Read-only checks and skill packaging

Shared brand footer

Maintain the footer's logo, display size and main official website in one template. Do not edit generated page footers individually. The generator calculates page-relative asset paths; the generated Markdown can be displayed without a client-side script. The common footer does not carry a product license. Keep the repository's root LICENSE and third-party notices intact; GitHub detects the repository license from that file. All page languages share the same brand footer; no license translations are needed.

Preview one page's exact change before applying it:

python3 -B scripts/sync_footers.py --page docs/zh-CN/onedev.md --diff
python3 -B scripts/sync_footers.py --page docs/zh-CN/onedev.md --write
python3 -B scripts/sync_footers.py --page docs/zh-CN/onedev.md --check

After reviewing the affected pages, synchronize all public reader pages:

python3 -B scripts/sync_footers.py --all --diff
python3 -B scripts/sync_footers.py --all --write
python3 -B scripts/sync_footers.py --all --check

Without --write, the tool does not change files. It replaces only a recognized complete footer block, preserves the preceding page body and rejects missing or ambiguous markers, unexpected content after a footer, and paths outside this project. The common template is the maintained source; generated pages refresh when the tool runs. The bulk selection excludes machine-facing SKILL.md instructions, GitHub form and PR-template files under .github/, and reusable template bodies; their directory README pages still receive the shared footer except under .github/. Website resource pages supply branding in their shared reader wrapper. This tool performs no Git or network operations.

test_sync_footers.py uses temporary sample bodies for all eight languages to check root/nested page paths, shared-template updates, repeat runs, and body/newline preservation. It also checks rejected unsafe inputs and exclusion of reusable instruction/template bodies. These samples are test data, not published page content or claims of native-language editorial review. Run it with:

python3 -B -m unittest discover -s scripts -p test_sync_footers.py -v

Public-content checks and packaging

Python 3.9 or newer; standard library only. Set PYTHONDONTWRITEBYTECODE=1 when running tools from a clean source tree.

python3 -B scripts/check_public_content.py --profile gitguard
python3 -B scripts/check_public_content.py --root /path/to/public-project --staged
python3 -B scripts/check_public_content.py --root /path/to/public-project --history origin/main..HEAD
python3 -B scripts/prepare_release.py --output ../gitguard-release --media-review /private/media-review.json
python3 -B scripts/check_public_content.py --root ../gitguard-release/source --profile gitguard --media-review /private/media-review.json
python3 -B scripts/package_skills.py --root ../gitguard-release/source --output ../gitguard-skill-packages --media-review /private/media-review.json

Replace the example project path and range with the actual reviewed target. Working-tree, index and history checks are separate. The history mode examines every selected commit snapshot, including files absent from the final tree, and refuses ranges larger than 200 commits rather than silently truncating. It never fetches or pushes.

The content checker reports paths/categories/line numbers, never the matching secret text. Exit 0 means no findings within its stated heuristics; 1 means findings; 2 means an incomplete check. It cannot recognize all private information, inspect binary attachments semantically, validate external links/fragments or choose a license. It refuses oversized/nonregular files and calls out binary material for manual review. Working-tree review includes ignored files except Git metadata, because ignored secrets can still be copied accidentally into a release.

Use the generic profile for adopting projects. The stricter gitguard profile additionally rejects this package's excluded actual project-management and website paths. Educational templates remain permitted.

The package builder makes a new output directory outside the source, preserves a hash manifest and includes a complete public reference snapshot so a generated skill works without the original repository. It performs no installation, network access, Git writes, deployment or backup.

prepare_release.py uses this project's explicit public root-file and directory list, including the reviewed GitHub community templates under .github/. It preserves but omits development directories such as .vs, .vscode, .idea and bytecode caches, recording exclusions in the output's private REVIEW.json. Only its source/ directory is a public candidate. Unknown root files, linked paths, oversize files and unresolved scanner findings stop the operation. An omitted development directory is not certified clean. The default whole-working-directory scanner still examines ignored files and reports unreadable files as incomplete while preserving other findings.

The optional --media-review argument accepts a private, previously reviewed receipt. Its JSON shape is {"status":"reviewed","files":{"assets/example.png":"SHA256"}}, where each value is the actual 64-character SHA-256 digest of an individually reviewed file. Keep the real receipt outside source, packages and Git history. It resolves only binary-review findings for the exact PNG/MP4 bytes, reports those dispositions separately, and never dismisses text secrets, unreadable files, oversize files or changed media. Do not create a receipt merely to make a check pass; without prior review, the findings remain outstanding. Exit 0 can include explicitly recorded reviewed media, not a claim that binaries were automatically inspected. Private secret-matching lists remain unsupported.

Each skill ZIP now includes the full checked public reference set, including LICENSE, posters and video targets. Unzip one package into an independent directory to check its references and bundled checker; original workspace paths are not required. Existing output directories are never silently reused.

Origin and sensitive inputs

The checker is a new Git Guard helper, not a migrated account-list or history-cleaning script. It does not read private matching lists. It recognizes a limited set of generic patterns, not arbitrary passwords or account numbers. See public-content review: exact private matches and real account data belong outside every public source/package/history; tests use invented values only.