skillZs
★ LIVE SKILL TAGS ★
>>> LIVE SKILLS INDEX <<<
* OPEN SOURCE *
NO LOGIN, NO TRACKING
※ REAL INSTALL DATA ※
← back to all skills
levnikolaevich/claude-code-skills70 installs

ln-022-researchgraph

Indexes and queries project research graphs backed by hex-research MCP. Use for hypotheses, goals, benchmark runs, evidence depth, derived goal metrics, lineage, generated research maps, and graph audits.

How do I install this agent skill?

npx skills add https://github.com/levnikolaevich/claude-code-skills --skill ln-022-researchgraph
view source ↗

Is this agent skill safe to install?

  • Gen Agent Trust Hubpass

    This skill provides tools for indexing and querying project research data, such as hypotheses, goals, and benchmark evidence. It operates on local markdown and manifest files and uses a specialized set of MCP tools for graph analysis. No malicious behaviors or security vulnerabilities were identified.

  • Socketpass

    No alerts

  • Snykpass

    Risk: LOW · No issues

What does this agent skill do?

Paths: File paths are relative to this skill directory.

Research Graph

Type: Standalone Utility Category: 0XX Dev Environment

Indexes and queries canonical project research files through hex-research-mcp: hypotheses, goals, tasks, sources, source library entries, benchmark run manifests, evidence links, derived goal metrics, lineage, generated research maps, and audit gaps.

Mandatory Read

MANDATORY READ: Load references/researchgraph_mcp_usage.md

Inputs

InputRequiredSourceDescription
project_pathyesargs or CWDProject root containing research graph files
commandnoargsSpecific action: audit, inspect H##, goal G##, runs, evidence, lineage, progress, proposed, canvas

When to Use

Use this skill when the project has any of:

  • docs/hypotheses/*.md
  • docs/goals/*.md
  • benchmark/runs/*/manifest.yaml

Use it when the user asks about research status, live or pending hypotheses, goal readiness, benchmark evidence, source quality, evidence depth, proposal readiness, implementation gaps, drift, field-level changes, or a visual/generated research map.

Do not use this skill for code symbol ownership, call graphs, references, or edit blast radius. Use ln-021-codegraph or host hex-graph policy for semantic code questions.

MCP Availability

Use hex-research when available. If unavailable, manually read the split research files with Glob/Grep/Read, answer from direct file evidence, and mark confidence as degraded.

Treat STALE as actionable graph debt or missing project metadata, not as a tool failure. Treat INVALID from verify_index as diagnostic unless a write/index operation fails.

Workflow

Phase 1: Detect Graph Layout

Check for the canonical layout:

  • docs/hypotheses/*.md
  • docs/goals/*.md
  • benchmark/runs/*/manifest.yaml

If none exist, stop and report that the project does not expose a research graph yet. Do not invent hypotheses or goals.

Phase 2: Verify or Index

Start read-only:

verify_index({ path: "{project_path}" })

Rebuild only when the index is missing, stale, explicitly requested, or the task requires current graph state:

index_hypotheses({ path: "{project_path}" })

The local index is stored under .hex-skills/researchgraph/ in the target project.

Phase 3: Route Intent

User intentTool
live, pending, rejected, mixed, not-started hypothesesfind_hypotheses
inspect H##inspect_hypothesis
inspect G## or goal statusinspect_goal
evidence, sources, benchmark prooffind_evidence
targeted or comprehensive benchmark runsfind_runs
parent/child hypothesis chaintrace_lineage
graph hubs, disconnected nodes, shapeanalyze_topology
missing goals, sources, task gaps, driftaudit_orphans
goal treetrace_goal_tree
goal readiness, derived metrics, and metric coverageinspect_goal, audit_goal_alignment
source quality or evidence depthinspect_hypothesis, find_evidence, analyze_proposed
changed research files and field-level deltas in current diffanalyze_progress
readiness of proposed next hypothesesanalyze_proposed
JSON Canvas exportexport_canvas
generated docs/research-map.md exportexport_research_map

Phase 4: Present Results

Report:

  1. tool status (OK, STALE, INVALID, ERROR)
  2. key hypotheses/goals/runs with IDs
  3. actionable graph debt separately from tool failures
  4. follow-up tool calls when the graph points to a narrower next question

For export_canvas and export_research_map, run dry-run first. Write only after dry-run output is useful and the user requested the file. Treat split files as canonical; generated research-map.md is not a second manual source of truth.

Manual Fallback

When MCP is unavailable:

  1. Read docs/goals/*.md and docs/hypotheses/*.md frontmatter first.
  2. Read benchmark/runs/*/manifest.yaml only after narrowing by H/G IDs or status.
  3. Prefer exact IDs (H04, G1) over broad scans.
  4. State that lineage, topology, and drift findings are manual approximations.

Definition of Done

  • Graph layout detected or absence reported
  • Index verified or rebuilt when needed
  • User intent routed to the smallest relevant tool
  • STALE / graph debt separated from MCP/tool failure
  • Manual fallback confidence stated when MCP was unavailable
  • Canvas writes use dry-run before write

Version: 0.1.0 Last Updated: 2026-05-08

Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.

<a href="https://skillzs.dev/skills/levnikolaevich/claude-code-skills/ln-022-researchgraph">View ln-022-researchgraph on skillZs</a>