What it implements
An SCI-grounded (SCI = (E × I + M) / R, per the GSF SCI specification) scoring tool that evaluates a codebase's sustainability posture and produces a single, versioned, trackable score plus a prioritized recommendations report. It treats static-analysis signals (code complexity, dependency weight, Docker image composition, CI/CD pipeline design, IaC configuration) as labeled-confidence proxies for the SCI E term, never presented as audited carbon accounting, following the same estimate-with-disclosed-confidence posture the SCI spec itself recommends.
What form it takes
A GitHub Action / CLI tool that can run in two modes: (1) a one-time full-repo scan producing a baseline score, and (2) a PR-triggered delta scan that comments the sustainability-score change caused by that diff (the same workflow shape as existing coverage/security bots). Deterministic static analyzers do the measurable-fact extraction; an LLM pass (a documented, versioned system prompt) reasons over the structured findings to write the human-readable report and recommendations. The LLM never invents or re-derives numbers itself.
Workload/use case
General-purpose application repositories (language-agnostic core, initial reference implementation targets JS/TS + Python as the two most common ecosystems) that use Docker, a CI/CD pipeline (GitHub Actions initially), and optionally IaC (Terraform/Helm). Not scoped to a specific cloud provider for v1; cloud-billing integration is called out below as an explicit future phase, not part of this proposal.
Scope statement
In scope for v1: static repo scan, weighted rubric across five pillars (code/algorithmic efficiency, containerization, CI/CD, cloud/infra config as declared in code, SRE/ops signals such as health checks and rollback config), single numeric score plus confidence-tagged findings, and a Markdown report with PR-comment delta. Out of scope for v1: live cloud billing/telemetry integration and live grid carbon-intensity API calls (both flagged as data gaps in v1 output, not silently defaulted), and any claim of audited/precise carbon-emissions accounting.
How it relates to existing entries
Complements rather than duplicates tools/gsf/green-code-skill: that project is a design/review-time pattern advisor with no scoring or reporting output (explicitly out of scope per its own README); this project is the scoring/reporting layer that could sit downstream of it, potentially consuming its pattern catalog as one input source. Distinct from the archived Green-Software-Foundation/carbon-ci project, which attempted a narrower static-analysis-only version of PR carbon scoring; this proposal extends that idea to also cover containerization, CI/CD design, and SRE/ops practices in one unified score, and documents the estimate-confidence methodology carbon-ci did not fully specify before archival.
What it implements
An SCI-grounded (SCI = (E × I + M) / R, per the GSF SCI specification) scoring tool that evaluates a codebase's sustainability posture and produces a single, versioned, trackable score plus a prioritized recommendations report. It treats static-analysis signals (code complexity, dependency weight, Docker image composition, CI/CD pipeline design, IaC configuration) as labeled-confidence proxies for the SCI E term, never presented as audited carbon accounting, following the same estimate-with-disclosed-confidence posture the SCI spec itself recommends.
What form it takes
A GitHub Action / CLI tool that can run in two modes: (1) a one-time full-repo scan producing a baseline score, and (2) a PR-triggered delta scan that comments the sustainability-score change caused by that diff (the same workflow shape as existing coverage/security bots). Deterministic static analyzers do the measurable-fact extraction; an LLM pass (a documented, versioned system prompt) reasons over the structured findings to write the human-readable report and recommendations. The LLM never invents or re-derives numbers itself.
Workload/use case
General-purpose application repositories (language-agnostic core, initial reference implementation targets JS/TS + Python as the two most common ecosystems) that use Docker, a CI/CD pipeline (GitHub Actions initially), and optionally IaC (Terraform/Helm). Not scoped to a specific cloud provider for v1; cloud-billing integration is called out below as an explicit future phase, not part of this proposal.
Scope statement
In scope for v1: static repo scan, weighted rubric across five pillars (code/algorithmic efficiency, containerization, CI/CD, cloud/infra config as declared in code, SRE/ops signals such as health checks and rollback config), single numeric score plus confidence-tagged findings, and a Markdown report with PR-comment delta. Out of scope for v1: live cloud billing/telemetry integration and live grid carbon-intensity API calls (both flagged as data gaps in v1 output, not silently defaulted), and any claim of audited/precise carbon-emissions accounting.
How it relates to existing entries
Complements rather than duplicates tools/gsf/green-code-skill: that project is a design/review-time pattern advisor with no scoring or reporting output (explicitly out of scope per its own README); this project is the scoring/reporting layer that could sit downstream of it, potentially consuming its pattern catalog as one input source. Distinct from the archived Green-Software-Foundation/carbon-ci project, which attempted a narrower static-analysis-only version of PR carbon scoring; this proposal extends that idea to also cover containerization, CI/CD design, and SRE/ops practices in one unified score, and documents the estimate-confidence methodology carbon-ci did not fully specify before archival.