OpenSSF Scorecard, check by check
OpenSSF Scorecard scores a repository from 0 to 10 on each of eighteen checks and publishes a weighted average. .github/workflows/scorecard.yml runs it on every push to main and weekly, uploads the findings to code scanning and publishes the result to api.scorecard.dev, which the badge in the README reads. It runs only while the repository is public.
This page states, for each check, what it measures, what the repository has, the score to expect and what would raise it. The rules are those of Scorecard v5.6.0 (checks/evaluation/ in its source); the scores marked measured are the output of scorecard --local v5.6.0 on the published files, which runs the eleven checks that need no GitHub API: 7.0 over those eleven before these changes, 10.0 after. The others are worked out from the rules and are expected, not measured, until the repository is public.
Summary
| Check | Weight | Today | With this page's set-up | Highest reachable | What the last step needs |
|---|---|---|---|---|---|
| Binary-Artifacts | high | 10, measured | 10 | 10 | n/a |
| Branch-Protection | high | 0, expected | 8 with one required review; 4 with none | 10 | two required reviewers: three people with write access |
| CI-Tests | low | not scored: no merged pull request | 10 once changes arrive by pull request | 10 | n/a |
| CII-Best-Practices | low | 0 | 5, the passing badge | 10 | the gold badge: several unassociated contributors, two-person review, 90% statement coverage |
| Code-Review | high | 0, expected | rises to 10 as approved pull requests replace the last 30 commits | 10 | a second person who approves every pull request |
| Contributors | low | 3, expected | 3 | 10 | commits from people of three organisations |
| Dangerous-Workflow | critical | 0, measured | 10, measured | 10 | n/a |
| Dependency-Update-Tool | high | 10, measured | 10 | 10 | n/a |
| Fuzzing | medium | 0, measured | 10, measured | 10 | n/a |
| License | low | 10, expected (9 measured offline) | 10 | 10 | n/a |
| Maintained | high | 0 | 0 until 13 February 2027 | 10 | the repository to be 90 days old, then one commit or maintainer issue reply a week |
| Packaging | medium | not scored | 10, expected, once release.yml has completed one run | 10 | n/a |
| Pinned-Dependencies | medium | 10, measured | 10, measured | 10 | n/a |
| SAST | medium | 10, measured | 10 | 10 | n/a |
| Security-Policy | medium | 10, measured | 10 | 10 | n/a |
| Signed-Releases | high | not scored: no release | 10 from the first release | 10 | the first release to be cut while the repository is public |
| Token-Permissions | high | 10, measured | 10, measured | 10 | n/a |
| Vulnerabilities | high | 5, measured | 10, measured | 10 | n/a |
A check that is not scored is left out of the average. The average weights a critical check 10, a high one 7.5, a medium one 5 and a low one 2.5.
| State of the repository | Expected average |
|---|---|
| public as it was before this page's changes | 4.9 |
| public, the changes merged, the ruleset with one required review, the passing badge, one release | 8.1 |
| the same, 90 days old, the last 30 commits all approved pull requests | 9.6 |
| also two required reviewers, the gold badge and three contributing organisations | 10 |
Scorecard also has two experimental checks, SBOM and Webhooks, that are off unless asked for and are not part of the published score. Releases carry an SPDX 2.3 SBOM (first-bite-<version>.spdx.json), which is what the SBOM check looks for; the repository defines no webhook.
The checks
Binary-Artifacts
Looks for executables and compiled objects committed to the tree. There are none: the binary files are docs/brief/first-bite-brief.pdf, icon/first-bite.png and examples/weather-station.pcap, none of which is executable code. The WebAssembly decoder is built by make wasm into site/pkg/, which .gitignore excludes.
