Researchstack

Researchstack

Use this root skill only as the routing and cross-stage coordination layer. Route a stage-specific request to the narrowest skill.

Runtime Bootstrap

Run this first when the host supports shell preambles:

_ROOT=$(git rev-parse --show-toplevel 2>/dev/null || pwd)
RESEARCHSTACK_ROOT="$HOME/.codex/skills/researchstack"
[ -d "$_ROOT/.agents/skills/researchstack" ] && RESEARCHSTACK_ROOT="$_ROOT/.agents/skills/researchstack"
[ -d "$_ROOT/bin" ] && RESEARCHSTACK_ROOT="$_ROOT"
RSTACK_BIN="$RESEARCHSTACK_ROOT/bin"
RSTACK_HOST=$("$RSTACK_BIN/researchstack-host-detect" 2>/dev/null || echo "unknown")
RSTACK_PROJECT=$("$RSTACK_BIN/researchstack-project-slug" --root "$_ROOT" --print-json 2>/dev/null || echo "{}")
RSTACK_MEMORY=$("$RSTACK_BIN/researchstack-memory-status" --root "$_ROOT" --print-json 2>/dev/null || echo "{}")
RSTACK_CONFIG=$("$RSTACK_BIN/researchstack-config" show 2>/dev/null || echo "{}")
echo "HOST: $RSTACK_HOST"
echo "PROJECT: $RSTACK_PROJECT"
echo "MEMORY: $RSTACK_MEMORY"
echo "CONFIG: $RSTACK_CONFIG"

Interpret the bootstrap:

• Existing project plus uncertainty about the next action: researchstack-next-step.
• Broad area and freedom to choose a problem: researchstack-idea-finder.
• Fixed topic that needs a new method: researchstack-method-synthesis.
• Durable direction intended to support several papers: researchstack-program-map.
• Rough one-paper thesis: researchstack-lab-intake.
• Missing or stale installs: run the doctor from the source checkout and compare managed-version markers before replacing anything.

Read references/workflow.md when a request spans multiple stages. Read references/memory.md when prior project decisions matter.

Routing

• researchstack-next-step: choose the next research action in an ongoing project.
• researchstack-program-map: design a multi-paper agenda around one durable topic or grant direction.
• researchstack-idea-finder: discover paper-sized problems when the topic is open.
• researchstack-method-synthesis: generate and stress-test mechanisms while keeping the topic fixed.
• researchstack-lab-intake: turn a rough direction into a paper brief.
• researchstack-idea-review: judge topic significance and method novelty separately.
• researchstack-idea-refine: repair a weak thesis or method without automatically changing topic.
• researchstack-literature-map: map competing mechanisms, claims, assumptions, and evidence.
• researchstack-learn: maintain durable project and researcher memory.
• researchstack-paper-reproduction: reconstruct an existing paper and its evidence path.
• researchstack-experiment-design: design baselines, rival predictions, ablations, workloads, and failure tests.
• researchstack-experiment-ops: manage runs and claim promotion.
• researchstack-artifact-audit: trace claims and figures to code, configs, data, and logs.
• researchstack-code-review: inspect research-code correctness and claim alignment.
• researchstack-paper-write: draft and revise evidence-aligned sections.
• researchstack-figure-studio: plan figures and tables; route rendering to specialized visual tools when available.
• researchstack-paper-layout: improve LaTeX structure and submission presentation.
• researchstack-submission-gate: make a final submit, delay, or retarget decision.
• researchstack-peer-review: simulate strict conference review.
• researchstack-rebuttal-coach: build a point-by-point evidence-backed response.

Non-Negotiable Research Behavior

• Topic overlap does not imply method overlap. Judge topic state and method state separately.
• When a topic matters but the method is weak, route to researchstack-method-synthesis before recommending a new topic.
• Investigate established mechanisms before inventing new ones; state what is reused, adapted, and still missing.
• Preserve the real semantic unit and end-to-end lifecycle. Moving required work outside the measured boundary is not an optimization or contribution.
• Prefer falsifiable hypotheses and discriminating experiments over inspirational framing.
• Keep claims no broader than the current evidence and record source-read and claim-promotion states.
• Treat shared infrastructure as an enabling asset, not a repeated contribution across papers.
• Distinguish systems, networking, and ML evidence standards rather than mixing them loosely.