localize-skill

Fills a generalized skill's context slots with one organization's confirmed local values.

2|2|Updated Aug 29, 2026
One-click install
npx skills add https://github.com/AI-Lab-for-Cities-at-Harvard/civic-skill-exchange --skill localize-skill-ai-lab-for-cities-at-harvard
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: localize-skill
Source: https://github.com/AI-Lab-for-Cities-at-Harvard/civic-skill-exchange/tree/main/skills/civic-skills/localize-skill
Command: npx skills add https://github.com/AI-Lab-for-Cities-at-Harvard/civic-skill-exchange --skill localize-skill-ai-lab-for-cities-at-harvard

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) and assets (resource) components.

What problem does it solve? Generalized civic skills ship with placeholder slots (dataset identifiers, field names, approving officials) that adopters often cannot answer, so adoption stalls or slots get filled with plausible wrong values. This Skill derives what it can from public sources and the organization's own data, asks only what no source can answer, and confirms every machine-consumed value before shipping. ## Core Features & Use Cases - Derive, propose, confirm: Looks up dataset identifiers, schema field names, and portal endpoints from public open-data catalogues, then asks the adopter to confirm by recognition rather than recall. - Org profile reuse: Maintains an <org>.profile.yml of organization-wide answers (jurisdiction, fiscal year, portal) so each subsequent skill adoption asks fewer questions. - Contract versioning and verification: Checks contract_version before filling, refuses files newer than it understands, and verifies no {{slot}} or TODO survives and every exact: true value was confirmed. - Use Case: A town adopts a generalized 311 pothole-analysis skill. This Skill finds the town's open-data portal, proposes the open-date column as case_opened_dt for confirmation, asks one batched set of remaining questions, and hands back a self-contained localized skill plus a provenance record. ## Quick Start Localize this generalized skill for the Town of Brookline using its context template and confirm each value with me before writing it in.

Frequently Asked Questions about localize-skill

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

FAQPage Schema
How do I localize a generalized civic skill for my organization?

Run localize-skill against the skill's context.template.yml. It derives values from public sources and your data, asks one batched set of questions for the rest, writes confirmed values into the skill body, and produces a standalone skill plus a provenance record.

What is the difference between generalize-skill and localize-skill?

generalize-skill takes a skill built for one organization and separates it into a portable skill plus a context file. localize-skill is the other half: it fills that context file for a new organization. If a skill was never generalized, run generalize-skill first.

Does the localized skill need the context file at run time?

No. Values are written directly into the skill body, references, and scripts. The context file is a record of decisions and provenance, kept so the skill can be re-localized later, not a runtime dependency.

What happens when a context file has a newer contract_version?

The Skill stops and fills nothing. A newer file may contain breaking changes that would be misread, producing plausible wrong values in a skill that still runs. It names the gap and asks for a newer localize-skill.

Can unconfirmed or derived values be written into the skill?

Derived values from public catalogues are proposals until a person confirms them. Values marked exact: true are never written unconfirmed; unanswered ones are left as slots and recorded in LOCALIZE-NOTES.md with what they block.

When should I not use localize-skill?

Do not use it on a skill that was never generalized, since there is no context to fill. It also does not review whether a skill is correct or worth adopting, only that every written value was confirmed.