Branch-Protection
Reads the rules on the default branch and scores them in five tiers; a tier counts only when the one before it is complete.
| Tier | Points | Rule |
|---|---|---|
| 1 | 3 | no force push; no deletion |
| 2 | 3 | at least one approving review; pull request required; branch up to date before merging; approval of the last push |
| 3 | 2 | at least one required status check |
| 4 | 1 | two approving reviews; review by a code owner |
| 5 | 1 | stale approvals dismissed on a new push; no one may bypass the rules |
Today the only rule on main is the organisation's, which requires signed and verified commits: 0 points, because tier 1 is not met.
rulesets/ holds three bodies for a repository ruleset on main, and one for release tags. All three forbid deletion and force pushes, require a pull request, a linear history, squash merging, resolved review threads and these status checks on an up-to-date branch: ci, leak-gate and contributor-trust. ci is the last job of ci.yml: it always runs and fails if any other job of that workflow failed, so a change that touches no code, whose Rust jobs are skipped, is not left waiting for checks that never report. None has a bypass list. They differ in the review they require:
| File | Approving reviews | Score | What it costs |
|---|---|---|---|
rulesets/main-no-review.json | none | 4 | nothing: the maintainer merges their own pull request when the checks pass. Code-Review stays at 0 |
rulesets/main-one-review.json | one, by a code owner, given after the last push | 8 | a pull request cannot be merged by its author alone. With one maintainer, every change waits for a second member of goat-owners; nobody can bypass |
rulesets/main-two-reviews.json | two | 10 | three people for every change |
CONTRIBUTING.md already states the middle one: "One approving review from a code owner merges it, by squash."
Adding a bypass actor (an administrator who may merge without review) to main-one-review.json keeps the 8, because the bypass rule is in tier 5; it then leaves Code-Review to fall with every change merged that way.
Apply one of the three, and the tag ruleset:
gh api -X POST repos/USAREUR-AF/first-bite/rulesets --input docs/security/rulesets/main-one-review.json
gh api -X POST repos/USAREUR-AF/first-bite/rulesets --input docs/security/rulesets/tags.json
rulesets/tags.json forbids deleting, moving or force-pushing a v* tag. It is not scored; it keeps a release tag pointing at the commit that was built. rulesets/tag-creation.json lets only the maintainers' team goat-owners create a tag, as the other goat repositories do; it is designed, not applied.
The status checks are named after the jobs of ci.yml, leak-gate.yml and contributor-trust.yml. A job that is renamed has to be renamed in the ruleset, or pull requests wait for a check that never reports.
CI-Tests
Of the pull requests merged among the last 30 commits, the share whose head commit had a successful check run. ci.yml runs on every pull request. The history so far was pushed, not merged from pull requests, so the check has nothing to count and reports no score; with the ruleset every change is a pull request with required checks, and the score is 10.
CII-Best-Practices
Reads the project's OpenSSF Best Practices badge: 2 points for a questionnaire in progress, 5 for passing, 7 for silver, 10 for gold. The project is not registered. A person has to register it and answer the questionnaire; best-practices-answers.md has an answer and its evidence for each of the passing criteria.
Silver also asks for statement coverage of at least 80%, measured, and for the project to carry on if one person leaves. Gold asks for at least two unassociated significant contributors, review of half of all changes by a second person, and 90% statement coverage. Coverage is not measured.
Code-Review
Of the changes among the last 30 commits, the share that are pull requests approved by someone other than their author. A change approved and written by a bot is not counted. Every commit so far was pushed by one maintainer: 0. The score is whatever share of the last 30 commits were approved pull requests, so with one required review it rises with each merge and reaches 10 after 30.
Contributors
Counts the organisations and companies of the authors of the last 30 commits who have at least five contributions, from the company and the public organisation memberships on their GitHub profiles: 10 for three or more, in proportion below. One organisation gives 3.
Dangerous-Workflow
Looks for a workflow that is triggered by pull_request_target or workflow_run and checks out the pull request's code, and for a script that interpolates text an outsider controls.
leak-gate.yml needs pull_request_target: it is what lets a pull request from a fork be checked against the maintainers' list of names, a secret. It used to check out the pull request beside the gate, as data. It now checks out the base branch only, fetches the pull request's commit into the object store and reads it from there with tools/check-public-tree.sh --tree <commit>: every blob is written under a number to a temporary directory and scanned; nothing of the pull request is checked out, built or run, and the check that runs is the one on main. The check no longer skips its own file, so a pull request cannot hide text in a changed copy of it.
