round 0 (Cold Start)**:

- - Call **round strategy** with `{ researchBrief, roundNumber: 0 }`

- - Runs: lit-survey(10 iter) → gap-analysis(6 iter) → insight(7-step pipeline) → ideation(5 iter) → review

+ DARE enforces this order and does not skip a phase:

- - **Hot Loop (Round 1-6)**:

- - Read review feedback

- - Brief user on strategy for this round

- - Call **round strategy** with `{ researchBrief, previousRoundState, roundNumber }`

- - Stop when: score >= 8/10 AND no CRITICAL issues, OR 7 rounds max

+ 1. **North Star**: if a confirmed North Star and ResearchBrief are absent, collect and crystallize them from the user request. If they are present, verify that they still describe the request.

+ 2. **Spec**: You MUST load skill `research-catalog` to expose the available tactics. You MUST load skill `write-research-spec` to turn the confirmed North Star, ResearchBrief, and user constraints into an executable Research Spec. Do not execute research while this phase is incomplete.

+ 3. **Execution**: after the user approves the Spec, You MUST load skill `execute-research-spec` to execute its phases and checkpoints. Do not select a tactic directly from the flat skill directory.

- ### Phase 3: Experiment Design (P2+)

- - STUB in P0. Will call experiment-design strategy.

+ At every phase boundary, record the decision and its reason in the current Phase checkpoint event stream. A plan change is an appended `decisions` event, never an edit to an earlier checkpoint. The product shell may present progress, but scientific conclusions remain in tactic/SOP Delta fields.

- ### Phase 4: Experiment Execution (P2+)

- - STUB in P0. Will call experiment-execution strategy with session-teleport.

+ ## Output Contract

- ## State Object

- ```json

- {

- "userTopic": "...",

- "researchBrief": {...},

- "knowledge": "...",

- "papersRead": [],

- "urlsVisited": [],

- "gaps": [],

- "ideas": [],

- "reviewHistory": [],

- "currentRound": 0

- }

+ ```yaml

+ produces:

+ - confirmed_north_star

+ - research_brief

+ - spec_view

+ - phase_execution_results

+ - final_research_summary

+ state_delta_fields:

+ - findings

+ - evidence_updates

+ - hypothesis_updates

+ - assumption_updates

+ - uncertainties

+ - decisions

+ - open_questions

+ - recommended_jumps

```

- ## P0 Implemented

- - [x] Intake strategy (Phase 1)

- - [x] Literature Survey strategy (Phase 2, stage 1)

- - [x] Round strategy skeleton (Phase 2)

- - [x] Self-review-quick SOP (lightweight validation)

- - [x] All 7 dare-agents tools (facet, digest, rate, self-review, C/D/J debate)

+ ## Completion Criteria

- ## P1 Implemented

- - [x] INSIGHT 7-step pipeline (root-cause → stakeholder → tension → hmw → abstraction → assumption → validation)

- - [x] Full multiagent debate (4 structured C→D rounds + Judge + 3 free rounds)

- - [x] Quality-diversity filtering (MAP-Elites niche selection)

- - [x] Gap-analysis strategy (6 iterations, calls insight tactic)

- - [x] Insight strategy (orchestrates INSIGHT pipeline + debate validation)

- - [x] INSIGHT SOPs (7 step SOPs + QD filtering SOP)

- - [x] Debate SOPs (debate-idea, debate-gap, debate-insight, debate-experiment)

- - [x] Tactics (insight, multiagent-debate, review)

+ - A confirmed North Star and ResearchBrief exist before spec construction.

+ - The Spec has explicit stages, inputs, tactics, completion gates, and backtrack conditions.

+ - Execution reaches a complete checkpoint for every required stage, or reports the first unmet gate with its evidence.

+ - The final summary cites the context files and checkpoints that support it.

- ## P2 Implemented

- - [x] IDEATION methods (5 categories, 31 dare-agents tools)

- - [x] IDEATION SOPs (31 SKILL.md files across 5 categories)

- - [x] IDEATION tactics (9: 2 operational + 5 categorical + 2 compound)

- - [x] Ideation strategy (iterative: generate → augment → filter → validate)

- - [x] Idea-generation tactic (5-category divergence + cross-pollination + convergence)

- - [x] Idea-augmentation tactic (reviewer2 + SCAMPER + surgery improvements)

+ ## Failure and Backtrack

- ## After P2

- - [ ] Experiment design strategy

+ Missing intent or scope blocks entry to specification and returns `NEEDS_CONTEXT`. An unmet completion gate keeps the current stage active. A triggered backtrack condition is presented for user confirmation and, if approved, is recorded as a new decision before execution returns to the target stage. DARE never treats an empty result, elapsed time, or an execution interruption as stage completion.

- ## P3 Implemented

- - [x] Paper-writing strategy (future v3.1+ — interface defined)

- - [x] Method-evolve strategy (future v3.2+ — tools implemented)

- - [x] Method-evolve dare-agents tools (3: mutate, crossover, evaluate)

- - [x] Method-evolve SOPs (3: method-mutate, method-crossover, method-evaluate)

- - [x] Method-evolution tactic (orchestrates mutation + crossover + evaluation cycle)

+ ## Boundary

+ This entry layer is installed under `skills/` alongside the 267 tactic and SOP nodes. It is not a graph node, does not create a third scientific layer, and does not replace tactic or SOP contracts.