construct-curation-cycle

Orchestrates graph maintenance by delegating curation runs to the Python curation capability.

Updated Apr 21, 2026
One-click install
npx skills add https://github.com/mabstruct/construct --skill construct-curation-cycle-mabstruct
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: construct-curation-cycle
Source: https://github.com/mabstruct/construct/tree/main/CONSTRUCT-CLAUDE-impl/claude/skills/construct-curation-cycle
Command: npx skills add https://github.com/mabstruct/construct --skill construct-curation-cycle-mabstruct

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Knowledge graphs accumulate stale cards, orphaned notes, and untyped connections over time, and manually reviewing every proposed lifecycle change is tedious and error-prone. This Skill runs a full maintenance pass while keeping a human approval gate over every write. ## Core Features & Use Cases - Delegated Curation Runs: Invokes the construct curation run CLI (or MCP tool) to execute integrity checks, decay detection, orphan detection, promotion evaluation, and connection typing without inline logic. - Consolidated Review Gate: Presents a single grouped queue of promotion, connection, archive, and escalation proposals, surfacing the method field so users can distinguish rule-based escalations from LLM judgments. - Gated Writes and Health Report: Resumes the run via construct curation review with approve-all, reject-all, or per-item decisions, then narrates the capability's graph health report. - Use Case: When a user says "clean up the graph", the Skill runs the maintenance pass, walks them through approving proposed promotions and archives, and reports the resulting graph health counts. ## Quick Start Ask the agent to run a curation cycle on the current workspace, for example by saying "curate my knowledge graph and show me the health report".

Frequently Asked Questions about construct-curation-cycle

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run a knowledge graph maintenance pass with the construct CLI?▼

Run `construct curation run --workspace . --json` to execute the deterministic maintenance pass covering integrity, decay, orphan detection, promotion evaluation, and connection typing. The run pauses at a review gate and returns a run_id and gate queue before writing anything.

How do I approve or reject proposed card promotions and archives?▼

Resume the paused run with `construct curation review --workspace . --run-id <run_id>` using `--approve-all`, `--reject-all`, or a per-item decisions JSON payload. Only approved items are written to the workspace; rejected items are dropped.

Can the curation cycle run through MCP instead of the CLI?▼

Yes, start the server with `construct mcp` and invoke the `construct_curation_run` tool with a workspace_path argument. The MCP path follows the same gate-and-review flow as the CLI.

What is the difference between rule-based and llm-judgment escalation methods?▼

A rule-based method means the gate escalated because the LLM evaluation failed or retried out, making it a mechanical escalation. An llm-judgment method indicates a genuine borderline decision the model reasoned about, helping reviewers prioritize attention.

Does the curation skill write directly to the knowledge graph?▼

No, the skill performs no direct workspace writes. The Python curation capability owns every write behind the human approval gate, applying promotions, connections, and archives only after review, and logs curation_cycle_complete to the event log.