Dependency-Update-Tool
Looks for a Dependabot or Renovate configuration. .github/dependabot.yml covers the workspace's crates, the fuzz targets' crates, the GitHub Actions, the Dockerfile, the fuzz builder image and the mock images, weekly.
Fuzzing
Looks for an OSS-Fuzz project, a ClusterFuzzLite configuration (.clusterfuzzlite/Dockerfile), or fuzz functions in the source (for Rust, libfuzzer_sys).
fuzz/ has three libFuzzer targets for the code that reads input the project does not control: decode_payload, load_dictionary and frame_pipeline. ClusterFuzzLite builds them from .clusterfuzzlite/ and runs them for 5 minutes on pull requests that change code (cflite_pr.yml) and for 30 minutes every night (cflite_batch.yml).
Measured on one core each of an AMD Ryzen 7 8845HS, AddressSanitizer build, from the seed corpus:
| Target | Time | Executions | Edges covered | Crashes |
|---|---|---|---|---|
decode_payload | 601 s | 2,963,994 | 633 | 0 |
load_dictionary | 601 s | 1,642,518 | 5,562 | 0 |
frame_pipeline | 601 s | 17,989,010 | 2,628 | 0 |
The ClusterFuzzLite build was run in its builder image and each target then run for a minute in its runner image, on the build host. The two workflows have not run on GitHub.
License
Looks for a licence file at the top level that GitHub recognises as an FSF or OSI licence. LICENSE is the Apache License 2.0. Offline the check cannot ask GitHub which licence it is and gives 9.
Maintained
0 for a repository created less than 90 days ago, whatever its activity; the repository was created on 15 November 2026. After that: the commits plus the issues a maintainer answered in the last 90 days, out of 12, so one a week gives 10.
Packaging
Reads every workflow file for a job that publishes a package, and asks GitHub whether that workflow has a successful run: 10 when both hold, no score otherwise. A job counts when one of its steps matches a list in IsPackagingWorkflow (checks/fileparser/github_workflow.go); for a container image the two entries are a step that uses docker/build-push-action, and a run step in which the text docker is followed on the same line by push. Release archives on the releases page are not on the list, and the crates are not published to crates.io (publish = false).
The image-publish job of release.yml pushes the container image of the Dockerfile to ghcr.io/usareur-af/first-bite with docker push in a run step, which is the second entry. The run that the check asks for is any successful run of release.yml (ListWorkflowRunsByFileName with Status: "success", in clients/githubrepo/workflows.go): the dry run counts, so the score does not wait for a release.
gh workflow run release.yml --ref main -f static=true
That builds, runs and packages every archive and both images and publishes nothing. The job has not run on GitHub.
Pinned-Dependencies
Looks at every GitHub Action, container image and downloaded tool in the workflows, Dockerfiles and shell scripts for a reference that can change under the project: a tag where a commit or digest could be given.
| What | How it is pinned |
|---|---|
| GitHub Actions | full commit SHA, checked by tools/check-pins.py |
images in Dockerfile, .clusterfuzzlite/Dockerfile, the Makefile, the mock compose files | tag and sha256 digest, checked by tools/check-pins.py |
libpcap source (tools/libpcap-static.sh), Npcap SDK (release.yml) | version and SHA-256 of the download |
cargo-sbom (release.yml), cargo-audit (rust-audit.yml) | exact version; cargo install --locked |
yamllint (ci.yml) | exact version, run with pipx |
| crates | Cargo.lock and fuzz/Cargo.lock, with --locked |
Not pinned: the packages apt-get installs in the two Dockerfiles and on the runners (Debian and Ubuntu remove superseded versions, so a pinned version stops installing); the two container images inside ClusterFuzzLite's actions, which the actions name by the tag v1; and the *_RUNS_IN images of the Makefile, which exist to try the release on each distribution's current image. Scorecard counts none of these.
