context-graph

Drafts and reviews semantic relationship proposals through the deterministic context-graph CLI.

Updated Jul 9, 2026
One-click install
npx skills add https://github.com/t-step/bindle-archive --skill context-graph-t-step
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: context-graph
Source: https://github.com/t-step/bindle-archive/tree/main/skills/context-graph
Command: npx skills add https://github.com/t-step/bindle-archive --skill context-graph-t-step

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Authoring semantic relationships in a Bindle context graph requires strict endpoint legality, canonical candidate keys, and human-confirmed acceptance; doing this by hand risks invalid edges, fabricated keys, or silent repairs that corrupt the graph. ## Core Features & Use Cases - Proposal drafting: Builds proposal.json envelopes with source, relationship, target, basis, and explanation, grounded in real node IDs from the CLI's preview output. - CLI-driven validation: Runs propose, confirm, and apply so the deterministic CLI remains the sole authority for endpoint legality, candidate keys, and acceptance. - Ontology safety: Refuses reserved relationships like implements, normalizes symmetric contradicts to one canonical candidate, and surfaces the CLI's exact rejection findings instead of silently repairing invalid proposals. - Use Case: While reviewing a design decision, you ask the agent to record that a decision constrains an assumption; the skill inspects the graph, drafts the proposal, validates it via propose, and presents the candidate for your explicit confirmation. ## Quick Start Ask the agent to propose a semantic relationship between two nodes in your Bindle context graph and confirm the validated candidate it returns.

Frequently Asked Questions about context-graph

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

FAQPage Schema
How do I propose a semantic relationship in a Bindle context graph?▼

Ground the proposal in the CLI's preview output so source and target are real node IDs, write a proposal.json with producer set to skill, then run propose. The CLI validates endpoint legality and returns a candidate with its canonical key for human review.

How do I confirm a candidate edge in the context graph?▼

Run confirm with the exact candidate key emitted by propose and your accept, reject, or retire decision. Confirm re-validates against the current graph under lock, and only then does a judgment enter the ledger.

Can I use the implements relationship between a decision and a pull request?▼

No, implements is reserved and the CLI rejects it. Implementation attribution uses decision implemented_by github_pr instead, which propose validates as a legal endpoint combination.

Does the skill decide whether a relationship is valid?▼

No, the deterministic CLI is the sole authority for endpoint legality, candidate keys, and acceptance. The skill only drafts proposals and surfaces the CLI's findings; a rejected proposal is never silently repaired into a different valid one.

Why does the skill refuse to derive the project slug from the git remote?▼

Project identity is an opaque project_id allocated once by init and stored in config.json, plus a configured slug. Deriving identity from a git remote or owner/repo pair conflates repository hosting with project identity, so the skill only displays and passes through configured identity.