Structured Research En

Structured Research

Produce traceable, resumable, stoppable research judgments — not more searching. Every study runs the same pipeline; the core move is expanding circles around a phenomenon and backfilling the analysis (defined in references/expansion-and-review.md).

Entry points for every task

Paths below are relative to this skill's base directory, not the research working directory; the links open each file directly.

1. workflows/research-cycle.md: the single research pipeline; read first on every task, then choose a new study or a continuation of the same instance.
2. references/research-identity.md: identity and isolation; read before first locating an instance, resuming, switching, or delegating.
3. references/expansion-and-review.md: circle expansion and audit; read when planning search or auditing counterexamples.
4. references/evidence-and-storage.md: evidence and storage; read before registering evidence, editing artifacts, or closing a round.
5. references/worked-example.md: a full execution picture of one expansion study; read to calibrate a first round, a stalled continuation, or the expected shape of each step's output — from a rough request to stop-loss and mapping.
6. references/guided-entry.md: guided discussion; when the user's thinking is stuck (stale field, detail fixation, no concrete phenomenon, field anxiety), use the matching discussion move to widen the view.

Re-match the entry point on every new task; do not mechanically re-read every file while context is still complete. After context compaction, relocate the instance.

Identity, expansion, evidence, and quality red lines live in the entry files above (identity and write boundaries in references/research-identity.md; expansion and stop-loss in references/expansion-and-review.md). SKILL.md navigates; it does not copy their definitions.

Tools and runtime boundaries

Scripts require PowerShell 7.2+, with no third-party dependencies. Commands use this skill's actual install directory and do not depend on the current working directory.

• scripts/init-research-workbench.ps1: create an independent instance and bind it to this session.
• scripts/resolve-research.ps1: accurately resume, switch, or explicitly take over; decide write access only after CanWrite.
• scripts/append-source-index.ps1: register a source, checking identity and duplicate source IDs.
• scripts/refresh-research-index.ps1: refresh the digest first, then regenerate the layer index.
• scripts/validate-research-round.ps1: structure and reference validation; it does not replace semantic review.
• scripts/test-research.ps1: executable regression check for installs and maintenance.

Native plans, tasks, and team state stay with the host; research files hold only goals, evidence, and round results. Do not auto-create teams, assign models, name agents, or create human leases, and do not modify global rules.

Delivery

Complete evidence review and structure validation inside the pipeline, then give the user the judgments and boundaries they need, the research ID, and openable entry points to the artifacts. Present deliverable files per host protocol. Distinguish clearly between "this round is finished", "limited conclusion", and "the research goal is actually met".