SAST
Looks for CodeQL or another static analyser, and at whether each merged pull request was analysed. codeql.yml analyses the Rust and Python sources and the workflows. It ran only on pull requests that changed one of those; it now runs on every pull request and every push to main, so that every merged change has an analysis. clippy with warnings as errors runs in ci.yml.
Security-Policy
Looks in SECURITY.md for a link or address to report to, text beyond it, and the words and numbers of a disclosure process. SECURITY.md has the private reporting form, what to include, and the times to expect.
Signed-Releases
Looks at the assets of the last five releases for a signature (.sig, .asc, .minisig, .sigstore, .sigstore.json: 8 points) or SLSA provenance (.intoto.jsonl: 10 points). It does not verify them.
release.yml already attested every asset with actions/attest-build-provenance: a signed in-toto statement with a SLSA provenance predicate, kept by GitHub. It now also attaches that statement to the release as first-bite-<version>.intoto.jsonl, so the provenance travels with the files:
gh attestation verify first-bite-1.0.1-x86_64-linux-gnu.tar.gz --repo USAREUR-AF/first-bite --bundle first-bite-1.0.1.intoto.jsonl
GitHub issues attestations for public repositories; a release cut while the repository is private has no provenance and would score 0. There is no release yet.
The check reads the last five releases, pre-releases included, and gives the average. A pre-release such as v1.0.1-rc.1 is built by the same workflow and carries its own first-bite-1.0.1-rc.1.intoto.jsonl, so it scores 10 like any other release.
Token-Permissions
Looks for workflows whose token can write. Every workflow sets permissions: contents: read at the top; write scopes are on the one job that needs each: security-events for CodeQL and Scorecard, id-token for Scorecard, Pages and the release attestation, pages for the deployment, contents and attestations for publishing a release, packages and attestations for publishing the image. The check leaves packages: write out of account in a workflow that IsPackagingWorkflow matches (createIgnoredPermissions in checks/raw/permissions.go).
Vulnerabilities
Asks OSV for open advisories against the lock files: 10, less one for each. Five were reported. One is a name collision: the advisory is for the crates.io crate bite, and the bite in the lock file is this repository's decoder. The other four are in crates that only the optional zenoh feature builds (one vulnerability in lz4_flex, three notices that a crate is no longer maintained); the default build and the release binaries do not contain them. osv-scanner.toml lists each with its reason and what removes the entry, as .cargo/audit.toml does for cargo audit.
What is left to do, in order
| # | Who | What |
|---|---|---|
| 1 | maintainer | Choose the review requirement: one of the three ruleset files under Branch-Protection. |
| 2 | administrator | Create the ruleset on main and the tag ruleset with the two gh api calls above. |
| 3 | administrator | Make the repository public. CodeQL, Scorecard and Pages start running, and attestations become available. |
| 4 | administrator | Run cargo-deny and cflite-batch once from the Actions tab, so that their badges have a result on main. |
| 5 | maintainer | Register for the Best Practices badge with best-practices-answers.md and put the project number in the README badge. |
| 6 | maintainer | Run the release workflow as a dry run, after step 3: gh workflow run release.yml --ref main -f static=true. Nothing is published; Packaging scores from this run. |
| 7 | maintainer | Cut the first release by tag: it carries the provenance and publishes the image. Then add the release badge to the README. |
| 8 | administrator | Set the visibility of the package first-bite to public in the organisation's package settings, once, after the first image is pushed: a new package is private. |
The workflow
.github/workflows/scorecard.yml pins ossf/scorecard-action v2.5.0 by commit, publishes the results, uploads the SARIF to code scanning, and gives security-events: write and id-token: write to its one job. It reads rulesets with the workflow token; no personal token is configured.
The source of this page is a Markdown file in the repository.