Open-Earth-Foundation

Guide non-technical contributors through making safe code changes to CityCatalyst. Use when someone without coding experience wants to update text, translations, colors, copy, or small UI details, and needs the agent to handle git, verification, and PR creation for them.

Create or update pull requests with repository-derived context and concise title/body standards. Use when creating, updating, automating, drafting, or polishing PRs, or preparing GitHub PR tool payloads.

Review pull requests for high-impact issues only. Use when the user asks to review a PR, especially requests like "do a review of PR <id>", "review PR <id>", or "PR review". Evaluate security, AGENTS.md alignment, repository structure, reuse over reimplementation, complexity, and regression risk. Return only actionable comments worth changing.

Run a read-only Kubernetes health audit against an explicit readonly cluster context such as `dev-cluster-readonly` or `prod-cluster-readonly`. Use when the user wants an automated cluster/pod health check, diagnosis of failing workloads, or a report of likely causes without making any write changes.

Answer a single Kubernetes question using readonly investigation and evidence-first output.

Generate structured Linear tickets (stories, tasks, bugs, spikes) with AI-enriched code references and acceptance criteria from a brief description. Use when the product team asks to create a ticket, write a story, draft a task, report a bug, or plan a spike for CityCatalyst.

Enrich an existing ticket with detailed technical breakdown, sub-tasks, code references, and estimation guidance. Use when an engineer asks to refine a ticket, break down a story, add technical details, estimate a task, or decompose work into sub-tasks.

Create or update any LLM-related prompt file in this repository using the required `<role>`, `<task>`, `<input>`, and `<output>` structure, with optional `<tools>` when tool policy is needed, and explicit model-aligned field contracts. Use for prompts stored in markdown, Python prompt templates, YAML/JSON config prompts, and any other runtime prompt definitions for large language models (LLMs).

Manufacture a structured staging dataset by compiling many public sources (award lists, registries, reports, portals) into schema'd, provenance-flagged, verified records — for cases where no single external dataset exists to adopt. Use whenever the user wants to compile, harvest, scrape, assemble, or "go collect examples/records" from scattered public sources; build a dataset that doesn't exist yet; or turn many announcements into one queryable table — even if they never say "compile". Sibling to dataset-discovery: discovery *finds* an existing dataset, compile *builds* one. Writes only to dataset-compile/<slug>/, never the knowledge base or the catalog.

Build the Mage pipeline that ships a vetted dataset into the modelled database — the third stage after discover → review. Use whenever the user wants to build/create a pipeline, ship or promote a reviewed dataset to production, load a cleaned dataset into modelled.*, register a dataset in the catalog, upload source/cleaned data to S3, or "wire up the pipeline for <dataset>". Orchestrates the stage and defers the how-to to knowledge-base/topics/engineering-standards/; it does not restate those standards.

Validate and profile a dataset after a pipeline has loaded it into the modelled database. Use when the user wants to test/validate loaded data, check a modelled table, write or run data expectations, check nulls / value sets / ranges, produce a data-quality or validation report, or "say something about the data". The post-load layer that complements Mage's in-pipeline @test checks; runs against the live DB.

Deep-dive review of a dataset for CityCatalyst: verify access and license, profile the actual data, read the methodology, run the fit-for-purpose check, and draft the review entry in dataset-review/reviews/. Use whenever the user wants to review, assess, vet, or "go deeper into" a dataset; promote a discovery candidate; check whether a dataset is fit for purpose; or create/update an entry under dataset-review/ — even if they just say "let's look at this dataset properly". Follows dataset-discovery in the pipeline: discover → review → pipeline.

Group the dataset catalog into collections by theme and by geography, and keep theme tagging honest. Use whenever the user wants to find datasets by theme or by country/global, see what's available in a topic area (finance, energy, transport, land use…), regenerate the by-theme or by-geography collections, reconcile or clean up theme tags, or check which review folders are missing from the catalog. Writes dataset-review/collections/by-theme.yaml and by-geography.yaml; reads the catalog but never edits it. Complements dataset-review, which produces the individual entries this skill groups.

Find and triage candidate datasets for CityCatalyst data needs. Use whenever the user wants to find, source, scout, or search for data; mentions a data gap; asks what data exists for a sector, theme, or geography; or wants to open, update, or resolve a data need — even if they never say "search" or "discovery". Also use when deciding whether a dataset someone suggested is worth a full review. Produces a need folder in dataset-discovery/needs/ containing need.md, search.yaml, and candidates.yaml.

Review pull requests for high-impact issues only. Use when the user asks to review a PR, especially requests like "do a review of PR <id>", "review PR <id>", or "PR review". Focus on security, repository conventions, reuse over reimplementation, structure consistency, and unnecessary complexity. Return only actionable comments worth changing.

Create or update any LLM-related prompt file in this repository using the required `<role>`, `<task>`, `<input>`, and `<output>` structure, with optional `<tools>` when tool policy is needed, and explicit model-aligned field contracts. Use for prompts stored in markdown, Python prompt templates, YAML/JSON config prompts, and any other runtime prompt definitions for large language models (LLMs).

Review LLM pipeline changes that add or modify retry logic, batching, multi-pass aggregation, fallback flows, or persisted diagnostics/artifacts. Use when an agent, orchestrator, writer, or API path changes how LLM calls are retried, split, combined, failed closed, or written to/read from run artifacts.

Use when a user asks to run Expect or expect-cli, validate the frontend in a real browser, or surface Expect's native replay output for this repository.

Review generated, imported, or cleaned-up Markdown artifacts for conversion damage before changes are applied. Use when Codex needs to inspect Markdown files for encoding corruption, broken tables, whitespace merges, repeated extraction fragments, malformed blockquotes, or similar document-conversion artifacts and propose exact fixes before editing.

Analyze GitHub pull request review threads against the current local branch. Use when the user asks to list unresolved PR comments, quote the exact review text, explain the code under review, determine whether the issue still exists, and state whether a fix or only a reply is needed.

Adds a dataset entry to catalog/datasets.yaml in the geospatial-data repository. Use when the user asks to add a dataset to the catalog, create a catalog entry, or register a new dataset in datasets.yaml.

Generates aws s3 sync and cp commands to upload geospatial transformation outputs to geo-test-api. Use when the user asks for S3 upload commands, wants to upload a dataset to S3, or needs commands to deploy transformation outputs. Does not run commands—outputs them for the user to execute.

Adds a new dataset to the geospatial-data repository with transformation pipeline, catalog entry, and collection. Use when adding a new GEE ImageCollection, raster, or vector dataset, or when the user asks to create a transformation, add a dataset to the catalog, or generate COGs and tiles.

Draft PRs for agentic-coder. Base branch is main.

Generate Conventional Commits messages for agentic-coder.