peterbamuhigire Research Meta-Initialization
Use this skill immediately after python -m engine new-project ... and before
any research sub-task is dispatched.
Required Load Order
skills/source-evaluation/SKILL.mdskills/source-evaluation/references/evidence-discipline.mdreferences/decision-tree.md
Objective
Convert the scaffolded _context/ TODO files into a concrete project operating
plan:
- research type
- domain and geography
- audience
- output family and variant
- methodology mix
- primary-research need
- monetization or reuse intent
- roadmap and validation gates
Intake
Ask the user only for facts that cannot be inferred from the existing brief.
If a field is unknown, mark it as TODO or (gap) instead of inventing detail.
Required fields:
| Field | Destination |
|---|---|
| Research type | _context/project-profile.md |
| Domain | _context/project-profile.md |
| Geography | _context/scope.md |
| Audience | _context/audience.md |
| Output family | _context/output-plan.md |
| Methodology mix | _context/methodology.md |
| Cohorts | _context/cohorts.md |
| Exclusions | _context/exclusions.md |
| Hypotheses | _context/hypotheses.md |
| Success criteria | _context/success-criteria.md |
| Monetization intent | _context/monetization.md |
Roadmap Output
Write _context/research-roadmap.md with:
- initiation tasks
- research waves
- verification pass
- synthesis pass
- output assembly
- validation and release-pack steps
Dispatch Gate
Research dispatch is blocked until the following are no longer blank:
_context/brief.md_context/project-profile.md_context/methodology.md_context/audience.md_context/output-plan.md_context/scope.md_context/success-criteria.md
Evidence Clause for Delegation
Every delegated research prompt must include the hard constraint from
source-evaluation/references/evidence-discipline.md verbatim. Do not shorten,
paraphrase, or soften it.
Use When
Use for a new or materially re-scoped research project before collection begins.
Do Not Use When
Do not rerun for routine work in an initialised project; use research-orchestration to revise active waves.
Inputs
| Input | Source/provider | If absent |
|---|---|---|
| Brief, audience, decision, output family | Requester or project files | Stop and request the missing decision or audience |
| Existing workspace state | Repository inspection | Record a new-project state; never invent files or prior decisions |
Workflow
- Inspect the workspace and stop if user-owned state would be overwritten.
- Resolve project type, audience, decision, evidence risk, method, and outputs.
- Choose cohorts and waves; recover a partial intake by recording open fields as gaps.
- Apply the evidence clause to every delegated brief and release the roadmap only when dispatch gates pass.
Outputs
| Artefact | Consumer | Acceptance condition |
|---|---|---|
| Project profile and roadmap | Research orchestrator | Audience, decision, method, cohorts, outputs, gaps, and gates are explicit |
Initiation Evidence Guidance
The completed intake, workspace inventory, decision record, and dispatch checklist form the initiation evidence.
Capability Contract
Workspace inspection is read-only by default. Creating or changing project files requires explicit authority; deletion, publication, spending, and external communication are outside this skill.
Degraded Mode
If files or tools are unavailable, return a qualified intake and wave outline with unverified fields marked gap; do not claim the project is dispatch-ready.
Decision Rules
| Choice | Action | Failure/risk avoided |
|---|---|---|
| Decision or audience missing | Stop intake | Aimless research |
| Existing project materially changed | Re-profile and preserve prior record | Silent scope drift |
| Evidence gate incomplete | Hold dispatch | Unsourced collection |
Quality Standards
The profile is traceable to supplied context, unresolved fields are visible, and no delegated task omits the evidence clause.
Initialization Pitfalls
- Guessing the audience; ask or mark the gap.
- Starting collection before scope approval; hold dispatch.
- Overwriting existing project state; reconcile it.
- Treating a method label as a method rationale; record the choice and reason.
- Omitting stop conditions; define them before waves begin.
Worked Example
A brief that names a topic but no decision produces a provisional profile and a blocked dispatch gate, not an invented executive audience.
References
- Initialization decision tree
Evidence Produced
| Evidence | Consumer | Acceptance condition |
|---|---|---|
| Intake record and dispatch checklist | Orchestrator and reviewer | Every supplied fact maps to its source and every unresolved field is marked as a gap |
Anti-Patterns
- Guessing the audience. Fix: ask the requester or mark the audience as a blocking gap.
- Starting collection before approval. Fix: stop dispatch until scope and gates are accepted.
- Overwriting project state. Fix: reconcile existing files and preserve prior decisions.
- Naming a method without rationale. Fix: record why it fits the question and constraints.
- Omitting stop conditions. Fix: define completion, escalation, and evidence thresholds before waves begin.