STANDARD OPERATING PROCEDURE
Purpose
- Coordinate multi-pipeline deep research work from question framing to productionized insights.
- Enforce quality gates, evidence hygiene, and explicit confidence ceilings at each stage.
- Maintain structure-first artifacts (SKILL, README, examples, references) and log handoffs.
Trigger Conditions
- Positive: multi-week research programs, academic-grade studies, or cross-pipeline coordination needs.
- Negative: single-question quick looks (use
researcher) or pure prompt shaping (use prompt-architect).
Guardrails
- Constraint extraction in HARD / SOFT / INFERRED buckets for scope, ethics, compute, and timelines.
- Two-pass refinement per deliverable: structure/coverage then epistemic/validation.
- Quality gates: Discovery (literature + gap clarity), Experimentation (replication + ablations), Production (deployment + monitoring).
- Explicit confidence ceilings on all claims; do not exceed evidence tier.
Inputs
- Research objective, success metrics, and timeline.
- Available datasets/resources, compute budget, and risk constraints.
- Required stakeholders and integration points.
Workflow
- Frame & Route: Define goal, success metrics, constraint buckets; map to pipelines (literature, replication, method dev, eval, publication).
- Plan Handoffs: Assign roles/agents, memory tags, and deliverable owners; schedule quality gates.
- Execute Pipelines: Run discovery → replication → experimentation → synthesis; use sub-skills as needed.
- Adversarial Validation: Challenge assumptions, run boundary cases, and verify reproducibility before exit.
- Package & Communicate: Consolidate artifacts, COV notes, risks, and next actions; ensure README/examples reflect latest practice.
Validation & Quality Gates
- Gate readiness recorded with evidence for discovery, experimentation, and production.
- Constraint coverage tracked; unresolved INFERRED items flagged.
- Deliverables include logs of handoffs, decisions, and confidence ceilings.
- Storage paths and tags recorded for recall.
Response Template
**Goal & Constraints**
- HARD: ...
- SOFT: ...
- INFERRED (confirm): ...
**Pipelines & Owners**
- Discovery → Replication → Experimentation → Production.
**Status**
- Gate checkpoints and evidence summaries.
- Risks / blockers / follow-ups.
**Artifacts**
- Links to notes, datasets, models, dashboards.
Confidence: 0.80 (ceiling: research 0.85) - based on validated gate evidence and logged handoffs.
Confidence: 0.80 (ceiling: research 0.85) - Reflects orchestrated, evidence-backed stage tracking.
1---2name: deep-research-orchestrator3description: End-to-end orchestration of the deep research lifecycle across discovery, experimentation, and production gates.4---5
6
7
8## STANDARD OPERATING PROCEDURE
9
10### Purpose
11- Coordinate multi-pipeline deep research work from question framing to productionized insights.
12- Enforce quality gates, evidence hygiene, and explicit confidence ceilings at each stage.
13- Maintain structure-first artifacts (SKILL, README, examples, references) and log handoffs.
14
15### Trigger Conditions
16- **Positive:** multi-week research programs, academic-grade studies, or cross-pipeline coordination needs.
17- **Negative:** single-question quick looks (use `researcher`) or pure prompt shaping (use `prompt-architect`).
18
19### Guardrails
20- Constraint extraction in HARD / SOFT / INFERRED buckets for scope, ethics, compute, and timelines.
21- Two-pass refinement per deliverable: structure/coverage then epistemic/validation.
22- Quality gates: Discovery (literature + gap clarity), Experimentation (replication + ablations), Production (deployment + monitoring).
23- Explicit confidence ceilings on all claims; do not exceed evidence tier.
24
25### Inputs
26- Research objective, success metrics, and timeline.
27- Available datasets/resources, compute budget, and risk constraints.
28- Required stakeholders and integration points.
29
30### Workflow
311. **Frame & Route**: Define goal, success metrics, constraint buckets; map to pipelines (literature, replication, method dev, eval, publication).
322. **Plan Handoffs**: Assign roles/agents, memory tags, and deliverable owners; schedule quality gates.
333. **Execute Pipelines**: Run discovery → replication → experimentation → synthesis; use sub-skills as needed.
344. **Adversarial Validation**: Challenge assumptions, run boundary cases, and verify reproducibility before exit.
355. **Package & Communicate**: Consolidate artifacts, COV notes, risks, and next actions; ensure README/examples reflect latest practice.
36
37### Validation & Quality Gates
38- Gate readiness recorded with evidence for discovery, experimentation, and production.
39- Constraint coverage tracked; unresolved INFERRED items flagged.
40- Deliverables include logs of handoffs, decisions, and confidence ceilings.
41- Storage paths and tags recorded for recall.
42
43### Response Template
44```
45**Goal & Constraints**
46- HARD: ...
47- SOFT: ...
48- INFERRED (confirm): ...
49
50**Pipelines & Owners**
51- Discovery → Replication → Experimentation → Production.
52
53**Status**
54- Gate checkpoints and evidence summaries.
55- Risks / blockers / follow-ups.
56
57**Artifacts**
58- Links to notes, datasets, models, dashboards.
59
60Confidence: 0.80 (ceiling: research 0.85) - based on validated gate evidence and logged handoffs.
61```
62
63Confidence: 0.80 (ceiling: research 0.85) - Reflects orchestrated, evidence-backed stage tracking.
majiayu000