forgeguard

Security and trust evidence documents the project's policies and automated checks. No certification or badge level is claimed.

Read-only posture assessment for one explicitly authorized self-hosted Gitea or Forgejo instance, or an explicitly supplied anonymized configuration snapshot.

Multi-Forge Support — 0.5.0

Adds Gitea or Forgejo as separate providers, with operator-declared product identity, finite provider-specific advisory catalogs, exposure intent, request profiles, offline configuration review of an explicitly supplied snapshot, and JSON/Markdown/SARIF exports. It does not scan application source code or private artifact contents.

Qualification targets are Gitea 1.26.4 / 1.27.3 and Forgejo 15.0.8 / 16.0.4 — these are the exact tested targets, not a universal support promise; see release notes and the CI evidence linked from the release.

Install

python -m pip install forgeguard

Or pin the exact release:

python -m pip install forgeguard==0.5.0

Usage

forgeguard providers
forgeguard checks
forgeguard scan --url https://git.example.com/team --authorized --product forgejo --profile standard --policy public --dry-run
forgeguard scan --url https://git.example.com/team --authorized --product forgejo --profile standard --policy public --format json
forgeguard config review --snapshot anonymized-snapshot.json --policy private --format json

Use minimal for one version request, standard for five bounded existing requests, or extended for those requests plus root. The profiles change scope, not aggressiveness. Exposure intent (public/private/unspecified) is independent; unspecified is the default.

The operator declares product identity. Compatible APIs and inventory versions alone do not identify a product. Opposing markers, conflicting versions, unknown syntax and unsupported evidence remain explicitly incomplete. No Gitea advisory is applied to Forgejo.

Evidence boundaries

HTTP 200 on a checked path is only a status observation, not proof of repository readability, private data access or loaded configuration. Public-by-design status observations are informational. Private intent creates a bounded review warning. HTTP 401/403 may support denial on that path only. Errors, redirects, 404, 429, malformed version JSON, timeouts and truncation remain incomplete.

Incomplete assessments have null score, N/A grade and assessed=false, even when another check warns. Skipped checks remain listed. Version 2 scores are not comparable with 0.2.2. No score is security certification or a claim of no vulnerabilities.

Offline review performs zero requests and evaluates only the supplied snapshot. It supports a closed key list for registration, sign-in, new-repository privacy and product-specific MFA. It never reads app.ini, production files, tokens, databases or private keys. See config review.

Transport and reports

One target, GET only, finite allowlist, no retries or redirects, serial requests, at most 12 requests, 10 seconds each and 60 seconds overall. Responses are streamed with a 256 KiB decompressed limit. HTTPS verifies certificates; --ca-bundle explicitly supplies a private CA. Ambient proxy/CA settings are ignored. HTTP never transmits a token. Anonymous requests do not inherit authentication or cookies.

Prefer FORGEGUARD_TOKEN for the optional version-read token. The legacy --token option warns because shell history/process listings may retain it. Reports do not contain the token.

Use --format md,json,sarif with --out reports/result.md for separate files. The directory must exist. Existing files, symlinks and collisions are refused, and writes are atomic. Multiple formats require files; single machine-readable stdout is clean.

JSON uses forgeguard.assessment.v1 with a packaged schema. SARIF 2.1.0 is schema-validated against frozen OASIS errata01. No source locations are invented. No DevGuard or GitHub Code Scanning importer integration is claimed.

Documentation

• Migration and exit codes
• Security model and authorized use
• Scoring, checks, upstream evidence
• Provider guide and integration lab
• Candidate notes and release checklist

Historical 0.2.2 before/after examples remain in examples/ for migration context. Current examples use the golden- prefix and are synthetic offline fixtures, not production scans.

Commercial hardening

Gexiro offers a paid Hardening Report for authorized Gitea operators who want human review of bounded evidence, explicit limitations, prioritized findings, and a practical remediation plan.

A Remediation Sprint is a separate engagement: scope and acceptance criteria are frozen first, changes begin only after explicit approval, and the result includes before/after verification. Scope expansion, third-party targets, and remote changes require separate written authorization.

ForgeGuard remains usable as a free OSS CLI without a hosted account or control plane. Heavy SaaS and recurring monitoring are not offered at this stage.

Request a scoped ForgeGuard Hardening Report

License

Apache-2.0. See LICENSE.

Built and maintained by Gexiro Global Enterprises Ltd.

Part of the Gexiro open-source toolkit.

ForgeGuard by Gexiro

Not affiliated with Gitea, Forgejo, Codeberg, GitHub, or GitLab.